Supply Chain Transformation: Why the Technology Is the Easy Part

Let's get the uncomfortable truth out of the way first.

The supply chain AI market is growing rapidly. Investment is accelerating. According to a Deloitte survey, 85% of leaders increased their AI investment over the past year. The Australian Industry Group reports that 27% of industrial businesses rank AI-powered solutions as a top investment priority for 2026. Platforms like o9 Solutions, Kinaxis, Blue Yonder and SAP IBP are competing aggressively for enterprise adoption. The technology itself — for demand forecasting, inventory optimisation, logistics routing, supplier risk, warehouse automation — is genuinely capable.

And yet, the results are overwhelmingly disappointing.

An MIT study published in mid-2025 found that 95% of enterprise generative AI pilots delivered zero measurable return on investment. Forbes reports that upwards of 85% of AI initiatives don't make it into production, and of those that do, a significant portion fail to deliver the expected value. McKinsey's 2025 State of AI survey found that while 88% of organisations use AI in at least one function, only about a third have managed to scale it beyond experiments or pilots — and only 39% report any improvement to EBIT at all, with most of those seeing less than 5% impact.

For supply chain specifically, the picture is equally stark. McKinsey research found that only 16% of companies have successfully scaled AI in supply chain operations, despite 68% claiming to use it. PwC's 2025 Digital Trends in Operations Survey reported that 92% of operations and supply chain leaders cite at least one reason why their technology investments haven't fully delivered expected results, with integration complexity (47%) and data issues (44%) the most common culprits.

These numbers should give every Australian supply chain leader pause — not because AI doesn't work, but because something between the investment decision and the operational outcome is consistently going wrong. Understanding what that something is, and how to avoid it, is worth more than any vendor demonstration or conference keynote.

The gap isn't technology. It's everything around the technology.

The most important thing to understand about supply chain AI failure is that the technology itself is almost never the problem. The algorithms work. Machine learning can genuinely improve forecast accuracy. Optimisation engines can identify inventory savings that humans miss. Predictive models can flag supplier risks earlier than manual monitoring. Computer vision can improve quality control. Natural language processing can accelerate document review and contract analysis.

The failures happen in the space between the technology's capabilities and the organisation's ability to use them. As one analysis put it, the high failure rate observed today reflects implementation challenges, not inherent limitations of AI. Companies that mistake these struggles for permanent constraints risk missing the window to build a competitive advantage.

This is a critical distinction. When an AI project fails, the instinct is often to blame the tool — the vendor oversold it, the model wasn't accurate enough, the platform was too complex. Sometimes those things are true. But far more often, the project failed because the organisation wasn't ready for it — and nobody assessed that readiness honestly before committing budget.

Let's look at the specific patterns.

Pattern one: starting with the technology instead of the problem

This is the most common and most expensive mistake. An organisation sees a compelling vendor demonstration, reads about what a competitor is doing, or faces board pressure to "do something with AI." So they select a platform, launch a pilot, and then try to find a problem for it to solve.

It happens more than you'd think. Gartner found that 37% of leaders in low-maturity AI organisations identified "finding the right use case" as one of their top three implementation barriers. That statistic should be alarming — it means more than a third of organisations are investing in AI without being clear about what they're trying to achieve.

In supply chain, the technology-first approach typically looks like this: a team selects an AI-powered demand planning tool, runs it alongside their existing forecast for a few months, gets results that look promising in a controlled test environment, and then struggles to integrate those results into their actual planning process. The forecast might be better in theory, but if the planning process doesn't change, if planners don't trust the output, if the data feeding the model isn't maintained to the standard the model requires, the theoretical improvement never translates into operational results.

The organisations that succeed with AI start from the other end. They identify the operational decisions that matter most and are currently made poorly — where is the most value being lost? — and then ask whether AI can improve those decisions. Sometimes the answer is yes. Sometimes the answer is that better data, cleaner processes, or more disciplined governance would improve the decision without any AI at all. Either way, the improvement is grounded in a real operational problem rather than a technology aspiration.

Pattern two: the data foundation isn't there

If there is one theme that dominates every credible analysis of AI failure in supply chains, it's data. Not as an abstract concept, but as a practical, operational blocker.

PwC's survey identified data issues as the second most common reason technology investments fail to deliver (44% of respondents). BCG research found that 61% of supply chain leaders cite poor data quality and system integration as the top barriers to successful AI implementation. Gartner's research confirmed that data availability and quality remain among the top challenges regardless of AI maturity — identified by 34% of leaders in low-maturity and 29% in high-maturity organisations.

In Australian supply chains, the data challenge has several dimensions that are worth spelling out because they're often underestimated.

Fragmented systems. Most organisations of any scale run multiple systems across their supply chain — an ERP (sometimes more than one, particularly after acquisitions), a WMS, a TMS, procurement systems, demand planning tools, spreadsheets that bridge the gaps, and manual processes that bridge the gaps between the spreadsheets. Each system holds part of the picture. AI models need the full picture. Getting data out of these systems, aligning it, cleaning it and maintaining it at the quality the model requires is not a small task. It's a foundational infrastructure project — and most organisations underinvest in it dramatically relative to what they spend on the AI tool itself.

Master data quality. Inaccurate lead times, incorrect unit conversions, outdated supplier records, inconsistent product hierarchies, duplicated customer records — these are mundane, unglamorous problems. But they directly determine whether an AI model produces useful output or garbage. When a demand forecasting model is trained on historical data where product hierarchies changed mid-period, where promotions weren't flagged consistently, where returns were recorded inconsistently, or where intercompany transfers contaminated demand signals — the forecast will be wrong in ways that are hard to diagnose and expensive to fix.

Historical data gaps. Machine learning models learn from patterns in historical data. If you've only been tracking a particular metric for twelve months, or if your data from two years ago is unreliable because it predates a system migration, or if a pandemic scrambled your demand patterns so thoroughly that three years of history are essentially unusable — the model has less to work with than you think. This is particularly relevant for Australian businesses with seasonal demand patterns that differ from Northern Hemisphere norms, or those with relatively short data histories in newer product categories.

Integration costs. The ASCLA highlighted that data quality across fragmented systems remains a persistent challenge for logistics operations, noting that when AI tries to optimise across disconnected sources, decisions can be dangerously flawed. ABI Research's 2025 supply chain survey recommended that organisations expect integration and change costs to exceed computational spend by a factor of 1.2 to 2.5 times. That's not the AI platform licence — that's the unglamorous plumbing work required to make the platform actually function.

The practical implication is straightforward but frequently ignored: the data foundation work needs to happen before the AI investment, or at minimum concurrently with a realistic assessment of how long it will take. Organisations that treat data quality as a problem to be solved during implementation rather than a prerequisite for implementation consistently overshoot timelines and budgets.

Pattern three: bolting AI onto broken processes

This one is subtle and therefore dangerous. An organisation has a demand planning process that doesn't work particularly well — forecasts are inaccurate, planners override the statistical baseline without discipline, the S&OP process is more presentation than decision-making, and inventory parameters haven't been reviewed in two years. They invest in an AI-powered planning platform expecting it to fix the problem.

It doesn't. The AI tool generates a better statistical baseline, but the same planners override it with the same undisciplined adjustments. The improved forecast feeds into the same S&OP process that doesn't connect demand to supply to finance. The inventory parameters are still wrong, so the system optimises around incorrect targets. The result: an expensive new platform producing results that aren't materially better than what came before.

KPMG identified this explicitly in their analysis of AI in supply chains: true scalability requires clean data, standardised processes, and disciplined governance. The technology layer sits on top of a process layer, which sits on top of a data layer. If any layer is broken, the layers above it can't function properly.

This is where the process design and transformation work becomes essential — not as a separate initiative from AI adoption, but as an integral part of it. The organisations that get value from AI in demand planning, for example, typically redesign their planning process concurrently: clarifying who owns the forecast, establishing governance for overrides, connecting the demand plan to supply and financial plans, and defining how the AI output is used in decision-making. The technology and the process change together.

Pattern four: the pilot trap

The data on this is unambiguous. Organisations are launching AI pilots at scale, but very few of those pilots reach production.

McKinsey's 2025 survey found that the majority of organisations are still in experimenting or piloting stages. BCG reported that 90% of vertical, function-specific AI use cases remain stuck in pilot mode. The MIT study's finding that 95% of enterprise AI pilots deliver zero measurable return isn't about pilot failure in the technical sense — many pilots "work" — it's about the gap between a successful pilot and a scaled deployment that delivers measurable business value.

The pilot trap works like this. An organisation identifies a use case, selects a vendor or builds something internally, runs a proof of concept in a controlled environment, and demonstrates that the model can produce better results than the current approach. The pilot is declared a success. Then the project stalls.

