15 Demand Planning KPIs and Metrics That Matter

The 15 demand planning KPIs that actually move turns and service fall into five buckets: accuracy (WMAPE, MAPE, MAE, forecast accuracy), bias (forecast bias, tracking signal), value-add (FVA, override hit rate), inventory and working capital (turns, days of supply, inventory accuracy, E&O), and downstream service (fill rate, OTIF, plan attainment). Lead with WMAPE for accuracy, forecast bias for cash leakage, and FVA to prove the human is helping. Measure them monthly, weighted by revenue, and pair every accuracy number with a dollar number.

Most demand planning dashboards I've inherited measure the wrong things, measure them the wrong way, or measure 40 things so nobody acts on any of them. The point isn't a pretty scorecard. It's to tell you where the plan is breaking and what that's costing in cash and service.

At a $250M manufacturer I cut our metrics from 30-plus down to the 15 below. Measuring these well, every month, weighted by revenue, did more for our turns and fill rate than any software did. Here are the ones that earn their place, grouped by what they actually tell you.

Accuracy and error metrics

These tell you how close the number was to reality. Accuracy is the entry fee, not the prize, so don't stop here.

1. Forecast accuracy. 1 − (forecast error / actual), as a percentage. The headline number. Useful only if everyone agrees on the formula and the time bucket, which they rarely do.

2. MAPE (Mean Absolute Percentage Error). avg( |actual − forecast| / actual ). The classic, but it blows up on low-volume SKUs: a 2-unit error on a 1-unit forecast reads as 100%. It also ignores SKU value, so don't lead with it at the portfolio level.

3. WMAPE (Weighted MAPE). sum|actual − forecast| / sum(actual), weighted by volume or revenue. This is the one to lead with. A 50% miss on a $4 SKU should not register the same as a 5% miss on your flagship.

In a benchmark study of demand planners, WMAPE was the single most common error metric, used by 52% of respondents precisely because it reflects business impact. For a working build, see MAPE vs WMAPE. Target: under 30% on A-items for most discrete manufacturers, under 20% if you're good.

4. MAE (Mean Absolute Error). Average absolute miss in units. Pairs with WMAPE when you need raw magnitude, not a percentage, for capacity and supply math.

Bias: the metric that quietly drains cash

5. Forecast bias. sum(forecast − actual) / sum(actual). Positive means you chronically over-forecast, building inventory you don't sell. Negative means under, which buys you stockouts and lost sales.

Bias is more dangerous than error because it's systematic. Random error averages out over time; bias compounds in the same direction every cycle and shows up on the balance sheet as dead stock. If you fix one metric first, fix this one. Target: within ±5%. The mechanics live in Forecast Bias: How to Measure and Eliminate It.

6. Tracking signal. Running sum of bias divided by MAE. When it drifts past ±4, your forecast is structurally off and needs intervention, not a tweak. I use it as the alarm that pulls a SKU into the demand review.

Value-add metrics: is the human helping?

This is the bucket almost nobody runs, and it's the one that exposes the most waste.

7. Forecast Value Add (FVA). Compares your final forecast accuracy to a naive baseline, usually last period equals next period or a seasonal naive. If your analysts and sales overrides aren't beating the dumb forecast, they're destroying value, and you're paying salaries to make the number worse.

This isn't a fringe worry. Lokad's writeup of FVA frames the naive forecast as the baseline every process step must beat. Steve Morlidge's research, summarized on the SAS Business Forecasting Deal blog, found that roughly half of real-world forecasts were worse than a naive random walk. Run FVA on every override. Our full method is in Forecast Value Added (FVA): A Practical How-To Guide.

8. Override hit rate. Of the manual adjustments made in demand review, what share improved accuracy versus hurt it? This tells you whose judgment to trust and whose to quietly discount. Track it by person and by reason code.

Inventory and working-capital metrics

The plan exists to deploy cash and service customers. Measure both, or accuracy becomes a vanity project.

9. Inventory turns. COGS / average inventory. The cleanest read on whether the plan converts to working-capital efficiency. Most mid-market manufacturers run 4 to 8; pulling one extra turn on a $40M inventory frees real cash.

The spread is wide. APQC benchmark data shows top performers turning raw materials about 16.5 times a year against roughly 6 for bottom performers, a gap of more than 30 extra days of held inventory. Benchmarks and the full formula are in Inventory Turnover Ratio: Formula and Benchmarks.

10. Days of Supply / Days Inventory Outstanding. Inventory on hand expressed in days of forward demand. The operational twin of turns. APQC defines inventory days of supply as average inventory at standard cost divided by (annual COGS / 365). Watch it by SKU segment; your C/Z tail is where the dead stock hides.

11. Inventory accuracy. If on-hand records are wrong, every downstream KPI is fiction. Cycle-count accuracy under 95% means your other metrics are noise, and 95% is the widely cited floor for A-items in a disciplined cycle-count program. Below that, fix the data before you trust the dashboard.

