The Demand Planning Process: 7 Steps for Manufacturers

The demand planning process for a manufacturer runs in seven steps: clean and segment the history, generate a statistical baseline, run the demand review with sales and marketing, build one consensus number, reconcile it against supply and finance, commit it in executive S&OP, then measure the miss and feed it back. Run on a monthly cycle, those steps turn scattered guesses into one signed demand number that supply, inventory, and finance build off. Skip any of them and you get three competing forecasts, a quarter-end inventory surprise, and a planner who can't tell you why the number missed.

This is the exact sequence I ran every month at a $250M manufacturer with 4,200 SKUs across 11 locations. Not a textbook diagram. The working version, with the owners, the inputs, and the places it actually breaks.

Each step below has one owner and one deadline. Demand planning fails when it's everyone's job, which means it's no one's. Assign names. For the wider context of where this process sits, see what demand planning is.

Step 1: Clean and segment the data

Before any math, fix the inputs. Pull 24-36 months of shipment or order history and scrub it. Most teams skip this and forecast on dirty history. Garbage in, expensive out.

What to scrub

• De-stockout the history. Past stockouts mean you sold what you had, not what customers wanted. Flag those periods and correct them, or your forecast inherits your old supply failures.
• Strip one-time events. A 2024 emergency buy from a customer who's since left isn't a pattern. It's noise dressed as signal.
• Net out returns and intercompany transfers. Both inflate apparent demand and quietly bias every model downstream.

Segment before you forecast

Run an ABC/XYZ cut: ABC by revenue contribution, XYZ by demand variability. Your A/X items (high value, stable) earn tight statistical forecasting. Your C/Z items (low value, lumpy) get simple rules, not babysitting.

This segmentation decides how much attention each SKU deserves and which model fits it. The full method is in our ABC-XYZ inventory analysis guide. Owner: demand planner. Deadline: day 2.

Step 2: Generate the statistical baseline

Now run the math. Match the model to the demand profile. There is no single best method, and the academic literature is blunt about it.

Demand profile	Model that fits	Why
Stable, seasonal (A/X)	Holt-Winters exponential smoothing, ARIMA	Captures level, trend, and seasonality with few parameters
Trending	Holt's linear / damped trend	Damping prevents runaway extrapolation
Intermittent, lumpy (C/Z)	Croston's method, SBA	Built for demand that is zero most periods
Rich external signal	ML forecasting (gradient boosting)	Absorbs price, weather, macro, and promo features

Exponential smoothing weights recent observations more heavily and decays older ones — a simple idea that still anchors most production forecasts, as Hyndman and Athanasopoulos lay out in their open textbook Forecasting: Principles and Practice (2018). For SKUs that sell in sporadic bursts, the standard is still Croston's method (1972), refined later by the Syntetos-Boylan approximation.

Let the system pick best-fit per SKU against a holdout period. Don't hand-tune 4,200 SKUs. Hand-tune the 200 that drive 80% of revenue. Owner: planning analyst. Deadline: week 1.

Step 3: Run the demand review with sales and marketing

This is where the forecast becomes a plan. Get sales, marketing, and the planner in a room — or a shared model — and layer in what the math can't see.

What the statistics miss

• Promotions and price changes already on the calendar
• Known account wins and losses, with confirmed POs where they exist
• New product launches and the SKUs they'll cannibalize
• Channel and customer mix shifts

Correct for optimism

The planner's real job here is to bias-correct sales optimism. Track each rep's historical bias and discount accordingly. If a region has run 18% high for six straight quarters, their input gets a haircut before it touches the plan.

Gartner's research is direct that supply chain leaders routinely overlook this, and that chronic over- or under-forecasting is one of the most fixable problems in planning (Gartner, demand forecast bias). For the mechanics of catching it, see how to measure and fix forecast bias. Owner: demand planner. Deadline: week 2.

Step 4: Build the consensus number

Reconcile the statistical baseline and the sales intelligence into one number. Document every override and the reason in plain language: "+2,000 units, Q3 Costco promo, confirmed PO." That override log is gold later. It tells you whether your human adjustments help or hurt.

The rule is simple. No orphan numbers. There is one demand plan — by SKU, by location, by time bucket. Not a sales number and a planning number living in separate spreadsheets. One.

This is the heart of consensus demand planning: a single forecast everyone owns, not the highest bidder's wish list. Owner: demand planner, arbitrated by the S&OP lead. Deadline: week 2.

Step 5: Reconcile against supply and finance

A demand plan you can't build or fund isn't a plan. It's a wish. Two reviews close that gap.

Supply review

Can the plant and the suppliers actually deliver this mix at this timing? Where they can't, the constraint goes back to demand for re-prioritization. You decide what to serve before the customer decides for you.