It stalls because the pilot environment was curated — clean data, engaged users, limited scope. Production is messy — inconsistent data quality, users who weren't involved in the pilot, integration with live systems, edge cases the pilot didn't encounter, organisational resistance from people whose workflows are being changed. The gap between pilot and production is not a small step — it's often a larger, more expensive, and more complex undertaking than the pilot itself.

The SupplyChainBrain analysis noted this directly: buying AI tools from specialised vendors succeeds about 67% of the time, compared to about one-third for internal builds. But even for vendor solutions, success depends on choosing the right product, configuring it for the business, and adopting it strategically — starting narrow and scaling from there rather than building something generic.

For Australian organisations, this has a practical implication. Don't conflate a successful pilot with a successful implementation. Budget and plan for the pilot-to-production journey explicitly, including data integration, process redesign, change management, user training, and ongoing model maintenance. If the business case only works at production scale, ensure there's a realistic plan and budget to get from pilot to production — or don't start the pilot.

Pattern five: underinvesting in people and change

Gartner predicts that by 2028, 60% of supply chain digital adoption efforts will fail to deliver promised value due to insufficient investment in learning and development. Their survey of 579 supply chain practitioners found that 58% identified rapid tech advancement as a major future challenge, while 40% believed hyperautomation was evolving skills requirements faster than they could adapt.

This is not a soft problem. It's the difference between success and failure.

When an AI tool changes how a planner does their job — from manually building a forecast in a spreadsheet to reviewing, adjusting and approving an AI-generated forecast — that's not a small change. It requires the planner to develop new skills: understanding what the model is doing well enough to know when to trust it and when to override it, interpreting confidence intervals and scenario outputs, managing by exception rather than rebuilding from scratch. If the planner doesn't develop those skills, they'll either ignore the AI output (making the investment worthless) or accept it blindly (introducing new risks).

The same applies across warehouse operations, procurement, logistics planning and supplier management. AI changes the nature of work in each of these functions. People need new skills, new workflows, and enough confidence in the technology to use it effectively. That requires deliberate investment in training, ongoing support, and an organisational culture that embraces experimentation and learning.

McKinsey's research confirmed that high-performing AI organisations are nearly three times more likely than others to fundamentally redesign their workflows. They don't just add AI to existing processes — they rethink how work is done. And high performers are more likely to have defined processes to determine how and when model outputs need human validation. The human-in-the-loop design is intentional, not an afterthought.

In the Australian context, where supply chain talent is already constrained — the ASCLA reports ongoing shortages of skilled drivers, warehouse staff and logistics professionals — the workforce dimension of AI adoption deserves even more attention. Organisations aren't just competing for AI-literate talent; they're asking existing teams with full workloads to adopt new ways of working while maintaining operational performance. That requires investment in capability building and organisational design that goes well beyond a vendor training session.

Pattern six: no clear success metrics

A surprisingly common contributor to AI "failure" is that nobody defined what success looks like before the project started.

Gartner's research found that mature AI organisations were more likely to define performance metrics early in the ideation phase of every use case. PwC identified unclear objectives and weak business rationale as one of the least-selected reasons for technology investment failure — which they interpret not as evidence that objectives are clear, but as a blind spot. Organisations don't realise their objectives are unclear until the project is underway and nobody can agree on whether it's working.

In supply chain, this plays out in several ways. A demand forecasting project might improve forecast accuracy at a national level but not at the SKU-location level where inventory decisions are actually made. An inventory optimisation project might reduce total stockholding but increase stockouts on critical lines. A logistics optimisation project might reduce total transport cost but create service problems for premium customers. Whether any of these outcomes counts as "success" depends entirely on what was defined upfront.

The most effective approach is brutally specific. Before committing to an AI project, define the operational metric you're trying to improve, the baseline you're measuring from, the target improvement, and the timeframe. If you can't do that — if the best you can say is "we want to use AI to improve our supply chain" — you're not ready to invest. You need a diagnostic first.

Pattern seven: trying to do too much at once

The final pattern is ambition that exceeds organisational capacity. An organisation commits to an enterprise-wide AI transformation — new demand planning, new inventory optimisation, new logistics routing, new supplier risk monitoring, all at once. The scope is overwhelming, the change burden on the organisation is unsustainable, and the project collapses under its own weight.

The evidence strongly favours a focused, sequential approach. The SupplyChainBrain analysis recommended starting with a single AI application and then adding more over time, building a solid foundation before expanding capabilities. Infios' leadership team noted that most AI initiatives fail because companies chase tools rather than solving specific problems. Supply chain AI trends research from Dataiku emphasised that organisations should start narrow, prove value, then scale — not attempt broad transformation from day one.

For Australian supply chain teams, the sequencing question is particularly important. Resources are constrained. Operational demands are relentless. Tariff disruption, Scope 3 reporting deadlines, and cost pressures are all consuming management bandwidth simultaneously. Trying to run a multi-workstream AI transformation on top of these priorities is a recipe for failure.

A phased approach works better. Start with one use case where the business case is clear, the data is relatively clean, and the process is well understood. Deliver measurable results. Build internal confidence and capability. Use the lessons from the first deployment to inform the second. This is slower than the ambition suggests, but it's how value actually gets captured.

What the organisations that succeed are doing differently

The patterns above describe how AI projects fail. The inverse — what successful organisations do — is equally instructive, and the evidence is remarkably consistent.

They start with operational problems, not technology. The organisations capturing value from AI identified specific decisions that were being made poorly, quantified the cost of those poor decisions, and then evaluated whether AI could improve them. The technology selection followed the problem definition, not the other way around.

They invest heavily in data foundations. Not as a separate initiative, but as an integral part of AI deployment. They clean master data, build integration pipelines, establish data governance, and create processes to maintain data quality over time. They treat this work as ongoing investment, not a one-time project.

They redesign processes concurrently. McKinsey's high performers are three times more likely to fundamentally redesign workflows. They don't bolt AI onto broken processes. They redesign the process to take advantage of AI capabilities — establishing new roles, new governance, new decision rights, and new ways of working.

They invest in their people. Training, change management, capability building, and ongoing support. They recognise that AI changes the nature of supply chain work and invest in helping their teams make that transition. They deploy human-in-the-loop controls deliberately, building trust in the technology over time.

They define success metrics upfront. Specific, measurable, operationally grounded. They know what they're trying to achieve before they start, and they measure progress against those targets relentlessly.

They start narrow and scale deliberately. One use case, well-executed, delivering measurable results. Then a second, informed by lessons from the first. This isn't lack of ambition — it's recognition that sustainable AI value comes from building organisational capability alongside technical capability.

They have senior leadership commitment. McKinsey found that high performers tend to have senior leaders who demonstrate strong ownership and commitment to AI initiatives. AI adoption is not delegated to IT or an innovation team — it's owned by the supply chain leadership as an operational priority.

What this means for Australian supply chain leaders

The Australian supply chain landscape creates both specific challenges and specific opportunities for AI adoption.

The challenges are real. Fragmented systems from years of organic growth and acquisition. Data quality issues that predate the AI conversation by a decade. Talent constraints that make it hard to build internal AI capability. Operational complexity driven by Australia's geography — vast distances, concentrated coastal populations, long inbound lead times, and relatively small domestic markets compared to the infrastructure required to serve them. Regulatory complexity, from Scope 3 reporting to modern slavery to product safety, consuming management attention that might otherwise go to technology transformation.

But there are opportunities too. Australian labour costs are among the highest in the world, which means AI tools that improve productivity have a shorter payback period than in lower-cost markets. The country's geographic isolation creates structural supply chain complexity — long lead times, concentrated shipping lanes, vulnerability to disruption — that AI-powered planning and risk management tools are well-suited to address. And the relatively concentrated nature of many Australian industries means that organisations that get AI right can establish meaningful competitive advantages quickly.

The practical question for most Australian supply chain leaders is not whether to invest in AI — it's how to invest in a way that actually delivers results rather than adding to the 85% failure rate.

Based on everything we've seen — in the research, in client engagements, in the pattern of successes and failures across the market — here's what we'd recommend.

Start with a diagnostic. Before committing to any AI investment, understand where your supply chain is actually losing value. Map your end-to-end supply chain costs, processes and capabilities. Identify the decisions that matter most and are currently made most poorly. Quantify the opportunity. This creates the fact base for prioritising AI investments — and it often reveals that the highest-value improvements don't require AI at all. Better data, cleaner processes, more disciplined governance, and stronger procurement practices can deliver substantial value without any algorithms.

Assess your readiness honestly. Data quality, system integration, process maturity, workforce capability — evaluate these honestly, not aspirationally. If your master data is unreliable, your first investment should be in fixing it. If your planning process doesn't work, redesign the process before layering AI on top of it. If your team doesn't have the skills to work with AI tools, invest in capability building before you invest in the tools.

