Demand Planning: How to Build a Function That Actually Drives Decisions
Demand planning is the process of forecasting what customers will buy, in what quantities, in which locations, and when, so that the rest of the supply chain can prepare accordingly. It sounds straightforward. In practice, it is one of the most consistently underperforming functions in Australian supply chains.
The symptoms are familiar. The forecast exists, but nobody trusts it. Sales overrides the numbers every month based on gut feel. Production plans against a different set of assumptions than procurement. The warehouse gets surprised by demand spikes that the commercial team saw coming but never communicated. Inventory is simultaneously too high in aggregate and too low in the specific SKUs that customers actually want. And the S&OP meeting, which is supposed to reconcile all of this, becomes a reporting exercise where people present last month's actuals rather than making forward-looking decisions.
The root cause is almost always the same: the organisation has a forecasting process but not a demand planning function. There is a difference. A forecasting process generates a number. A demand planning function generates a number that is trusted, challenged, enriched with commercial intelligence, and used as the single basis for procurement, production, inventory, logistics, and financial planning.
This article covers how to build a demand planning function that works: the right process, the right data, the right technology, the right organisational design, and the right connection to the rest of the supply chain.
Why Demand Planning Fails
Before building something better, it is worth understanding why the current state is so consistently poor.
The forecast is built in isolation. In most organisations, the statistical forecast is generated by a planner or analyst using historical shipment data and a forecasting tool. That forecast is then circulated for review, at which point sales, marketing, and finance each make adjustments based on their own information, their own biases, and their own incentives. The result is a forecast that has been adjusted multiple times by multiple people with different objectives, and that nobody fully owns or trusts. Sales wants to be conservative to overdeliver on targets. Marketing wants to be aggressive to justify promotional investment. Finance wants a number that reconciles to the budget. The demand planner is left trying to reconcile these competing inputs into a single number.
The wrong data is being used. Most demand planning processes are built on shipment or dispatch data, which measures what the organisation shipped, not what customers actually demanded. Shipment data includes the effect of supply constraints, promotional loading, and order phasing. If a product was out of stock for two weeks, the shipment data shows zero demand for that period, when in reality demand existed but could not be fulfilled. True demand data, clean of supply-side distortion, is what the forecast should be built on. For FMCG and retail businesses, point-of-sale data from the retailer is the closest proxy to true demand. For B2B businesses, order data (including lost orders and backorders) is the relevant input.
Promotional demand is not planned separately. In FMCG and retail, promotional activity can represent 30 to 50 percent of volume in some categories. Promotional demand behaves completely differently from baseline demand: it is lumpy, time-bound, and heavily influenced by the type of promotion, the retailer, the price point, and the in-store execution. Feeding promotional volume into the same statistical model that forecasts baseline demand produces a forecast that is wrong on both components. Best practice separates baseline demand (the steady-state sales pattern driven by distribution, ranging, and habitual purchasing) from uplift demand (the incremental volume driven by promotions, new product launches, and one-off events) and plans each using different methods.
Forecast accuracy is measured but not managed. Most organisations measure forecast accuracy at an aggregate level, typically monthly at a product family or brand level. At that level, the numbers often look acceptable: 70 to 85 percent accuracy is common. But the decisions that matter, how much of which SKU to produce, where to position inventory, what to order from suppliers, are made at a much more granular level: weekly, by SKU, by location. At that level, forecast accuracy drops dramatically. Measuring accuracy only at the aggregate level gives a false sense of confidence. Measuring it at the level where decisions are made reveals the true performance gap.
There is no accountability for the forecast. Who owns the demand forecast? In most organisations, the answer is unclear. The demand planner generates it. Sales adjusts it. Marketing provides promotional inputs. Finance signs off on it. But nobody is accountable for the quality of the final number or for the consequences of getting it wrong. Without clear accountability, there is no incentive to invest in improving the process.
What Good Demand Planning Looks Like
Organisations that do demand planning well share several characteristics, regardless of sector.
