What is demand planning? Demand planning is the process that turns sales history, a statistical baseline, sales input, and promotional plans into one demand number a business commits to. A demand plan can follow every step and still run 15% high for four quarters in a row. The cause is often one input nobody checked. Teams respond by returning the model or lengthening the monthly consensus meeting, and the defect survives both. This guide covers the definition, the steps, and the five inputs behind the miss.
What Is Demand Planning?
The process builds one agreed demand number for each item and location across a planning horizon. It combines statistics on past sales with the knowledge of people closest to the market. The output goes to S&OP, where supply is set against it. In practice, the demand planning meaning is one number per item and location that a named person owns.
A demand planner reconciles several imperfect views of the future into one number the business can commit to. The statistical baseline carries what history can explain. Sales input carries what the field knows before history shows it. Marketing adds the promotional uplift it has already funded. Each view is partly right. The planner decides how much weight each one earns this cycle. The result is a consensus forecast that production and procurement can act on. Across a monthly cycle, the planner’s week ends in the consensus meeting, with the baseline, the sales changes and the promotion plan on one screen.
The Demand Planning Process, Step by Step
The demand planning process runs on a monthly planning cycle, in step with S&OP. Refresh the baseline between cycles only if demand moves faster than that and the business can act on the change, as with weekly replenishment. These demand planning steps repeat every cycle, and each one inherits whatever the step before it left behind.
- Clean the demand history. Remove supply effects from invoiced or dispatched quantity before any model reads it.
- Build the statistical baseline. Fit a model to the cleaned history at the level of the decision, such as item and depot for replenishment.
- Collect sales input. Add tenders and new listings that history cannot contain yet.
- Layer the promotional plan. Enter each scheme with its dates and the expected uplift by item.
- Run consensus. Hold one meeting on one version of the number, with evidence attached to every change.
- Publish and hand off to S&OP. The agreed number enters supply review, and the demand-supply balance is set there against capacity and inventory.
Steps one and six get the least attention. Step one decides what the model is allowed to learn. The hand-off decides whether supply treats the published number as a commitment. Fix the calendar too. If sales input arrives after the baseline is locked, the plan that reaches supply review describes a different month from the one the model fitted. Noise needs a different fix. An item swinging week to week around a steady level needs a buffer sized on its forecast error, and a faster refresh would only make the plan chase the swings.
Which Five Inputs Do Demand Planning Teams Get Wrong?
Five inputs feed the published number in a make-to-stock business, where finished goods sit at a plant, depot or distributor and ship from stock. Demand history, the statistical baseline, sales input, promotional input and the consensus step each fail recognisably. Each failure leaves a symptom a planner can look for this week.
| # | Input | Why it goes wrong | Symptom |
|---|---|---|---|
| 1 | Demand history | Invoice or dispatch history not cleansed for supply effects | The forecast repeats last year’s shortages and spikes |
| 2 | Statistical baseline | Promotion spikes and one-off orders left in, or the wrong history window | A confident number built on the wrong pattern |
| 3 | Sales input | Upward adjustments carry optimism bias | Chronic over-forecast and excess stock |
| 4 | Promotional input | Added without checking what the baseline already contains | Uplift double-counted or missed |
| 5 | Consensus step | Run as a negotiation between departments | The final number reflects influence over data |
Demand history not cleansed for supply effects
The baseline in a make-to-stock business is built on invoiced or dispatched quantity. Invoiced quantity is a fair proxy for demand while the stocking point has stock. The trouble is narrower. An invoice records what was available to sell, and it says nothing about what the market wanted. Wherever the two differ, supply effects enter the forecast.
| Situation | What the invoice shows | Effect on the forecast |
|---|---|---|
| Stock-out at the depot | Invoice says 600, but the depot could have sold 1,000 | The lost 400 never enters history |
| Shortage allocation | Who received rationed stock, with no record of who needed it | Depots that were cut back look like low-demand locations |
| Substitution | Sales that shifted to another pack size or a nearby depot | One history is inflated and the other deflated |
| Month-end loading | Primary sales pushed to distributors against billing targets | The model learns the month-end push as consumption |
Correct invoice history and keep using it. Flag stock-out periods by item and location so the model skips them, and estimate lost sales from the sales rate on in-stock days. Record ordered against supplied quantity where depots or distributors place replenishment orders. Cross-check primary sales against distributor offtake where that data exists. Distributors over-order during shortages, so their order data needs cleansing too. Daniel Feiler, Jordan Tong and Richard Larrick reported this censorship bias in Management Science in 2013. People lean too heavily on the sales they can see, so demand beliefs run low when lost sales stay invisible.
Statistical baseline
Promotion weeks and one-off tender volumes sit in history as ordinary demand. The model fits a pattern that includes them and projects it forward. History length matters as well. A festive-season item with two years of monthly data gives a seasonal model only 24 points. An item whose distribution doubled 18 months ago has history that describes a smaller business. Flag the structural break and fit only the period after it.
Sales input
Fildes and colleagues studied more than 60,000 forecasts from four supply-chain companies for the International Journal of Forecasting in 2009. Upward adjustments were much less likely to improve accuracy than downward ones, and were more often made in the wrong direction, which the authors read as optimism bias. Larger adjustments more often improved accuracy. Small ones often damaged it and used planner time for nothing. So question the small upward nudge first, and keep the large downward call from someone who knows the account.