Choose one use case to start. Not the most ambitious one — the one with the clearest business case, the best data foundation, and the most receptive team. In most Australian supply chains, demand forecasting or inventory optimisation are strong starting points because the data typically exists (even if it needs cleaning), the business case is quantifiable (working capital, service levels, waste reduction), and the technology is relatively mature.

Plan for the full journey. Budget and plan not just for the pilot, but for the pilot-to-production transition, the integration work, the process redesign, the change management, and the ongoing model maintenance. If the total cost of that journey exceeds what the organisation can support, reduce the scope rather than reducing the investment in the things that determine success.

Get independent advice on technology selection. The supply chain technology market is crowded, vendor claims are aggressive, and the difference between a platform that's right for your organisation and one that isn't can be the difference between success and a multi-year, multi-million-dollar disappointment. Independent advisory — from a firm that doesn't sell software and doesn't have vendor partnerships that create conflicts — is worth the investment. Trace is independent of all technology vendors, which means our recommendations are based entirely on what's right for your organisation.

Connect AI to your broader supply chain strategy. AI isn't a standalone initiative. It connects to your network design, your procurement strategy, your planning processes, your warehousing and distribution operations, your resilience planning, and your workforce strategy. The organisations that get the most value from AI are those that integrate it into a coherent supply chain improvement program rather than treating it as a separate technology project.

The role of advisory in AI adoption

There's an irony in the supply chain AI market that's worth naming. The same organisations that are underinvesting in data foundations, process design and change management — the things that determine whether AI projects succeed — are overinvesting in technology licences and vendor engagements. They're spending millions on platforms and relatively little on the advisory work that determines whether those platforms deliver value.

This isn't an argument for hiring consultants instead of buying technology. It's an argument for getting the sequence right. A well-run diagnostic, a clear readiness assessment, a disciplined use case prioritisation, a structured technology selection process, a thoughtful process redesign, and a proper change management program — these are the things that determine whether an AI investment succeeds or fails.

At Trace, this is exactly the work we do. We're a specialist supply chain and procurement consulting firm — it's all we do, which gives us the operational depth that generalist firms and technology vendors can't match. We're independent of all technology vendors, which means we can assess platforms, evaluate vendors and make recommendations based entirely on what's right for your organisation. And we understand how AI fits into the broader supply chain context — because we work across strategy and network design, procurement, planning and operations, warehousing and distribution, organisational design, technology advisory, and resilience.

We've written previously about how to adopt AI practically in your supply chain and about the seven things that need to be right before the technology matters. This article adds the failure patterns — because understanding what goes wrong is often more instructive than understanding what to aim for.

The bottom line

The supply chain AI market is not going away. The technology is getting better. The use cases are getting clearer. The competitive pressure to adopt is intensifying. Gartner predicts that 70% of large organisations will adopt AI-based supply chain forecasting by 2030. BCG reports that agentic AI systems already account for 17% of total AI value and are projected to reach 29% by 2028.

But the gap between AI investment and AI results is a real and present problem. Four in five organisations are seeing no tangible bottom-line impact from their AI initiatives. Ninety-five percent of pilots deliver zero measurable return. The organisations that are capturing value are doing fundamentally different things — not buying better technology, but building better foundations, redesigning processes, investing in people, and approaching AI with the same operational discipline they apply to every other aspect of supply chain management.

The window for getting this right is narrowing. Organisations that build the foundations now — clean data, sound processes, capable people, clear strategies — will be positioned to capture genuine value from AI as the technology continues to mature. Organisations that keep launching pilots without addressing the foundations will keep getting the same disappointing results, while falling further behind competitors who've figured out that AI success is an operational challenge, not a technology one.

If you're an Australian supply chain or operations leader thinking about your AI strategy — or wondering why your current AI investments aren't delivering the results you expected — we'd welcome the conversation. Get in touch and let's talk about what's really going on and what to do about it.

Trace Consultants is an Australian supply chain and procurement consulting firm working across FMCG and manufacturing, retail and consumer, resources and energy, health and human services, and government and defence. We help organisations navigate the intersection of supply chain strategy, technology adoption and operational improvement — with the independence and depth that comes from doing nothing else. Visit our insights page for more articles on the challenges shaping Australian supply chains.

‍

There's a widening gap in Australian supply chains right now, and it's not the one you'd expect. It's not between organisations that have AI and those that don't. It's between organisations that have deployed AI and are getting genuine operational value from it, and organisations that have deployed AI and are quietly wondering why the results don't match the business case.

The Australian Government's National AI Centre tracks AI adoption across the economy, and its Q1 2025 data tells an instructive story. When asked whether AI had helped with supply chain and supplier management, only 14% of surveyed businesses said "definitely." Another 46% said "possibly" — the kind of answer you give when something is technically in place but the results aren't clear enough to point to. The remaining 40% said "unlikely." Meanwhile, industry data from the Supply Chain and Logistics Association of Australia suggests nearly 40% of supply chain leaders report measurable improvements from AI implementation — which means 60% haven't captured those benefits yet.

These numbers don't reflect a technology failure. The AI tools available today — for demand forecasting, inventory optimisation, logistics routing, predictive maintenance, warehouse automation and supply risk management — are genuinely capable. The algorithms work. The platforms are maturing. The cloud infrastructure to run them is accessible and affordable.

What these numbers reflect is an execution failure. Organisations are adopting AI without adequate preparation in the areas that actually determine whether it works: understanding their own readiness, selecting the right use cases, redesigning processes to consume AI outputs, choosing the right technology for their context, running pilots that test the full operating model, building the data foundations that AI depends on, and developing the human capability to work alongside AI tools effectively.

These seven disciplines aren't optional extras that you bolt on after the technology is deployed. They're the work that determines whether your AI investment delivers a 15% improvement in forecast accuracy and a $30 million reduction in inventory — or becomes a line item that finance questions at every budget review.

This article covers all seven, in the order they need to happen, with practical guidance for Australian supply chain and operations leaders who want to get this right.

1. Supply Chain AI Readiness Assessment

Every AI journey should start with an honest answer to a question most organisations skip: are we actually ready for this?

Readiness isn't about whether you can buy an AI tool. Of course you can — the market is flooded with options. Readiness is about whether your organisation has the foundations in place to deploy AI in a way that produces reliable outputs, integrates into operational decision-making, and sustains value over time.

A genuine AI readiness assessment evaluates five dimensions.

Data maturity. AI is only as good as the data it learns from and operates on. For demand forecasting, that means clean, granular historical transaction data at the right level of detail — typically SKU-location-week — with promotional activity, pricing changes, new product introductions and other demand-shaping events accurately captured. For inventory optimisation, it means reliable lead time data, supplier performance history and demand variability metrics. For logistics optimisation, it means accurate delivery windows, vehicle constraints, cost structures and geographic data.

Most Australian organisations have this data somewhere in their systems. The question is whether it's accessible, consistent, integrated and trustworthy. Data quality issues, system fragmentation, inconsistent master data and manual workarounds are the norm. A readiness assessment quantifies where you actually stand — not where you think you stand — against the data requirements of specific AI use cases.

Process maturity. AI tools produce recommendations, forecasts, optimisation outputs and alerts. Those outputs only create value if there's a business process that consumes them. If your S&OP process is dysfunctional, a better demand forecast won't fix it. If your warehouse has no standard operating procedures for pick path management, an AI-optimised slotting recommendation won't translate to throughput improvement. If your procurement team doesn't have a structured approach to supplier risk management, an AI early warning system will generate alerts that nobody acts on.

Process maturity assessment looks at the planning and operations processes that would need to consume AI outputs: demand planning, supply planning, inventory management, S&OP, logistics planning, warehouse management, procurement execution. For each, it evaluates whether the process is defined, followed, measured and governed — because these are the prerequisites for AI integration.

Technology landscape. What systems are currently in place? What's the ERP platform? Is there a warehouse management system? A transport management system? An existing planning tool? What are the integration points and constraints? What's the cloud posture? These questions determine what's technically feasible, what integration work is required, and whether AI tools can plug into the existing architecture or need a parallel infrastructure.

Organisational capability. Do your planners, analysts and operations managers have the skills to work with AI tools? Not to build models — that's a different skill set — but to interpret AI outputs, manage exceptions, configure parameters, and know when to trust the model versus override it. The readiness assessment should evaluate current capability levels across the roles that will interact with AI, identify gaps, and flag the training and development investment required.

Governance and culture. Does the organisation have a framework for managing AI responsibly? This includes data governance (who owns data quality, who authorises data use), model governance (who monitors performance, who recalibrates, who decides when a model should be retired), decision governance (who is accountable when AI-informed decisions go wrong), and cultural readiness (is the organisation open to changing established ways of working based on AI-generated insights, or will there be resistance?).

