Forecast accuracy

Forecast accuracy measures how close a forecast comes to actual demand. The most common form is one minus MAPE, expressed as a percentage:

Forecast Accuracy = 100% − MAPE

A forecast with MAPE of 20% has 80% accuracy. A forecast with 10% MAPE has 90% accuracy. Higher number = closer to actual demand. If you'd rather measure error in units than as a percentage, see MAE.

This is the planning metric that gets the most management attention and is the most often misunderstood. The headline number matters less than what you do with it.

What "good" looks like

Realistic targets vary wildly by SKU type:

• Mature high-volume SKUs: 85–95% accuracy
• Normal consumer SKUs: 75–85% accuracy
• High-variability SKUs (CV > 0.5): 50–70% accuracy
• New launches: 50–60% accuracy in the first 3 months
• Seasonal items in peak weeks: typically 10–15% lower than annual average

A category-level number averaging 85% sounds good. The same number can hide individual SKUs running at 50% accuracy that are responsible for most of the operational pain.

Why the headline number is the wrong focus

Three reasons:

• Aggregation hides variance. A high portfolio-level accuracy can mask terrible SKU-level accuracy on the items that actually matter.
• Accuracy without bias is incomplete. A forecast with 80% accuracy and zero bias is much healthier than one with 80% accuracy and chronic +15% bias.
• What you do with the number matters more than the number itself. A 75% forecast that's used to drive better safety stock decisions outperforms an 85% forecast that no one acts on.

How to actually use forecast accuracy

Three operational uses:

1. Watch the trend. Is your accuracy improving over time? Static accuracy means the planning process isn't learning.
2. Identify outlier SKUs. Sort by accuracy ascending. The bottom 5–10% of SKUs are usually responsible for most of the inventory pain. Focus the planning effort there.
3. Calibrate safety stock. Lower-accuracy SKUs need more safety stock. The safety-stock formula already implicitly accounts for this through demand variability, but explicit per-SKU accuracy buckets sharpen the calibration.

Forecast accuracy by horizon

Accuracy degrades with planning horizon:

This isn't a flaw — it's a fundamental property of demand. The further out, the more events can intervene. Setting accuracy targets that don't recognise this leads to manufactured planning despair.

Forecast accuracy vs forecast bias

Two different things, often confused:

• Accuracy = how close, in magnitude
• Bias = which direction the errors lean

A forecast can be:

• High accuracy + low bias = the goal
• High accuracy + high bias = consistent small errors in one direction (rare)
• Low accuracy + low bias = noisy but unbiased
• Low accuracy + high bias = noisy and skewed = worst case

Always report both.

Common ways accuracy gets gamed

When accuracy becomes a KPI, the planning team finds ways to optimise it without improving operational performance:

• Forecast at category level, not SKU. Category accuracy is always higher than SKU accuracy. Reporting category-level accuracy hides where the real problems are.
• Move forecasts late in the cycle. Updating the forecast 1 week before the period reduces error mechanically. The forecast becomes "what's about to happen" rather than "what we plan for."
• Apply accuracy adjustments to ignore outlier weeks. A promo-driven week is always a "non-standard" week if you're trying to make accuracy look better.
• Pick the easier metric. WAPE is usually higher than MAPE for the same forecast, because it weights large SKUs (which forecast better) more heavily.

The defence: track multiple metrics, report at SKU level, don't attach bonuses to the headline number, focus on the trend.