Demand Planning Metrics: The KPIs That Actually Measure Forecast Performance
Forecast accuracy, MAPE, bias, and fill rate — the demand planning metrics that tell you whether your planning process is working. How to calculate them, what benchmarks to target, and which combinations reveal the root cause of supply chain problems.
Photo: Unsplash
Table of Contents
Most supply chain teams track the wrong demand planning metrics, or track the right ones incorrectly, or track them correctly without using them to improve anything.
Demand planning metrics are not a reporting exercise. They are a diagnostic system. When forecast accuracy is low, metrics tell you whether the problem is the statistical model, commercial input quality, data integrity, or process discipline. When service levels drop, metrics tell you whether the cause is forecast error, supply constraints, or execution failures. The metrics only create value when they’re used to find and fix the root cause of planning failures.
This guide covers the demand planning KPIs that matter, how to calculate them correctly, what benchmarks to target, and how to use them together to diagnose specific problems.
The Core Demand Planning Metrics
1. Forecast Accuracy (FA)
What it measures: The percentage of periods where the forecast was within a defined tolerance of actual demand.
Formula:
Forecast Accuracy = 1 - (|Forecast - Actual| / Actual)Expressed as a percentage. A forecast of 100 units against actual demand of 110 units gives:
FA = 1 - (|100 - 110| / 110) = 1 - (10/110) = 90.9%How to aggregate: Forecast accuracy should be calculated at the SKU/period level and then aggregated. Aggregating demand and forecast before calculating accuracy (adding up all SKUs, then measuring accuracy of the total) is a common mistake that masks individual SKU-level errors through cancellation.
What to target: Industry benchmarks vary significantly by category, but general reference points:
| Category | Strong FA | Average FA |
|---|---|---|
| High-velocity stable products | 90%+ | 75–85% |
| Seasonal products | 80%+ | 65–78% |
| Promotional items | 75%+ | 55–70% |
| New product introductions | 60%+ | 40–55% |
| Slow-moving / intermittent demand | Context-dependent | — |
Limitation: Forecast accuracy alone doesn’t tell you whether errors are systematic (bias) or random. Two planning programs with identical average accuracy can have very different operational impacts depending on whether their errors are biased or unbiased.
2. Mean Absolute Percentage Error (MAPE)
What it measures: The average magnitude of forecast error expressed as a percentage of actual demand. MAPE is the complement of forecast accuracy — a MAPE of 15% corresponds to a forecast accuracy of approximately 85%.
Formula:
MAPE = (1/n) × Σ |Forecast - Actual| / Actual × 100Where n is the number of periods.
Why it’s the standard metric: MAPE is scale-independent, which allows comparison across SKUs with very different volumes. An error of 10 units on a 50-unit SKU (20% MAPE) is operationally very different from an error of 10 units on a 1,000-unit SKU (1% MAPE). MAPE captures this distinction.
MAPE benchmarks:
| MAPE | Interpretation |
|---|---|
| Under 10% | Excellent — uncommon outside stable, high-volume categories |
| 10–20% | Strong — achievable with good process and data quality |
| 20–35% | Average — most mid-market organizations |
| 35–50% | Below average — significant inventory and service level impact |
| Over 50% | Poor — typically indicates data, process, or model problems |
Known limitations of MAPE:
- Undefined for zero actuals: When actual demand is zero, MAPE is mathematically undefined. This creates problems for slow-moving or intermittent demand SKUs.
- Asymmetric penalty: MAPE penalizes over-forecasting less than under-forecasting when actuals are low. A forecast of 150 against an actual of 100 gives 50% error; a forecast of 50 against an actual of 100 also gives 50% error — but the operational consequences are completely different.
- Averages can mislead: MAPE averages can be driven by a small number of high-error, low-volume SKUs. Always review MAPE distributions, not just the mean.
Alternatives for intermittent demand: For SKUs with frequent zero-demand periods, WMAPE (weighted MAPE, weighted by volume) or MAD/Mean ratio (Mean Absolute Deviation divided by mean demand) are more appropriate.
3. Forecast Bias
What it measures: Whether the forecast systematically over- or under-predicts demand. Bias is arguably more operationally damaging than raw error, because it creates systematic inventory imbalances rather than random variability.