The Australian Government's 2025 Guidance for AI Adoption sets out six essential practices for responsible AI governance, reflecting a maturing regulatory environment that organisations need to align with. The readiness assessment should evaluate current governance maturity against these expectations.

What the output looks like. A well-executed readiness assessment produces a clear picture of where the organisation stands across all five dimensions, with specific gaps identified and prioritised. It should include a heat map of readiness by AI use case — showing which applications are ready to pursue now, which need foundation work first, and which should be deferred until earlier initiatives have matured. This becomes the basis for a sequenced AI roadmap that's grounded in reality rather than aspiration.

The organisations that skip this step — that jump straight to technology selection or pilot execution — almost always find themselves circling back to readiness issues 6-12 months later, having spent budget and credibility on initiatives that underperformed because the foundations weren't in place.

2. AI Use Case Identification and Business Case Development

With a clear readiness picture in hand, the next step is identifying which AI applications to pursue and building the business cases to justify the investment.

This sounds straightforward. It rarely is. The challenge isn't a shortage of potential use cases — it's an overabundance. Every AI vendor, every conference presentation, every industry report suggests dozens of ways AI could be applied across the supply chain. The risk is either trying to do too many things at once (spreading resources thin and delivering nothing well) or picking the wrong things to start with (choosing technically interesting applications that don't address the organisation's most material operational problems).

Starting with the operational problem, not the technology. The most reliable approach to use case identification starts with a simple question: which operational decisions are currently made most poorly, and which have the largest financial impact?

In supply chain and operations, the decisions that typically offer the highest leverage for AI-assisted improvement include demand forecasting (where machine learning can materially improve accuracy for products with complex demand patterns driven by promotions, seasonality, weather or external signals), inventory optimisation (where AI-driven parameter setting can improve service levels while reducing working capital — particularly valuable in the current interest rate environment), logistics and route optimisation (where AI can reduce transport costs by 5-15% through better vehicle allocation and delivery sequencing — particularly impactful given Australian distances), supply risk and disruption management (where AI can monitor external data sources to provide early warning of disruptions), warehouse operations (from predictive slotting through to demand-driven labour scheduling), and predictive maintenance (where AI can reduce unplanned downtime by identifying equipment failure patterns).

Each of these is a broad category. The use case identification process should drill into the specific version of each that's relevant to the organisation: which product segments have the worst forecast accuracy? Which inventory positions are structurally wrong? Which logistics lanes have the most inefficiency? Where are the biggest maintenance cost drivers?

Quantifying the opportunity. For each candidate use case, the business case needs to quantify the current cost of the problem (excess inventory, lost sales from stockouts, transport cost premium, unplanned downtime cost), the realistic improvement AI can deliver (based on benchmarks, published research and the organisation's specific data characteristics — not vendor claims), the investment required (technology, implementation, data preparation, process redesign, training, ongoing support), and the payback period and return profile.

This is analytical work that requires deep understanding of supply chain economics — understanding total cost of ownership for inventory, the service-level-to-stock-level trade-off, the relationship between forecast accuracy and safety stock requirements, the cost structure of logistics operations. It's strategy work as much as technology work.

Prioritisation. With quantified business cases for multiple use cases, prioritisation should consider both financial attractiveness and readiness. The best starting point is a use case that has high financial impact and where the organisation's data, process and capability foundations are strong enough to support successful deployment. Starting with a lower-impact but higher-readiness use case can also make sense if it builds confidence and capability that enables more ambitious applications later.

The output should be a prioritised portfolio of 3-5 use cases with detailed business cases, clear sequencing, identified dependencies, and a realistic timeline. This becomes the investment case for the AI program — and the accountability framework against which results are measured.

3. AI-Enabled Planning and Operations Process Design

This is where most AI initiatives fall down, and it's the area that gets the least attention. Organisations invest heavily in selecting and implementing AI technology, then try to insert it into planning and operations processes that were designed for a pre-AI world. The result is predictable: the AI tool produces outputs that don't fit the existing workflow, planners don't know what to do with them, and the system either gets ignored or creates more work rather than less.

AI-enabled process design means redesigning the planning and operations processes — not just adding AI as an input — so that the entire workflow takes advantage of what AI makes possible.

What changes when AI enters the planning process. Consider demand planning as an example. In a traditional process, planners spend the majority of their time generating the statistical baseline forecast, then adjusting it based on market intelligence, promotional plans, sales team input and other qualitative factors. The S&OP cycle revolves around reviewing and reaching consensus on the forecast.

When AI handles the statistical forecasting — and does it better, faster and more granularly than manual methods — the planner's role fundamentally changes. They shift from forecast generation to forecast management: reviewing AI outputs, focusing attention on the exceptions and outliers where human judgement adds value, incorporating market intelligence that the model can't access, and making the cross-functional trade-off decisions that require business context.

This is a better use of skilled planners' time. But it requires a redesigned process that defines how AI forecasts are generated and reviewed, what exception thresholds trigger human intervention, how the planner's adjusted forecast feeds into the S&OP cycle, what governance ensures the AI model stays calibrated, and how forecast accuracy is measured and reported.

The same principle applies across other domains. AI-optimised inventory parameters need a process for planner review and override. AI-generated route plans need a process for driver exceptions and real-time adjustment. AI supply risk alerts need an escalation and response framework.

Designing the target operating model. For each AI use case, the process design work should define the end-to-end workflow (how does the AI-enabled process work, step by step, from data input through to operational decision?), the role of the planner or operator (what does the human do that the AI doesn't, and how do they interact with the system?), the exception management framework (what triggers human intervention, and what's the process for handling exceptions?), the governance rhythm (how often are AI outputs reviewed, who is accountable for model performance, what are the escalation paths?), and the performance measurement framework (how do we know the AI-enabled process is delivering better outcomes than the previous approach?).

This is organisational design work applied to AI-enabled operations. It requires understanding both the technical capabilities of the AI tool and the operational realities of the planning or operations environment — how planners actually work, what information they need, where they add value, and what frustrates them about current processes.

The S&OP connection. For organisations with a formal S&OP or integrated business planning process, AI-enabled planning has significant implications. AI can accelerate the demand planning phase, improve the quality of scenario modelling, provide more granular inventory trade-off analysis, and enable faster response to demand or supply changes between formal planning cycles. But these benefits only materialise if the S&OP process is redesigned to take advantage of them — if the meeting cadence, the information packs, the decision frameworks and the accountability structures all reflect the new AI-enabled capability.

Trace has written extensively about S&OP and planning process design — organisations looking at this intersection should also review our thinking on planning and operations process maturity and our guide to Advanced Planning Systems which covers how APS capabilities integrate with planning processes.

4. AI Technology Selection and Vendor Advisory

With clear use cases, quantified business cases and a target process design, the organisation is ready to evaluate technology options. Not before.

The sequence matters because technology selection should be driven by defined requirements — not the other way around. Too many organisations start with a technology shortlist and then try to find use cases that justify the purchase. This leads to solutions looking for problems, features that don't map to actual decision-making needs, and implementations that are technically impressive but operationally irrelevant.

The AI technology landscape for supply chain and operations. The options fall into four broad categories.

Embedded AI within existing platforms. Your ERP vendor's demand planning module may now include ML-based forecasting. Your WMS might offer AI-powered slotting optimisation. Your TMS may have added AI routing capabilities. These embedded capabilities are typically the easiest to deploy (no new integration required) but may be less sophisticated than specialist tools.

Best-of-breed AI point solutions. Standalone platforms focused on specific use cases — demand sensing, inventory optimisation, logistics route optimisation, predictive maintenance, supply risk monitoring. These typically offer deeper functionality for their specific domain but create integration requirements and add another vendor relationship to manage.

Advanced Planning Systems with native AI. The dedicated supply chain planning platforms — Kinaxis, Blue Yonder, o9 Solutions, SAP IBP, RELEX, Logility and others — increasingly embed AI and ML capabilities across their planning modules. For organisations considering a broader planning transformation, these platforms offer comprehensive capability but require significant implementation investment.

General-purpose AI and analytics platforms. Cloud ML platforms (AWS SageMaker, Google Cloud AI, Azure ML) and business intelligence tools with predictive capabilities can be configured for supply chain use cases. These offer maximum flexibility but require more internal technical capability to deploy and maintain.

How to evaluate. The evaluation approach should mirror the structured RFx process we recommend for any significant technology selection, but with specific adaptations for AI.

First, evaluate against defined use cases, not feature lists. The question isn't "does this platform support demand forecasting?" (they all do). It's "does this platform's forecasting capability handle our specific demand patterns — high promotional variability, long-tail SKUs, new product introductions with no history — better than alternatives?"