12. Excess and obsolete (E&O) inventory. Dollars sitting above the demand that will ever consume them. This is stranded cash. Track it as a percentage of total inventory and trend it: rising E&O is chronic over-bias surfacing on the balance sheet. When it climbs, work How to Reduce Excess and Obsolete Inventory Fast.

Service and downstream-impact metrics

Forecasting is a means. Service is the end your customers actually feel.

13. Fill rate. Share of demand met from stock on the first pass. The direct customer-facing payoff of good planning. Watch the difference between unit fill and line fill; they tell different stories.

14. OTIF (On-Time In-Full). Did the customer get what they ordered, complete, on the promised date? APQC tracks this as the percentage of orders delivered complete and on time. It's the metric your biggest accounts score you on, and a great forecast that doesn't move OTIF isn't reaching the customer.

15. Plan attainment. Did actuals land within tolerance of the committed plan, not the statistical forecast? This separates a forecasting problem from a decision problem. If the forecast was good but attainment is bad, your S&OP meeting is breaking the number.

The dashboard, prioritized

Seven of the 15 carry most of the signal. Lead with these and pull the rest in by exception.

Metric	Formula	Target (mid-market)	What it tells you
WMAPE	sum|err| / sum(actual)	<30% A-items	Accuracy, value-weighted
Bias	sum(fcst−act) / sum(act)	±5%	Systematic over/under
FVA	final acc − naive acc	>0	Is human input helping
Inventory turns	COGS / avg inv	6-8+	Working-capital efficiency
E&O %	excess$ / total inv$	falling	Stranded cash
Fill rate	met / demanded	95-98%	Customer service
Plan attainment	act vs committed plan	within tolerance	Decision vs forecast problem

How to actually use these

Three rules separate scorecards from action. Skip any one and the dashboard goes back to being wallpaper.

• Weight by revenue, always. Unweighted accuracy flatters you and points you at the wrong SKUs.
• Segment by ABC/XYZ. A single portfolio number hides everything. A/X items should be tight; the C/Z tail never will be, and chasing it wastes effort.
• Pair an accuracy metric with a money metric. WMAPE next to E&O and turns shows whether better numbers are turning into freed cash.

Set the cadence, then hold it

A KPI you review quarterly is a KPI nobody owns. Run accuracy, bias, and FVA monthly, tied to the demand-review step of your S&OP cycle, with one named owner per metric.

The reason monthly matters: bias and tracking signal only catch a drift if you're looking often enough to act before the cash is already trapped. Quarterly review means you find the over-build a full quarter after you paid for it.

Don't measure accuracy in a vacuum

The trap is celebrating an accuracy win while the cash story gets worse. I've watched teams cheer a 5-point WMAPE improvement while E&O kept climbing, because they got better at forecasting the wrong, over-biased number.

This is also where the AI conversation gets real. McKinsey reports that AI-driven forecasting can cut errors by 20 to 50% and reduce lost sales from unavailability by up to 65% — but only if your bias and inventory metrics confirm the gain is reaching the balance sheet. Gartner predicts 70% of large organizations will adopt AI-based supply chain forecasting by 2030, which makes a trustworthy KPI baseline more urgent, not less. Accuracy plus bias plus cash, together, or you're flying blind.

Frequently asked questions

What is the single most important demand planning KPI?

Forecast bias, if you can only watch one. Bias is systematic, so it compounds in the same direction every cycle and quietly funds excess inventory or stockouts. WMAPE is the best accuracy headline, but bias is the one that drains cash without showing up on an accuracy report.

What is a good forecast accuracy for a manufacturer?

For A-items in discrete manufacturing, WMAPE under 30% is solid and under 20% is strong, while your C/Z tail will run far higher and shouldn't be chased. There's no universal benchmark; acceptable error varies by industry, product type, planning horizon, and volatility. Segment by ABC/XYZ and judge each tier against itself.

How is WMAPE different from MAPE?

MAPE averages percentage errors equally, so a tiny miss on a low-volume SKU can swamp a big miss on your flagship. WMAPE weights each error by volume or revenue using sum|actual − forecast| / sum(actual), which reflects real business impact. That's why a 2024 benchmark study found 52% of planners use WMAPE as their primary metric.

What is Forecast Value Added (FVA) and why does it matter?

FVA compares your final forecast accuracy against a naive baseline like last-period-equals-next-period. If a process step or override doesn't beat the naive number, it's destroying value. Research summarized on the SAS Business Forecasting Deal blog found roughly half of real-world forecasts were worse than a naive random walk, which is exactly the waste FVA exposes.

How often should demand planning KPIs be reviewed?

Monthly, tied to the demand-review step of your S&OP cycle, with one named owner per metric. Bias and tracking signal only help if you look often enough to act before the cash is trapped. Quarterly review means you discover an over-build a full quarter after you paid for it.