Formula:
Bias = Σ (Forecast - Actual) / Σ Actual × 100A positive bias means the forecast consistently over-predicted (leading to excess inventory). A negative bias means the forecast consistently under-predicted (leading to stockouts).
Why bias matters: Random forecast errors cancel out over time — sometimes too high, sometimes too low, with safety stock absorbing the variability. Biased errors accumulate: consistently over-forecasting a product category builds inventory systematically; consistently under-forecasting creates chronic stockouts. Bias also compounds across time horizons — a 5% monthly bias becomes a significant inventory imbalance over a quarter.
Target: Bias should be close to zero. A bias within ±5% is generally acceptable; above ±10% indicates a systematic problem requiring investigation.
Common sources of positive bias (over-forecasting):
- Commercial teams consistently adding volume to meet budget targets
- Failure to remove discontinued or declining products from the plan
- New product launches that underperform launch forecasts
Common sources of negative bias (under-forecasting):
- Demand planning teams anchoring to conservative budgets rather than market signals
- Failure to capture promotional lift in the baseline forecast
- Systematic under-reporting of early demand signals from sales
4. Demand Plan Attainment
What it measures: The percentage of the approved demand plan that was actually shipped.
Formula:
Demand Plan Attainment = Actual Shipments / Demand Plan × 100Why it’s useful: Demand plan attainment distinguishes between forecast failure and supply execution failure. If forecast accuracy is high but attainment is low, the planning process is producing accurate demand plans that the supply chain isn’t executing against — typically indicating supply constraints, procurement failures, or operational execution issues.
If both forecast accuracy and attainment are low, the problem may be that the demand plan is unreliable enough that operations doesn’t use it as the basis for supply decisions.
Target: 90%+ for a well-functioning supply chain. Chronic attainment below 85% with high forecast accuracy indicates supply execution as the primary failure mode.
5. Customer Service Level / Order Fill Rate
What it measures: The percentage of customer orders fulfilled on time and in full. This is the output metric — the customer-facing consequence of demand planning and supply planning performance.
Key variants:
- Line fill rate: Percentage of order lines fulfilled complete
- Order fill rate: Percentage of orders where every line is fulfilled complete
- On-time in-full (OTIF): Combines delivery timing with completeness
- Case fill rate: Percentage of cases or units ordered that were shipped
Target: Industry benchmarks vary by channel and customer type:
| Channel | Typical Target |
|---|---|
| Key accounts / retail | 96–99% OTIF |
| General trade | 92–96% |
| Ecommerce / B2C | 95–99% |
| B2B / industrial | 90–95% |
Major retailers increasingly mandate OTIF rates and impose financial penalties for shortfalls — making this metric directly connected to cost performance.
6. Inventory Days of Supply (DOS) / Inventory Turns
What it measures: How much inventory is held relative to demand rate. Inventory turns = Annual COGS / Average Inventory Value. Days of Supply = Current Inventory / Average Daily Demand.
Why it belongs with demand planning metrics: High-quality demand planning produces appropriate inventory levels — lean enough to minimize carrying costs, sufficient to meet service level targets. Monitoring inventory turns alongside forecast accuracy reveals whether the planning process is producing the right inventory outcome.
Target: Benchmarks vary widely by industry and supply chain structure:
| Industry | Typical Inventory Turns |
|---|---|
| Grocery / FMCG | 15–25× |
| Consumer electronics | 8–15× |
| Apparel / fashion | 4–8× |
| Automotive | 10–20× |
| Industrial / B2B | 4–8× |
Improving forecast accuracy without a corresponding improvement in inventory turns suggests safety stock assumptions are too conservative — the accuracy gain isn’t being translated into inventory reduction.
Using Metrics Together: Diagnostic Combinations
Individual metrics reveal symptoms. Metric combinations reveal causes.
High MAPE + High Bias → Process Problem
When both accuracy is low and the error is directional, the problem is usually in the human input to the demand plan — commercial over-optimism, political number-setting, or poor market intelligence. Fix the process (demand review discipline, decoupling forecast from budget) before optimizing the algorithm.