Second, insist on demonstrations with representative data. Canned demos tell you nothing about how the system will perform in your environment. Provide shortlisted vendors with a sample dataset that reflects your real data characteristics and ask them to demonstrate their system's outputs against defined scenarios.

Third, evaluate the implementation approach and partner ecosystem as heavily as the software. For AI tools specifically, the quality of model configuration, data engineering and calibration during implementation determines the majority of the outcome. A superior algorithm badly implemented will underperform an adequate algorithm well implemented.

Fourth, assess total cost of ownership, not just licensing. Implementation services, data engineering, integration development, training, change management and ongoing model monitoring and recalibration typically represent 60-70% of the total investment.

Fifth, evaluate local capability. For Australian organisations, the vendor's or partner's presence and experience in the ANZ market matters. Long inbound supply chains, Australian seasonal patterns, concentrated retail customers and specific regulatory requirements create a context that global vendors may not have deep experience with.

The procurement disciplines of structured evaluation, commercial benchmarking and negotiation are just as important for AI technology selection as for any other significant purchase — perhaps more so, given the complexity and the risk of vendor lock-in.

5. AI Pilot Design and Execution Support

Before committing to full-scale deployment, most organisations benefit from a structured pilot that tests whether the AI tool, configured for their data and integrated into their process, actually produces better operational outcomes than the current approach.

The key word is "structured." An AI pilot isn't a sandbox experiment where data scientists play with models to see what's possible. It's a controlled operational test with defined scope, clear success metrics, baseline measurement, and a rigorous evaluation framework that determines whether to scale, adjust, or stop.

Designing the pilot. A well-designed AI pilot defines several critical elements.

Scope. Which products, which locations, which planning horizon, which operational process? The scope should be large enough to be representative but small enough to be manageable. For a demand forecasting pilot, this might mean a specific product category across a defined set of locations over a 12-16 week period. For an inventory optimisation pilot, it might mean a subset of SKUs in a defined part of the distribution network.

Baseline. What does "current performance" look like, measured rigorously? If you're testing AI-driven demand forecasting, you need a clean baseline of current forecast accuracy by SKU, by location, by time horizon — measured consistently over a period long enough to account for normal variability. Without a solid baseline, you can't measure improvement.

Success metrics. What does "better" look like, quantified? A 10% improvement in forecast accuracy at SKU-week level? A 15% reduction in safety stock without service degradation? A 7% reduction in transport cost per delivery? The metrics should be defined before the pilot starts, agreed with stakeholders, and directly traceable to the business case.

Operating model for the pilot period. How will planners or operators interact with the AI tool during the pilot? Will they use AI outputs as their primary input (replacing the current approach) or run in parallel (comparing AI outputs against current methods)? Parallel running is safer but less realistic — planners who know they have a fallback behave differently than planners who are depending on the new tool. The pilot design should specify the operating model clearly.

Duration. The pilot needs to run long enough to test performance across different demand conditions — not just steady-state weeks but promotional periods, seasonal transitions, supply disruptions and other real-world variability. For most supply chain use cases, 12-16 weeks is a reasonable minimum.

Evaluation framework. How will the pilot be evaluated? Who reviews the results? What threshold of improvement justifies scaling? What happens if results are mixed — some metrics improve, others don't? The evaluation framework should be defined upfront, not designed retrospectively to fit the results.

Testing the full operating model. The most important — and most frequently missed — aspect of pilot design is testing the full operating model, not just the technology. This means evaluating whether planners are actually using the AI outputs in their decision-making, whether the process changes are working as designed, whether exceptions are being managed appropriately, and whether the results are genuinely better than the current approach across the full range of real-world conditions.

A pilot that demonstrates impressive algorithmic accuracy in a test environment but doesn't test whether planners trust and use the outputs is answering the wrong question. The question isn't "can this model forecast well?" It's "does this model, used by our planners, in our process, with our data, produce better operational decisions than what we do today?"

The scale decision. At the end of the pilot, the organisation faces a clear decision: scale, adjust, or stop. If the pilot demonstrates clear, measurable improvement against the defined success metrics — and the operating model is working — the case for scaling is strong. If results are mixed, the pilot data should inform what needs to change before scaling (data quality issues? process gaps? model configuration? planner training?). If results are poor despite good execution, that's also valuable information — it means this particular use case, with this particular tool, in this particular context, doesn't deliver the expected value.

Honest evaluation at this stage — rather than confirmation bias that justifies the investment already made — is what separates organisations that deploy AI effectively from those that scale failures.

6. Data Readiness and Foundation Work

If there's one section of this article that deserves to be read twice, it's this one. Data readiness is the single largest determinant of AI success in supply chain and operations, and it's the area where organisations most consistently underinvest.

Every AI model — whether it's forecasting demand, optimising inventory, routing deliveries or predicting equipment failures — depends on data. The quality, completeness, granularity and accessibility of that data determines the ceiling of what AI can achieve. No algorithm, however sophisticated, can overcome fundamentally poor data.

What "data ready" actually means. For supply chain AI applications, data readiness typically requires several things.

Clean historical transaction data. For demand forecasting, this means order or shipment history at the right level of granularity (SKU-location-day or week), with anomalies identified and addressed (one-off bulk orders, data entry errors, system migration artefacts), and sufficient history to train models effectively (typically 2-3 years minimum, more for seasonal products).

Accurate demand-shaping event data. Promotions, pricing changes, new product introductions, range deletions, competitor activity — the events that cause demand to deviate from underlying patterns. AI models can learn from these events, but only if they're captured accurately in the data. Most organisations have promotional calendars, but the linkage between promotions and transaction data is often incomplete or inaccurate.

Reliable master data. Product hierarchies, location hierarchies, supplier data, customer segmentation, lead times, minimum order quantities, shelf life constraints — the reference data that AI models use to structure their analysis. Master data quality is a pervasive challenge in Australian organisations, and it directly impacts AI model performance.

Integration and accessibility. Data often sits in multiple systems — ERP, WMS, TMS, point of sale, CRM, external data sources — and needs to be brought together in a form that AI tools can consume. This requires data integration pipelines, potentially a data warehouse or data lake, and APIs or connectors to the AI platform.

Timeliness. Some AI applications require near-real-time data (logistics optimisation, warehouse operations), while others work on daily or weekly data cycles (demand planning, inventory optimisation). The data infrastructure needs to support the refresh frequency that each use case requires.

The foundation work. Getting data ready is unglamorous, time-consuming work. It includes data profiling and quality assessment (understanding what you have, where the gaps are, and how severe the quality issues are), data cleansing and enrichment (fixing historical anomalies, filling gaps, enriching transaction data with event data), master data governance (establishing ownership, standards, processes and tools for maintaining data quality on an ongoing basis — because data quality degrades continuously without active management), integration development (building the pipelines that bring data together from multiple sources into a form AI tools can consume), and data architecture design (determining where data lives, how it flows, and what infrastructure supports it).

This work should begin early — ideally during the readiness assessment phase — and continue in parallel with use case development and technology selection. It's the longest-lead-time activity in most AI programs, and it's the one most commonly underestimated.

The ongoing challenge. Data readiness isn't a one-time project. It's a continuous discipline. Data quality degrades as products change, systems are updated, processes evolve and people make mistakes. AI models trained on historical data gradually lose accuracy as the underlying patterns shift. Master data requires ongoing maintenance as the business changes.

This is why data governance — the ongoing organisational capability to maintain data quality — matters as much as the initial data cleansing effort. Without it, the AI investment delivers diminishing returns over time as the data foundations erode.

For organisations across FMCG and manufacturing, retail and consumer, resources and energy and government and defence, data readiness challenges take different forms — but the fundamental principle is the same: invest in the data foundation before investing in the AI tool, and establish the governance to sustain it.

7. AI-Focused Capability Building and Training

The final discipline — and the one that determines whether AI delivers value in year one only or compounds value over time — is building the human capability to work effectively alongside AI tools.

This isn't about turning planners into data scientists. It's about developing a set of practical skills that enable supply chain and operations professionals to get the most from AI-enabled tools and processes.

What AI-literate supply chain professionals need to know. The capability requirements fall into several layers.

Interpreting AI outputs. Understanding what a demand forecast from an ML model represents — including its confidence intervals, its assumptions, and its known limitations. Understanding what an inventory optimisation recommendation means — why the model is suggesting a particular safety stock level, what inputs are driving it, and what would change if assumptions shifted. This isn't deep technical knowledge; it's practical literacy that allows professionals to use AI outputs intelligently rather than either blindly following them or reflexively ignoring them.

