SynapseOne /blog
es ← back to blog

MAPE, MAE, RMSE: How to measure your demand forecast accuracy

· SynapseOne Team

If your company does demand forecasting, you need to know if it’s working. It’s not enough to “feel” the forecast is good, you need objective metrics to measure how accurate it is.

The three standard industry metrics are MAPE, MAE and RMSE. In this article we explain what they are, how they’re calculated, when to use each one and what accuracy level is professional.

Why You Need to Measure Forecast Accuracy

A forecast without metrics is like a car without a speedometer: you don’t know if you’re going 30 km/h or 120 km/h.

Measuring precision allows you to:

  • Compare methods: Is AI better than moving averages?
  • Detect problem products: Which SKUs have very wrong forecasts?
  • Improve over time: Track if your model is getting better
  • Justify investments: Demonstrate ROI of forecasting tools

According to the International Journal of Forecasting, companies that systematically measure accuracy reduce forecast errors by 15-25% in the first year.

MAPE: Mean Absolute Percentage Error

What is it?

MAPE measures the average error as a percentage of actual demand. It’s the most used metric because it’s easy to interpret: “my forecast has a 15% error”.

Formula

MAPE = (100 / n) × Σ |Actual - Forecast| / Actual

Where:

  • n = number of periods (days, weeks, months)
  • Actual = actual demand
  • Forecast = predicted demand
  • | | = absolute value (unsigned error)

Step-by-Step Example

Sales forecast for PROD-001 over 5 days:

Day Forecast Actual Error % Error
1 100 95 5 5.3%
2 110 120 10 8.3%
3 105 100 5 5.0%
4 90 85 5 5.9%
5 115 110 5 4.5%

Calculation:

MAPE = (5.3 + 8.3 + 5.0 + 5.9 + 4.5) / 5
MAPE = 29.0 / 5 = 5.8%

Interpretation: Your forecast has an average error of 5.8%. If your average demand is 100 units/day, you’re off by ±6 units on average.

Advantages of MAPE

✅ Easy to communicate: “We have 8% error” is clear for managers ✅ Scale independent: You can compare products of 10 units vs 10,000 ✅ Industry standard: Everyone uses MAPE

Disadvantages of MAPE

❌ Problem with zero or very low demand: If Actual = 0, MAPE = infinity ❌ Penalizes overestimation more: Underestimating 100 units hurts more than overestimating ❌ Doesn’t work for intermittent products: Products with many days of 0 sales

More on MAPE limitations in Hyndman & Athanasopoulos - Forecasting: Principles and Practice

MAE: Mean Absolute Error

What is it?

MAE measures the average error in absolute units. Unlike MAPE (which is percentage), MAE tells you: “your forecast is off by 8 units on average”.

Formula

MAE = (1 / n) × Σ |Actual - Forecast|

Example with Same Data

Day Forecast Actual Absolute Error
1 100 95 5
2 110 120 10
3 105 100 5
4 90 85 5
5 115 110 5

Calculation:

MAE = (5 + 10 + 5 + 5 + 5) / 5
MAE = 30 / 5 = 6 units

Interpretation: Your forecast is off by 6 units on average, regardless of whether it underestimates or overestimates.

Advantages of MAE

✅ Easy to understand: “I’m off by 6 units” ✅ Works with low demand: No division by zero ✅ Penalizes errors linearly: A 10-unit error weighs twice as much as a 5-unit error

Disadvantages of MAE

❌ Scale dependent: You can’t compare a product of 10 units/day with one of 1,000 ❌ Requires context: Is MAE of 6 good? Depends on whether you sell 50 or 500 units

RMSE: Root Mean Square Error

What is it?

RMSE is similar to MAE but penalizes large errors more. It’s used when you want to be especially aggressive against extreme errors.

Formula

RMSE = √[(1 / n) × Σ (Actual - Forecast)²]

Example with Same Data

Day Forecast Actual Error Error²
1 100 95 5 25
2 110 120 10 100
3 105 100 5 25
4 90 85 5 25
5 115 110 5 25

Calculation:

RMSE = √[(25 + 100 + 25 + 25 + 25) / 5]
RMSE = √[200 / 5]
RMSE = √40
RMSE ≈ 6.3 units

Interpretation: RMSE is 6.3 units, slightly higher than MAE (6.0). This indicates you have some larger errors that inflate the metric.