High MAPE + Near-Zero Bias → Model Problem
Random errors with no systematic direction typically indicate the statistical model is a poor fit for the demand pattern — not capturing seasonality, failing on promotional products, or applying the wrong algorithm to intermittent demand SKUs. Review model selection and consider alternative forecasting approaches.
High Forecast Accuracy + Low Fill Rate → Supply Problem
If the forecast is accurate but orders aren’t filled, the supply chain isn’t using the demand plan effectively — procurement is running on reorder points rather than demand plan signals, or supplier lead times have extended without compensating safety stock. The fix is in supply planning and procurement execution, not demand forecasting.
High Forecast Accuracy + High Fill Rate + Low Inventory Turns → Safety Stock Problem
The demand plan is working and customers are served, but inventory is too high. Safety stock parameters are too conservative relative to actual demand variability. Reduce safety stock targets and monitor service levels to find the right balance.
Good Aggregate Metrics + Poor SKU-Level Performance → Mix Problem
Aggregated MAPE can look acceptable while individual SKU performance is highly variable — high-volume, stable SKUs carry the metric while high-value, volatile SKUs perform poorly. Always segment metrics by ABC category and demand pattern type to identify where to focus improvement.
Segmenting Metrics by ABC Category
Treating all SKUs equally when measuring forecast performance is a mistake. A-class SKUs drive most volume and most inventory value — they deserve more scrutiny and justify more sophisticated forecasting investment. C-class SKUs may have individual MAPE of 80%+ with minimal operational consequence.
Recommended segmentation:
| ABC Class | Measurement Frequency | Accuracy Target | Review Protocol |
|---|---|---|---|
| A (top 20% of volume) | Weekly | 85%+ accuracy / under 15% MAPE | Individual SKU review; planner-managed overrides |
| B (next 30% of volume) | Monthly | 80%+ accuracy / under 25% MAPE | Exception-based review; significant changes flagged |
| C (bottom 50% of volume) | Monthly/Quarterly | 70%+ accuracy / under 40% MAPE | Automated with statistical override only |
See our guide to ABC inventory analysis for the full methodology of ABC classification.
Measuring Forecast Accuracy: Common Mistakes
Measuring at the aggregate, not the SKU level. Aggregate accuracy looks better than SKU-level accuracy because errors cancel out. A demand plan that over-predicts product A by 100 units and under-predicts product B by 100 units looks accurate in aggregate — but creates real inventory imbalances at the product level.
Not measuring bias. Most organizations measure MAPE but not bias. Bias is the more actionable metric for process improvement — it identifies whether errors are systematic and which direction they run.
Using shipment data instead of demand data. Forecast accuracy should be measured against actual customer demand (orders placed), not shipments. If supply constraints caused partial fulfillment, measuring forecast vs. shipments will artificially inflate accuracy during periods of supply shortage.
Not tracking accuracy by planning horizon. A 12-week forward forecast is inherently less accurate than a 4-week forward forecast. Measuring accuracy without specifying the horizon creates comparisons that don’t distinguish between planning process performance and inherent demand uncertainty.
Measuring without acting. Forecast accuracy that’s reported but not used to investigate root causes and improve the process is a reporting cost, not a business investment. The metric creates value when it’s connected to structured root cause analysis and process improvement.
Connecting Metrics to Demand Planning Software
At mid-market scale and above, tracking demand planning metrics effectively requires dedicated software that integrates with your transaction history, runs statistical forecasts, and provides an analytics layer for performance measurement.
Manual spreadsheet tracking of MAPE, bias, and fill rate across thousands of SKUs is error-prone and time-consuming. Purpose-built platforms automate the metric calculation, surface exceptions, and provide historical trend views that make it possible to identify whether process changes are improving performance.
For an overview of the platforms that support demand planning measurement at different market segments, see our guide to best demand planning software.
Demand planning metrics are the diagnostic system for your planning process. For the foundational overview of how demand planning works, see what is demand planning. For the statistical methods that drive the baseline forecast these metrics measure, see demand forecasting methods.
Supply Chain Desk Editorial
The Supply Chain Desk editorial team covers logistics, freight management, warehouse operations, and supply chain technology. Our guides are written for operations professionals who need practical, data-backed insights to improve efficiency and reduce costs.