Managing exceptions. Knowing when and how to override AI recommendations. AI models work well for the majority of routine decisions but struggle with genuine exceptions — unprecedented events, data quality issues, business context the model can't see. Building the judgement to recognise these situations — and the confidence to intervene — is a critical skill that comes from training, practice and organisational support.

Configuring and monitoring. Understanding how to adjust parameters that the AI tool exposes — service level targets, demand segmentation rules, exception thresholds, scenario assumptions. Knowing how to monitor whether the model is performing as expected and recognising the signs of degradation (declining accuracy, increasing exceptions, outputs that don't align with business reality).

Asking better questions. Perhaps the most valuable capability shift is moving from "generating the answer" (which AI now handles) to "asking better questions." What scenarios should we test? What assumptions should we challenge? What trade-offs should we explore? What risks aren't reflected in the data? This is where experienced supply chain professionals add the most value in an AI-enabled world — and it's a capability that needs to be actively developed, not assumed.

How to build it. Capability building for AI-enabled operations typically involves several components.

Assessment. Understanding the current capability baseline across the roles that will interact with AI tools — planners, analysts, operations managers, supply chain leaders. What's the current level of data literacy? Analytical skill? Comfort with technology? Openness to changing established ways of working?

Training programs. Structured learning that covers the practical skills outlined above — tailored to specific roles and specific AI tools. This isn't generic "AI awareness" training; it's hands-on instruction using the actual systems and processes the team will work with. It should include real data, real scenarios and real decision-making practice.

Playbooks and reference material. Documentation that supports ongoing performance — standard operating procedures for AI-enabled processes, exception management guides, parameter configuration guides, troubleshooting resources. These should be living documents that evolve as the organisation's AI capability matures.

Coaching and support. Particularly in the early stages of AI adoption, planners and operators benefit from accessible support — someone who can help when the model produces an output they don't understand, when they're unsure whether to override a recommendation, or when they encounter a situation the training didn't cover.

Communities of practice. As AI adoption scales across the organisation, connecting practitioners — planners in different sites, analysts in different categories, operators in different facilities — creates a peer learning network that accelerates capability development and shares best practice.

The leadership dimension. Capability building isn't just about frontline users. Supply chain leaders need their own capability development — not in how to use AI tools, but in how to lead AI-enabled organisations. This includes understanding what AI can and can't do (to set realistic expectations), how to interpret AI-related performance metrics (to make informed governance decisions), how to allocate resources between AI investment and other priorities (to make sound trade-offs), and how to create a culture that embraces AI-enabled ways of working while maintaining appropriate scepticism and human oversight.

Strategic workforce planning for AI-enabled supply chains should address this leadership dimension alongside the frontline capability requirements — because leadership buy-in and capability is what sustains AI adoption beyond the initial implementation.

The partnership model: where to build and where to partner

One of the most important strategic decisions for organisations adopting AI in supply chain and operations is what to build internally versus what to source from external partners.

In our view, the answer is clear for most Australian organisations. The capabilities that should be built internally — because they're core to operational performance and need to be sustained over time — are AI literacy across the planning and operations team, process design and governance for AI-enabled operations, data governance and quality management, and performance monitoring and continuous improvement.

The capabilities that are typically better sourced from external partners — because they require specialist skills that aren't needed on a full-time basis — are AI model development and engineering (the deep technical work of building, training and deploying models), technology platform implementation and integration, advanced data engineering for initial setup and migration, and strategic advisory on AI roadmap, technology selection and organisational readiness.

This is where the distinction between AI engineering firms and supply chain advisory firms matters. AI engineering firms bring deep technical capability in building and deploying AI systems — custom model development, agentic AI frameworks, large language model integration, deployment infrastructure. They're the right partner for the model development, platform engineering and technical deployment work. Supply chain advisory firms bring the domain expertise to ensure AI is applied to the right problems, embedded in the right processes, and supported by the right organisational capability.

The most effective AI programs use both: domain experts who understand the operational context and define what needs to happen, partnered with technical experts who know how to make it happen. Neither alone is sufficient. An AI engineering firm without supply chain domain expertise will build technically impressive solutions that don't address the most valuable operational problems. A supply chain advisory firm without AI engineering capability can design the right solution but can't build the models. The partnership model brings both together.

For Australian organisations specifically, this model has a practical advantage. The local talent market for deep AI engineering skills is tight and expensive. Building a permanent internal team of ML engineers, data scientists and AI architects is feasible for the largest organisations but impractical for most. A partnership model lets organisations access world-class AI engineering capability on a project basis while building the internal domain expertise and operational capability that sustains value over time.

How Trace Consultants can help

At Trace Consultants, we sit firmly on the domain expertise side of this partnership model. We don't build AI models. We make sure the AI investments you make actually work in your operations.

Our role in AI adoption spans all seven of the disciplines covered in this article:

AI readiness assessment. We assess your data maturity, process maturity, technology landscape, organisational capability and governance readiness — producing a clear, honest picture of where you stand and what needs to happen before AI can deliver value. This is strategy work grounded in practical supply chain experience.

Use case identification and business case development. We identify the highest-value AI applications for your specific operation, quantify the opportunity with rigour, and build business cases that stand up to scrutiny. Our deep understanding of supply chain economics — from inventory optimisation to logistics cost structures to procurement spend analysis — ensures use cases are anchored in real operational value.

AI-enabled process design. We redesign planning and operations processes to take advantage of what AI makes possible — defining workflows, roles, exception management, governance and performance measurement for AI-enabled operations. This is organisational design work that bridges the gap between technology capability and operational reality.

Technology selection and vendor advisory. We help you navigate the AI technology landscape with an independent, informed perspective — evaluating embedded capabilities, best-of-breed tools and APS platforms against your specific requirements through structured procurement processes. Our technology advisory ensures you select the right tool for your context, not the most impressive demo.

Pilot design and execution support. We design structured AI pilots with clear scope, success metrics, baseline measurement and evaluation frameworks — ensuring you test the full operating model, not just the technology. Our project and change management capability ensures pilots run smoothly and produce the information needed for sound scaling decisions.

Data readiness and foundation work. We support the data quality, master data governance, integration planning and data architecture work that determines whether AI tools produce reliable outputs. This isn't glamorous work — but it's the work that protects your AI investment.

Capability building and training. We develop and deliver practical training programs that build AI literacy across supply chain and operations teams — from frontline planners to senior leaders. Our strategic workforce planning expertise ensures capability development is structured, role-specific and sustainable.

Our independence matters. We don't sell AI software. We don't build AI models. We don't have partnership arrangements with AI vendors that influence our recommendations. Our advice is based entirely on what will deliver the best operational outcome for your organisation — which technology, which use cases, which sequencing, which partnerships.

We work across FMCG and manufacturing, retail and consumer, resources and energy, health and human services, and government and defence — bringing cross-sector perspective on what works in practice, not just in theory.

Getting started

AI in supply chain and operations isn't a future state. It's happening now, in Australian organisations across every sector. But the organisations capturing value aren't the ones with the most ambitious AI strategies or the largest technology budgets. They're the ones that got the foundations right: honest readiness assessment, disciplined use case selection, thoughtful process design, structured technology evaluation, rigorous pilots, serious data work, and genuine investment in human capability.

Every one of those seven disciplines is within reach of any Australian organisation with the commitment to do them properly. The technology will follow — and when it does, it will land on foundations that actually support it.

If your organisation is ready to move beyond AI aspiration and start building the operational capability that makes AI work, we'd welcome the conversation.

‍

Choosing an Advanced Planning System is one of those decisions that looks straightforward on the surface and turns out to be fiendishly difficult in practice. The vendor landscape is broad. The functionality overlaps are significant. Every platform claims AI-powered demand forecasting, scenario modelling, inventory optimisation and seamless integration with your ERP. The demos are polished. The reference customers are carefully curated.

And yet, a significant proportion of APS implementations either fail outright, deliver far less value than the business case projected, or land in a state where the organisation is technically "live" but planners are still running their actual decisions through spreadsheets because the system doesn't work the way they need it to.

The root cause, in most cases, isn't bad technology. It's a selection process that failed to answer the right questions. The RFx was structured as a technology procurement exercise — feature lists, architecture requirements, integration specifications, pricing schedules — when it should have been structured as a planning capability assessment: does this system, configured for our data, our products, our network and our planning process, actually produce better plans than what we're doing today?

That distinction — between buying a technology product and selecting a planning capability — is what separates organisations that get genuine value from their APS investment from those that spend 18 months implementing a system nobody trusts.

Why APS selection is different from other technology procurement

Before diving into how to structure the RFx process, it's worth understanding why demand planning and APS selection doesn't follow the same playbook as procuring an ERP, a WMS, a TMS or most other enterprise technology.