Promotional input
Marketing plans an uplift for a festive build. The baseline was fitted on history that already holds last year’s festive build, so the uplift gets counted twice. A scheme agreed after the baseline locks can be missed entirely. The check takes one query, comparing last year’s baseline volume in the same weeks with the uplift marketing has entered for each promoted item.
Consensus step
A consensus meeting run as a negotiation records who argued hardest. Treat it as a test against evidence. Record every change as a signed difference from the baseline with a stated reason. Limit discussion to items where sales input and baseline differ by more than a threshold set in advance, for example 15% or one week of cover. Everything inside the threshold publishes unchanged.
Why Do Demand Planning Errors Stack Up?
The errors stack because several inputs push the same number in the same direction. Take a festive-season item. The item’s history still carries last year’s month-end loading, and the baseline was fitted on promotion weeks. Sales then adds an optimistic lift, marketing adds uplift the baseline already holds, and the consensus meeting settles on the highest figure anyone defended.
Every step looks reasonable on its own. Say the baseline runs 5% high, sales input adds 8% and the double-counted scheme adds 6%. The published plan then sits about 20% above what the market will buy, and nobody made a large change. Independent errors would partly cancel. Shared bias compounds. The same stacking runs downward after a stock-out, when every input inherits the low history.
Check the direction of each input’s error before its size.
How Do You Find the Input That Hurts? Use Forecast Value Added
Forecast value added shows which step in the process improves accuracy and which step makes it worse. The Institute of Business Forecasting describes it as a metric for judging each step and each participant in the forecasting process. It answers the question of where to look first.
Run it over your last 12 planning cycles. Measure accuracy at four checkpoints for the same items and horizon:
- A naive forecast that carries last period’s actual forward.
- The statistical baseline forecast.
- The number after sales input.
- The number after consensus.
Use one error measure throughout and keep the lag fixed, so a one-month-ahead figure is never compared with a three-month-ahead one. A step that beats the one before it has earned its place. Repair first any step that loses accuracy against the checkpoint before it. Say the naive forecast misses by 30%, the baseline by 22%, the number after sales input by 25%, and the consensus number by 24%. The sales step added 3 points of error, so repair starts at the judgmental adjustment.
Where Does Oritiq Fit in Demand Planning?
Oritiq is the decision layer above the ERP. Before any model reads the history, it marks stock-out periods for each item and location and estimates the sales those gaps cost. The record is corrected first, so the baseline learns from demand with the supply gaps taken out.
Its demand forecasting engine picks a statistical model for each SKU from that item’s own demand pattern. An override always carries a reason category and the name of the person who made it. The platform measures forecast value added at each stage, from the model through planner and sales changes to the executive override, and flags planners whose changes keep costing accuracy. S&OP reconciliation then carries the agreed number into supply review. Data refreshes from ERP transactions on a set schedule.
FAQs on Demand Planning
1. What is demand planning?
Put simply, it is how a business settles on the demand number it will build and buy against. A consumer goods maker might agree on 2,000 item-depot numbers each month in one consensus meeting. Planners then hand those numbers to S&OP, where capacity and stock are set against them.
2. What are the steps in the demand planning process?
Six steps repeat each month. The first four build the number, from cleaning the history and fitting the baseline to adding sales input and the promotional plan. Consensus and the hand-off to S&OP close the cycle. Extra refreshes in between pay off only when demand shifts within the month and replenishment can respond, for instance on a 7-day cycle.
3. What is the difference between demand planning and demand forecasting?
Forecasting produces a statistical number from history. A demand plan turns it into a number the business will act on, by adding market knowledge, promotional commitments, and a consensus step. A forecast of 2,400 units can become a plan of 2,650 once a confirmed tender is added.
4. Do manual sales adjustments help or hurt forecast accuracy?
Both, depending on the type. In a 2009 study of more than 60,000 forecasts, upward adjustments improved accuracy far less often than downward ones. Large adjustments more often helped. Small ones often made the forecast worse. Question the small upward nudge first and keep the large, informed downward call.
5. What is a consensus forecast?
A consensus forecast is the single demand number the commercial and planning teams agree on for a cycle. Its quality shows in the record. Every change from the baseline should carry a stated reason, so the plan can be checked 6 weeks later against what actually sold.
6. How often should a demand plan be updated?
Once a month for the full cycle, alongside S&OP. Weekly replenishment justifies a quicker baseline refresh, but only for demand that really shifts within the month. A noisy item needs a buffer sized on its forecast error. The plan must also look far enough ahead to cover the longest lead time, such as a 10-week casting.
7. What makes a demand planning process fail even when the steps are followed correctly?
A defective input. Invoice history with stock-outs left in, or a baseline fitted on promotion weeks, produces a wrong number through a correct process. So does uplift counted twice. Forecast value added across the last 12 cycles shows which input is responsible.
What Should You Count in Your Demand History First?
Pull 12 months of demand history for your top 50 items. Count the periods each one spent at zero stock at its stocking point. That count tells you whether your baseline has been reading demand or reading supply, and it takes an afternoon.
Talk to the Oritiq team about a scoped pilot that starts with your stock-out history.