A structured, repeatable monthly process. The demand planning cycle runs on a fixed calendar with defined steps, clear inputs, and specific outputs. A typical monthly cycle involves four steps. First, generate the statistical baseline forecast using clean demand data, refreshed with the most recent actuals. Second, enrich the forecast with commercial intelligence: promotional plans, new product launches, distribution changes, pricing changes, customer wins and losses, and known market events. Third, review and challenge the forecast in a demand review meeting that brings together demand planning, sales, marketing, and finance. Fourth, publish the consensus demand plan as the single version of the truth that drives all downstream supply chain decisions.
Baseline and uplift are planned separately. The statistical forecast drives the baseline. Promotional uplift, new product volumes, and event-driven demand are layered on top using different methods: promotional models, analogue analysis (comparing a new promotion to a similar historical promotion), or sales team input for customer-specific activity. This separation allows each component to be measured, managed, and improved independently.
Forecast accuracy is measured at the right level. The metrics that matter are forecast accuracy at the SKU-location-week level (or the most granular level at which decisions are made), forecast bias (whether the forecast is systematically over or under), and forecast value added (whether each step in the process, from statistical forecast to consensus plan, improves or degrades accuracy). Forecast value added is particularly important because it tells you whether the human adjustments being made to the statistical forecast are actually adding value. In many organisations, the adjustments make the forecast worse, not better.
The demand plan drives supply decisions. The consensus demand plan is the single input to production planning, procurement, inventory policy, and logistics planning. If the supply chain is planning against a different number than the demand plan, the demand planning function has failed in its most important objective. The connection between the demand plan and the supply plan, typically formalised through the S&OP or IBP process, is what turns a forecast into a decision-making tool.
Technology supports the process, not the other way around. Demand planning technology ranges from Excel (still the most common tool in Australian mid-market businesses) through specialist demand planning software (tools like SAP IBP, o9, Kinaxis, Blue Yonder, RELEX) to AI-powered forecasting platforms. The technology should be selected based on the maturity of the process and the complexity of the demand signal. An organisation that does not have a structured demand review process will not benefit from a $500,000 AI forecasting platform. The process and people need to be in place before the technology investment makes sense.
The Australian Context
Several characteristics of the Australian market make demand planning both more important and more challenging.
Concentrated retail. In grocery FMCG, Woolworths and Coles together account for approximately 65 percent of the market. This concentration means that a ranging decision, a promotional programme, or an inventory policy change at one major retailer has a disproportionate impact on supplier demand. FMCG manufacturers that do not have strong demand planning processes are perpetually surprised by swings that could have been anticipated through better collaboration with their retail customers.
Long inbound lead times. Australia's distance from most manufacturing sources means inbound lead times for imported goods are typically four to twelve weeks, depending on origin. For products sourced from Europe or North America, lead times can extend further. Long lead times amplify the cost of forecast error: if you get the forecast wrong and your replenishment cycle is eight weeks, you have eight weeks of misaligned inventory before you can correct it. This makes forecast accuracy structurally more important for Australian businesses than for businesses in markets closer to their supply base.
Seasonal and promotional volatility. Australian retail is heavily promotional. End of financial year sales, Black Friday, Boxing Day, and seasonal events like Christmas and Easter create demand peaks that require specific planning. For categories like beverages, sunscreen, and outdoor products, seasonal demand variation can be two to three times the baseline. Planning for these peaks requires dedicated promotional and seasonal demand processes, not just a statistical model that smooths the peaks into the average.
Thin domestic manufacturing base. For many product categories, Australian businesses are importers, not manufacturers. This means the demand plan is the primary input to purchase order placement and shipping decisions, not production scheduling. The consequence of poor demand planning is not just an inefficient factory run; it is a container sitting on the wrong side of the world or an airfreight bill to cover a stockout.
Building the Capability
For organisations that want to move from a basic forecasting process to a genuine demand planning function, here is a practical sequence.
Start with the data. Before investing in process, technology, or people, get the data right. Identify your cleanest source of demand data: point-of-sale data if available, order data if not. Clean it: remove anomalies, account for stockouts, separate promotional volume from baseline. Establish a data foundation that the statistical forecast can be built on with confidence. This step alone can take four to eight weeks for a complex business, and it is the step that determines the quality of everything that follows.
Define the process. Document the monthly demand planning cycle: who does what, when, with what inputs, producing what outputs. Assign clear ownership for each step. Define the demand review meeting: who attends, what they bring, what decisions are made. Keep it simple. A well-run monthly cycle with four clear steps is better than an elaborate weekly process that nobody follows.