The problem you're solving is harder to define. When you select a warehouse management system, the functional requirements are relatively concrete: receive stock, put it away, pick it, pack it, ship it, count it. The processes are physical, observable and well-understood. When you select a demand planning or supply chain planning system, the requirements are inherently more abstract. You're trying to improve the quality of decisions about uncertain futures — how much to make, where to hold stock, when to order, how to allocate constrained supply. The "right answer" changes daily. The value is in the system's ability to help planners navigate ambiguity, not execute a defined process.

Vendor capabilities genuinely vary in ways that are hard to assess. The Gartner Magic Quadrant for Supply Chain Planning Solutions now includes more than 20 vendors. The Leaders — Kinaxis, Blue Yonder, o9 Solutions, Logility and others — all claim comprehensive functionality across demand planning, supply planning, inventory optimisation, S&OP and production scheduling. But underneath those claims, the actual capabilities differ substantially. Some platforms are genuinely strong in statistical and machine-learning forecasting but weaker in constraint-based supply planning. Others excel at interactive scenario modelling but struggle with the algorithmic optimisation that drives inventory decisions. Some are purpose-built for specific industries — retail, FMCG, process manufacturing — while others are horizontal platforms that require significant configuration to work well in any specific context.

A standard RFx feature checklist — "does the system support promotional uplift modelling? Yes/No" — doesn't surface these differences. Every vendor will tick the box. The question isn't whether the feature exists but whether it works well enough, in your context, with your data, to make materially better decisions than your current approach.

The implementation matters as much as the software. APS aren't systems you install and switch on. They're systems you configure, calibrate and continuously tune — and the quality of that work determines whether the system produces plans worth following. The same platform can deliver transformative results in one organisation and be a $2 million shelf ornament in another, depending on how it was implemented, how the data was structured, how the planning processes were designed, and whether planners were genuinely brought along through the change.

This means the RFx process needs to evaluate not just the software but the implementation approach, the partner ecosystem, the support model and the vendor's track record in environments comparable to yours.

The six mistakes that derail most APS selection processes

Having supported organisations across FMCG and manufacturing, retail and consumer and other sectors through APS selection and implementation, we see the same patterns of failure repeatedly. Understanding them is the first step to designing a process that avoids them.

Mistake one: starting with the vendor shortlist instead of the planning problem

Too many APS selection processes begin with a list of vendors and a request for information rather than a clear articulation of what the organisation actually needs from its planning capability. What are the specific planning decisions that need to improve? Where is forecast accuracy weakest and why? Which inventory positions are structurally wrong — too much in some places, not enough in others? Where are supply and demand decisions disconnected? What does the S&OP process need that it doesn't currently have?

Without clear answers to these questions — grounded in data, not assumptions — the RFx process has no anchor. It becomes a general-purpose technology evaluation rather than a targeted assessment of which system best addresses the organisation's specific planning challenges.

This upfront work — what we'd describe as planning maturity assessment and requirements definition — is the most important phase of the entire selection process. It determines everything that follows: the evaluation criteria, the weighting, the scenarios for demonstration, and ultimately the commercial justification for the investment.

Mistake two: evaluating against feature checklists

The traditional approach to technology RFx involves compiling a requirements matrix — sometimes running to hundreds of rows — with each requirement scored on a scale (e.g. "fully met, partially met, not met, requires customisation"). Vendors complete the matrix, an evaluation panel scores them, and the highest-scoring platform is identified.

This approach works reasonably well for transactional systems with well-defined functional requirements. It works poorly for APS, for two reasons. First, the requirements that matter most in planning — forecast accuracy, scenario modelling quality, user experience for planners, speed of exception management — are qualitative and contextual, not binary. Second, vendors are skilled at completing these matrices in ways that maximise their score without necessarily reflecting how the system will perform in your environment. A planning system might technically "support" promotional uplift modelling, but if it requires three months of custom configuration and a data science team to maintain, that's a very different proposition from one where it's a native, well-tested capability.

Mistake three: accepting canned demos

Every APS vendor has a demonstration environment loaded with sample data and pre-configured scenarios designed to make the system look spectacular. These demos are useful for understanding the user interface and general approach, but they tell you almost nothing about how the system will perform with your data, your products, your demand patterns and your network complexity.

The organisations that make the best APS selection decisions insist on scripted demonstrations using representative samples of their own data. This doesn't mean a full proof-of-concept at the RFx stage — that comes later — but it does mean providing shortlisted vendors with a defined dataset and a set of planning scenarios that reflect your real-world challenges, then evaluating how each system handles them. The difference in insight between a canned demo and a data-driven demonstration is enormous.

Mistake four: underweighting implementation and change

Most RFx evaluation frameworks allocate the majority of their weighting to software functionality and price. Implementation approach, partner capability, change management methodology and ongoing support model might get 10-15% of the total score. This is backwards for APS, where the implementation and change journey typically determines 60-70% of the eventual outcome.

A system with excellent algorithmic capabilities but a weak implementation partner, a rigid configuration approach, or no credible change management methodology is a worse bet than a system with good-enough algorithms backed by a team that understands your industry, your data challenges, and how to get planners actually using the system to make decisions.

Mistake five: treating cost as licensing cost

APS pricing models vary enormously — from per-user licensing to volume-based pricing to flat annual subscriptions with modular add-ons. But the licensing cost is typically 30-40% of the total cost of ownership. Implementation services, data engineering, integration development, testing, training, change management, and ongoing support and optimisation make up the rest.

RFx processes that evaluate cost on licensing alone — or that allow vendors to quote implementation at unrealistically low levels to win the deal — set the organisation up for budget overruns and scope compromises during implementation. The RFx should require vendors and implementation partners to provide a realistic total cost of ownership estimate across a defined horizon (typically five years), including all the components that actually drive spend.

Mistake six: ignoring the planning process and people dimension

An APS is a tool. Its value depends entirely on the planning process it supports and the people who use it. Yet most RFx processes focus almost exclusively on the tool and pay limited attention to whether the organisation is ready to use it effectively.

Questions about planning and operations process design — how will the S&OP process change? What will the planner's daily workflow look like? How will exceptions be managed? What governance will ensure the system stays calibrated? — are rarely part of the vendor evaluation. Neither are questions about strategic workforce planning — do we have the right planning roles? Do our planners have the analytical skills to use an advanced system effectively? What training and development is needed?

These aren't implementation details to be figured out later. They're fundamental to whether the investment will deliver returns, and they should inform the selection decision.

How to structure an APS RFx process that works

With those pitfalls in mind, here's a practical framework for running an APS selection process that actually surfaces the information needed to make a good decision.

Phase one: planning capability assessment (4-6 weeks)

Before engaging the market, invest in understanding your own planning capability — where it's strong, where it's weak, and where the biggest opportunities for improvement sit.

This means conducting a structured assessment of current planning processes: demand planning, supply planning, inventory management, S&OP, production scheduling. It means analysing forecast accuracy by product segment, customer and time horizon. It means quantifying the inventory opportunity — how much working capital is tied up that shouldn't be, and where service levels are being missed despite high stock levels. It means mapping the data landscape: what's available, what's reliable, what's missing, and what integration is needed.

The output of this phase is a planning capability baseline and a set of prioritised requirements — not a generic feature list, but a specific articulation of the planning decisions that need to improve, the data and analytical capabilities required to improve them, and the process and organisational changes that will be needed alongside the technology.

This is strategy and network design work as much as it is technology work. It ensures the RFx is anchored in business value rather than vendor marketing.

Phase two: market scan and shortlisting (2-3 weeks)

With clear requirements in hand, conduct a structured market scan to identify the vendors and implementation partners most likely to be a good fit. This should consider the vendor's strength in your specific industry (FMCG, manufacturing, retail, resources), the maturity of their demand planning and forecasting capabilities relative to your needs, their track record in the ANZ market — including local implementation partners and support, their architecture and integration approach relative to your ERP and data landscape, and their pricing model and how it scales with your business.

The goal is to shortlist three to four vendors for detailed evaluation — enough for genuine competitive tension, not so many that the evaluation becomes unmanageable. Gartner's Magic Quadrant and Critical Capabilities reports are useful inputs here, but they should be one data point among several, not the sole basis for shortlisting. As independent analysis has noted, the Magic Quadrant format tends to reward vendors with broad market presence and comprehensive feature claims without necessarily distinguishing between deep capability and surface-level functionality.

Phase three: RFx design and issue (3-4 weeks)

The RFx document itself should be designed to generate the specific information needed for a high-quality evaluation — not to demonstrate procurement rigour through volume of documentation.