Advantages of RMSE

✅ Penalizes outliers: A 20-unit error weighs much more than two 10-unit errors ✅ Useful for ML: Many algorithms minimize RMSE by default ✅ Sensitive to large errors: Detects forecasts with very wrong spikes

Disadvantages of RMSE

❌ Difficult to interpret: The square root complicates explanation ❌ Less intuitive than MAPE: It’s not a percentage or direct units ❌ Scale dependent: Like MAE, you can’t compare products of different magnitude

More on RMSE in Journal of Business Forecasting

Comparison Table: MAPE vs MAE vs RMSE

Metric Unit Main Pro Main Con When to Use
MAPE Percentage Easy to communicate Fails with low demand Products with stable demand >10
MAE Units Works with low demand Not scalable All products (default)
RMSE Units Penalizes large errors Difficult to explain When outliers are critical

What’s a Good Forecast Accuracy?

It depends on the industry and product, but here are general benchmarks:

By MAPE

MAPE Interpretation Action
< 10% Excellent ✅ Keep current method
10-20% Good ✅ Acceptable for most cases
20-30% Acceptable ⚠️ Consider improvements
30-50% Poor ❌ Forecast not very reliable
> 50% Very poor ❌ Almost like flipping a coin

Industry specific:

  • High-turnover retail: 5-15% (stable products like basic foods)
  • Fashion: 20-40% (high variability due to trends)
  • Spare parts: 30-60% (very sporadic demand)
  • OTC pharmacy: 10-20% (relatively stable demand)

Source: APICS - Supply Chain Benchmarking

By MAE

There’s no universal standard because it depends on the product scale. Use this criterion:

Good MAE ≤ 10% of Average Demand

Example:

  • Average demand: 100 units/day
  • Acceptable MAE: ≤ 10 units

Combining Metrics: The Professional Approach

Advanced companies don’t use a single metric, they use all three in combination:

Recommended Strategy

  1. MAPE as primary metric: To communicate accuracy to the team
  2. MAE for debugging: Identify products with large errors in units
  3. RMSE for optimization: When training ML models

Real Example

Product MAPE MAE RMSE Diagnosis
PROD-001 8% 5 u 6 u ✅ Excellent, consistent errors
PROD-002 25% 8 u 18 u ⚠️ Has outliers (RMSE >> MAE)
PROD-003 45% 2 u 3 u ⚠️ High MAPE but low MAE = low demand

Insights:

  • PROD-001: Good forecast, keep it up
  • PROD-002: Investigate why there are wrong spikes (missed promotions?)
  • PROD-003: High MAPE is misleading, absolute error is actually small

How SynapseOne Calculates Accuracy Automatically

SynapseOne calculates all three metrics (MAPE, MAE, RMSE) for each product automatically:

  1. Historical tracking: Compares forecast vs actual sales every day
  2. Rolling metrics calculation: Last 30 days, 60 days, 90 days
  3. Problem product detection: Alert if MAPE > 30%
  4. Model adjustment: If accuracy drops, try another ML algorithm

Dashboard:

📊 PROD-001 - Forecast Accuracy (last 30 days)

  • MAPE: 8.2% ✅ Excellent
  • MAE: 5.1 units
  • RMSE: 6.3 units
  • Trend: Improving (+2% vs previous month)

All calculated automatically, without spreadsheets.

Common Mistakes When Measuring Accuracy

❌ Mistake 1: Using only MAPE for low-demand products If you sell 5 units/day and your forecast says 8, MAPE is 60% but real error is only 3 units. Use MAE.

❌ Mistake 2: Not filtering anomalous data If you had a sales spike from a non-repeatable promotion, including it in the calculation skews metrics. Clean outliers.

❌ Mistake 3: Measuring accuracy only once Accuracy should be monitored continuously. A model with MAPE 10% today can deteriorate to 25% in 3 months if demand patterns change.

❌ Mistake 4: Comparing products of different scales with RMSE/MAE A product of 10 units/day with MAE = 2 is better than one of 1,000 units/day with MAE = 50, even though the absolute number is higher. Use MAPE to compare.

When to Recalculate Accuracy

  • Weekly: For critical or high-turnover products
  • Monthly: For the rest of the catalog
  • After changes: New promotion, supplier change, new season
  • When performance drops: If you notice more stockouts or overstock than normal

Conclusion

MAPE, MAE and RMSE are the three essential metrics to measure if your forecast is working.

  • MAPE: Your primary metric, easy to communicate (goal: <20%)
  • MAE: For low-demand products or debugging
  • RMSE: To optimize ML models

A forecast without metrics is like navigating without a compass. Measuring accuracy allows you to continuously improve and justify investments in forecasting technology.

Want us to calculate your forecast accuracy automatically?

SynapseOne tracks MAPE, MAE and RMSE for each product and alerts you when accuracy drops. No spreadsheets, no manual formulas.

Try SynapseOne free →

References

← back to blog