Finance review

Does the plan roll up to the revenue and margin the P&L needs? If there's a gap to target, name it now. Don't let it surface in the board meeting three weeks later.

This is the S&OP handshake — the demand plan, supply plan, and financial plan are the same plan in three views. ASCM defines sales and operations planning as exactly this: one integrated set of plans across sales, operations, and finance. Owner: S&OP lead. Deadline: week 3.

Step 6: Commit in executive S&OP

Leadership signs the number. This is a decision meeting, not a status update. The agenda is exceptions and trade-offs only.

What lands on the agenda

• SKUs where demand and supply can't reconcile: what do we serve, what do we let slip?
• The gap to the financial plan and the specific moves that close it
• Risk items and the scenarios bracketing them

When the executive team commits, the number locks for the period. Everyone downstream executes off it. No re-litigating in week 2.

Skipping the executive commit is a fast track to the bullwhip effect — the demand distortion Lee, Padmanabhan, and Whang documented back in 1997, where small swings at the top whip into wild inventory swings down the chain. A single committed signal is the cheapest dampener you have. Owner: GM / VP Supply Chain. Deadline: week 4.

Step 7: Measure, then close the loop

The step everyone skips. Compare the committed plan to actuals and dissect the misses. This is where compounding teams separate from the rest.

The four metrics that matter

• Forecast accuracy / WMAPE, weighted by revenue so your big SKUs dominate the score. The MAPE vs WMAPE trade-off decides which version you report.
• Bias — chronic over or under is your single most fixable error.
• Plan attainment — did the committed decision hold, or did execution drift?
• Forecast value added — did human overrides beat the naive statistical baseline?

That last one humbles people. SAS analyzed forecasts across companies and found a majority were worse than a naive no-change forecast — meaning the process actively destroyed accuracy (SAS, Forecast Value Added analysis). If your overrides make accuracy worse, stop making them. Our forecast value added how-to walks the calculation.

Feed every lesson back into Step 1. Demand planning is a loop, not a line. Owner: demand planner. Deadline: week 1 of next cycle.

The cadence at a glance

Week	Steps	Output
Week 1	Data clean, statistical baseline, prior-month measure	Clean forecast + last cycle's accuracy
Week 2	Demand review, consensus number	One reconciled demand plan
Week 3	Supply + finance reconciliation	Buildable, fundable plan
Week 4	Executive S&OP commit	Signed, locked number

The whole thing runs in four weeks and resets. This monthly demand cadence is the front half of the broader S&OP monthly cycle that links it to supply and finance.

Where the process leaks — and where AI helps

The discipline beats the sophistication every time. A simple process run religiously crushes a fancy model run sporadically. Most accuracy problems I've torn down trace to a missing owner or a skipped Step 7, not a weak algorithm.

That said, the math has moved. McKinsey's work on AI-driven supply chain planning reports forecast-error reductions of 20-50% and inventory cuts up to 20% when machine learning replaces spreadsheet forecasting (McKinsey, 2021). The gain is real, but it lands only on top of a clean process. Bolt an ML model onto dirty history and an absent commit step, and you've automated the same chaos faster.

So run the seven steps first. Earn the right to the algorithm by getting the loop honest.

Frequently asked questions

What are the 7 steps of the demand planning process?

The seven steps are: clean and segment the data, generate a statistical baseline, run the demand review with sales and marketing, build a single consensus number, reconcile against supply and finance, commit the number in executive S&OP, and measure actuals to close the loop. Each step has one named owner and a deadline inside a four-week monthly cycle. The output is one signed demand number the whole organization plans against.

How long does a demand planning cycle take?

A standard monthly cycle runs about four weeks: data and baseline in week 1, demand review and consensus in week 2, supply and finance reconciliation in week 3, and the executive commit in week 4. The cycle then resets and repeats every month. The cadence matters more than the duration — the discipline of running it the same way every month is what compounds accuracy.

What is the difference between demand planning and demand forecasting?

Demand forecasting is the statistical step that projects future demand from history and signals. Demand planning is the full cross-functional process that turns that forecast into one committed, reconciled number everyone executes against. Forecasting is one input; planning is the decision, the consensus, and the commit around it.

Who owns the demand planning process?

The demand planner owns the day-to-day steps — data cleaning, the baseline, the demand review, and the consensus number. The S&OP lead arbitrates conflicts and runs the supply and finance reconciliation. Final commitment belongs to leadership, typically the GM or VP of Supply Chain, in the executive S&OP meeting.

How do you measure if your demand plan is working?

Track four metrics every cycle: weighted forecast accuracy (WMAPE), forecast bias, plan attainment, and forecast value added. Forecast value added is the most revealing — it tests whether your human overrides actually beat a naive baseline, since research shows many forecasts perform worse than doing nothing. Feed each finding back into the next cycle's data-cleaning step.