The most effective APS RFx documents we've seen include several key components. First, a business context section that gives vendors genuine insight into the organisation's planning challenges, data environment and improvement priorities — because better-informed vendors produce better responses. Second, a focused set of functional requirements organised around planning capability areas (demand planning, supply planning, inventory optimisation, S&OP, analytics) rather than exhaustive feature checklists. Third, a set of planning scenarios that vendors will be asked to demonstrate during evaluation — specific, data-driven scenarios that reflect the organisation's real challenges, such as "forecast a product with high promotional variability," "optimise safety stocks across a three-tier distribution network," or "model the impact of a supply constraint on customer service." Fourth, clear requirements for implementation approach, team composition, methodology, timeline and change management. Fifth, a total cost of ownership template that requires vendors to itemise licensing, implementation, integration, data engineering, training, change management and ongoing support costs over a defined horizon.

The scenarios are critical. They're what transform the evaluation from a paper exercise into a practical test of capability. Providing a representative dataset to shortlisted vendors — anonymised if necessary — and requiring them to demonstrate their system against defined scenarios reveals more in a two-hour session than a 200-page written response ever will.

Phase four: evaluation (4-6 weeks)

The evaluation phase should combine three streams of assessment.

Written response evaluation against the RFx criteria, focused on approach, methodology, team capability and total cost rather than feature compliance. This provides the baseline comparative view.

Scripted demonstrations where each shortlisted vendor works through the defined planning scenarios using the provided dataset. These sessions should be attended by planners, not just IT and procurement, because planners are the ones who will recognise whether the system handles their real-world challenges effectively. Evaluation should focus on the quality of outputs (does the forecast look sensible? Does the inventory recommendation align with service targets?), the usability of the planner interface (can a planner understand and act on what the system produces?), and the transparency of the system's logic (can planners see why the system is recommending what it's recommending, and override it when they have better information?).

Reference checks and site visits with comparable organisations that have implemented the same platform. These should go beyond the vendor's curated reference list — ask for references in your industry, your geography, your scale, and at a similar stage of planning maturity. The questions that matter most are: how long did implementation actually take versus plan? What was the real total cost versus budget? How quickly did planners adopt the system? What worked well and what would they do differently?

The evaluation criteria should weight practical demonstration performance, implementation credibility and total cost of ownership at least as heavily as written response quality. A common weighting structure might allocate 30% to demonstrated planning capability, 25% to implementation approach and team, 20% to total cost of ownership, 15% to architecture, integration and scalability, and 10% to vendor viability and support model.

Phase five: proof of concept (4-8 weeks, optional but recommended)

For high-value APS investments, a proof of concept with the preferred vendor — or in some cases, two finalists — provides a final layer of validation before committing. The PoC should use a defined subset of the organisation's actual data and test a specific set of planning capabilities in realistic conditions.

This isn't a full implementation. It's a structured test that answers the question: does this system, with our data, produce materially better planning outputs than our current approach? If the answer is yes — demonstrably, measurably — the business case for investment is validated. If the answer is equivocal, that's critical information to have before signing a multi-year contract.

Phase six: commercial negotiation and contracting (3-4 weeks)

With a preferred vendor identified and validated through demonstration and potentially PoC, the commercial negotiation phase should address the full scope of the relationship — not just licensing terms.

Key negotiation areas include licensing structure and scalability (what happens as user numbers or data volumes grow?), implementation pricing and risk sharing (fixed fee for defined scope versus time-and-materials), performance commitments (are there contractual commitments to system performance, forecast accuracy improvement, or implementation timeline?), data ownership and portability (if you leave, can you take your configuration, models and data?), support model and SLAs (response times, availability commitments, upgrade frequency), and contract term and exit provisions.

This is procurement work that benefits from experience with technology contracts specifically — understanding the commercial levers that matter in SaaS agreements, the risks that need to be allocated, and the governance mechanisms that protect the organisation's interests over a multi-year relationship.

The Australian and New Zealand context

For organisations operating in Australia and New Zealand, several characteristics of the local environment make APS selection particularly consequential.

Long inbound supply chains with variable lead times — particularly for imported products coming through congested port infrastructure — mean that demand planning accuracy and inventory optimisation have a disproportionate impact on both working capital and service levels. The buffer you need across a 12-week ocean lane from Asia is fundamentally different from what a European business needs across a 3-day road lane from a neighbouring country.

Seasonal and promotional demand patterns in Australian markets — particularly in FMCG, retail and agriculture — create concentrated demand peaks that expose the limitations of basic forecasting approaches. An APS that handles promotional uplift well can be worth millions in reduced waste and improved on-shelf availability.

The relatively small scale of the ANZ market means that some global APS vendors have limited local presence — fewer implementation consultants, smaller support teams, longer response times. The RFx process should explicitly evaluate local capability, not just global credentials. An outstanding platform with no credible ANZ implementation partner is a risk, not an asset.

Working capital pressure in the current interest rate environment makes inventory optimisation — one of the core APS capabilities — directly material to the balance sheet. Every dollar of unnecessary inventory carries a real financing cost that didn't matter as much when rates were near zero.

These factors reinforce why the RFx process needs to be tailored to ANZ operating conditions rather than run as a generic global technology evaluation. The scenarios used in evaluation should reflect Australian supply chain realities: long lanes, seasonal peaks, promotional volatility, concentrated retail customers and distributed geographic networks.

For more on how APS capabilities map to ANZ supply chain challenges, Trace has published a detailed guide on Advanced Planning Systems and how they transform supply chain planning in Australia and New Zealand.

Process readiness matters as much as system selection

One of the most important — and most frequently overlooked — aspects of APS selection is the question of process readiness. The best demand planning system in the world won't deliver value if the organisation doesn't have a functioning S&OP process, if planners don't have time to work with the system because they're buried in manual data manipulation, or if there's no governance rhythm for reviewing and acting on system outputs.

The RFx process should include an honest assessment of organisational readiness: do we have the planning processes, the roles, the data discipline and the executive sponsorship to implement and sustain an APS effectively? If the answer is "not yet," that doesn't mean abandoning the investment — but it does mean sequencing process improvement and capability building alongside (or ahead of) the technology implementation, and reflecting that reality in the implementation plan and timeline.

The organisations that get the most from their APS investment typically spend as much effort on project and change management — redesigning planning processes, upskilling planners, building governance mechanisms, securing stakeholder buy-in — as they do on configuring and deploying the technology. The RFx process should evaluate vendors and implementation partners on their ability to support this change dimension, not just their technical capability.

How Trace Consultants can help

At Trace Consultants, we help Australian organisations navigate APS selection and implementation from a position of genuine planning expertise — not just technology procurement capability.

We work across the full selection lifecycle:

Planning capability assessment. We assess your current demand planning, supply planning, inventory management and S&OP processes against best practice, quantify the improvement opportunity, and define the specific requirements that should drive your APS selection. This is planning and operations consulting grounded in practical supply chain experience.

Market scan and shortlisting. We help you navigate the vendor landscape with an informed, independent view — identifying which platforms are genuinely strong for your industry, your scale and your planning maturity, and which are better suited to different contexts.

RFx design and management. We design and manage the RFx process end to end — from document development through evaluation, demonstration management, reference checking and recommendation. Our approach emphasises scenario-based evaluation over feature checklists, ensuring the selection is based on demonstrated capability rather than written claims.

Commercial negotiation. We support procurement negotiations with APS vendors and implementation partners, bringing benchmarks and experience with technology contract structures to ensure the commercial terms protect your interests and reflect realistic expectations.

Implementation readiness. We help you prepare for implementation — designing the target planning processes, defining the organisational structure and roles, building the data readiness plan, and establishing the change management approach that will determine whether planners adopt the system or revert to spreadsheets.

Ongoing optimisation. Post-implementation, we support organisations in tuning and optimising their APS — recalibrating forecasting models, refining inventory parameters, improving S&OP process integration, and building internal capability to sustain and extend the system's value over time.

Our independence from any APS vendor means our recommendation is based entirely on which platform best fits your requirements, your data, your organisation and your budget — not on partnership arrangements or reseller margins. We've worked with organisations implementing platforms across the spectrum, from the major Leaders through to focused niche players, and our advice reflects that breadth of experience.

Getting this decision right

An APS investment is typically a multi-year, multi-million dollar commitment. The system you select will shape how your organisation plans demand, manages inventory, allocates supply and makes trade-off decisions for years to come. The planners who use it daily will either be empowered to make better decisions faster — or frustrated by a tool that doesn't fit their reality.

The RFx process is where you create the conditions for one of those outcomes or the other. A process that's rigorous about the right things — planning capability, practical demonstration, implementation credibility, total cost, change readiness — will consistently lead to better decisions than one that's rigorous about the wrong things: feature checklists, compliance documentation and licensing price.

If your organisation is considering an APS investment and you want to make sure the selection process is designed to deliver a system your planners will actually use, we'd welcome the conversation.

‍