Separate baseline and uplift. Build the statistical baseline forecast. Then establish a separate process for capturing and planning promotional and event-driven demand. In FMCG, this typically requires a promotional planning calendar linked to retailer activity, with uplift estimates based on historical promotion performance. In B2B businesses, it might involve capturing known project pipelines, contract renewals, or one-off orders separately from the recurring demand pattern.
Implement forecast accuracy measurement. Start measuring accuracy at the level where decisions are made. Publish the numbers monthly. Track bias. Introduce forecast value added measurement to assess whether each process step is improving accuracy. Make the numbers visible to the demand review meeting participants. What gets measured gets managed, and most organisations are surprised by how poor their granular forecast accuracy is when they first start measuring it properly.
Connect to supply. Ensure the consensus demand plan is the single input to procurement, production, and inventory planning. If the supply chain is using a different number, identify why and fix it. This connection is what makes demand planning operationally valuable rather than just an analytical exercise.
Then consider technology. Once the process is stable, the data is clean, and the team knows what they need from a system, evaluate technology options. For mid-market businesses, a well-structured Excel model or a lightweight planning tool may be sufficient. For larger businesses with complex demand patterns, a specialist demand planning platform will add value through better statistical modelling, automated baseline generation, and integrated promotional planning.
The Role of AI in Demand Planning
AI and machine learning are increasingly being applied to demand planning, and for good reason. Statistical forecasting models have inherent limitations: they are built on historical patterns and struggle with demand signals that are new or structurally different from the past. Machine learning models can incorporate a much wider range of demand signals, including weather, economic indicators, social media trends, competitor activity, and external events, and can identify patterns that traditional time-series models miss.
In practice, AI adds the most value in three specific areas. First, in generating the statistical baseline: ML models can improve baseline forecast accuracy by 10 to 30 percent over traditional exponential smoothing or ARIMA models, particularly for SKUs with erratic or intermittent demand. Second, in promotional uplift estimation: ML models trained on historical promotional data can predict uplift with greater accuracy than manual analogue-based methods. Third, in anomaly detection: identifying demand signals that are inconsistent with expected patterns and flagging them for planner review.
The trap is assuming that AI replaces the demand planning process. It does not. AI improves the statistical component of the forecast. It does not replace the commercial intelligence that comes from sales knowing a major customer is about to launch a new programme, or marketing knowing a competitor is withdrawing from a category. The best demand planning functions use AI to generate a better starting point, then layer human intelligence on top through a structured demand review process.
How Trace Consultants Can Help
Trace Consultants helps Australian organisations design, build, and improve their demand planning capability, from process design through to technology selection and implementation support.
Demand planning process design. We design end-to-end demand planning processes that connect statistical forecasting, commercial intelligence, demand review governance, and S&OP integration, tailored to your sector, complexity, and maturity level.
Forecast accuracy diagnostic. We assess your current forecasting performance at the level where decisions are made, identify the root causes of forecast error, and quantify the cost of those errors in terms of excess inventory, stockouts, and expediting.
Technology evaluation and selection. We help organisations evaluate and select demand planning technology, from lightweight tools for mid-market businesses to enterprise platforms for complex supply chains, ensuring the technology fits the process and the process fits the organisation.
S&OP and IBP integration. Demand planning does not exist in isolation. We design the connection between the demand plan and the supply response, ensuring the consensus forecast drives procurement, production, inventory, and logistics decisions through a structured S&OP or IBP process.
Explore our Planning & Operations services →Explore our Technology advisory services →Explore our FMCG & Manufacturing sector expertise →Speak to an expert at Trace →
Where to Start
If your organisation's demand planning consists of a spreadsheet that gets updated monthly and overridden weekly, start with the data and the process, not the technology. Clean your demand history, separate baseline from uplift, establish a structured demand review meeting, and start measuring forecast accuracy at the granular level. Those four steps cost very little and will tell you exactly where the improvement opportunity sits.
The organisations that get demand planning right reduce inventory by 10 to 25 percent while improving service levels. In an environment of high interest rates, long lead times, and volatile demand, that is not just an operational improvement. It is a material financial outcome.
Read more insights from Trace Consultants →Contact our team →