EN
Bookkeeping

How to Measure Sales Forecast Accuracy: Forecast Error, Weekly Error Table, WAPE and MAPE, Bias, the Cost of Over- and Under-Forecasting and Safety Margin Training

How do you measure sales forecast accuracy? Forecast error, WAPE, bias, the cost of over- and under-forecasting and the safety margin: an 8-lesson guide with examples.

Contents 14 sections
  1. What will you be able to do by the end of this training?
  2. Lesson 1: How is forecast error defined?
  3. Lesson 2: How do you build a weekly error table?
  4. Lesson 3: How are WAPE and MAPE calculated?
  5. Lesson 4: What is bias, and in which direction is the forecast wrong?
  6. Lesson 5: What is the cost of over- and under-forecasting?
  7. Lesson 6: How does a safety margin lower the cost of forecast error?
  8. Lesson 7: How does accuracy change by product group?
  9. Lesson 8: How do you monitor forecast accuracy?
  10. What are the most common mistakes when measuring forecast accuracy?
  11. How does NetKasam make forecast accuracy tracking easier?
  12. Mini quiz: Test what you have learned
  13. Conclusion: Write down the forecast, compare it with reality, know which direction you miss
  14. Frequently Asked Questions About Sales Forecast Accuracy

You made a sales forecast, placed your inventory order accordingly and set up your campaign calendar accordingly. So how close was the forecast? Most businesses make a forecast but never compare it with reality afterward; because the error stays invisible, they miss in the same direction every period. Measuring forecast accuracy means putting the size and direction of the error into figures, then calculating which direction of miss is more expensive and adjusting the forecast in that direction.

This article is not a narrative but an 8-lesson practical training. We will answer the question how to measure sales forecast accuracy using an example product group over eight weeks: the definition of forecast error, the weekly error table, the weighted absolute percentage error (WAPE) and the mean absolute percentage error (MAPE), forecast bias, the cost of over- and under-forecasting, the effect of a safety margin, the difference in accuracy by product group and monitoring indicators. At the end there is a mini quiz and an answer key. All figures are example values chosen to demonstrate the method; follow the same steps with your own data.

What will you be able to do by the end of this training?

In short: You will calculate forecast error weekly, find WAPE and bias, work out the monetary cost of the error and adjust the forecast in the direction that lowers cost.

Training mapSix steps to measuring forecast accuracy
  1. 01
    RecordWrite down each period's forecast in advance and do not change it.
  2. 02
    CompareAt period-end, calculate the difference from actuals.
  3. 03
    RateDivide the absolute error by total actual sales (WAPE).
  4. 04
    Find the directionSee whether the forecast is too high or too low.
  5. 05
    CostWork out the monetary cost of under- and over-forecasting.
  6. 06
    AdjustAdd a safety margin in the direction that lowers cost.

The 8 lessons of the training: (1) how forecast error is defined, (2) the weekly error table, (3) WAPE and MAPE, (4) bias: which direction the forecast misses, (5) the cost of error, (6) the safety margin, (7) accuracy by product group, (8) monitoring indicators.

Lesson 1: How is forecast error defined?

In short: Forecast error is the actual quantity subtracted from the forecast quantity. A positive error means over-forecasting; a negative error means under-forecasting.

FormulaThe four measures of forecast error
  1. Error = Forecast − Actual
  2. Absolute error = |Forecast − Actual|
  3. WAPE = Total absolute error ÷ Total actual
  4. Bias = Total error ÷ Total actual

It is important to keep the sign of the error. If you over-forecast by 100 units one week and under-forecast by 100 units the next, the sum of the two errors looks like zero; but you were wrong both weeks. That is why two separate measures are used: the error that shows direction (bias) and the absolute error that shows size (accuracy). A forecast can be measured only if it is written down in advance: saying "I was expecting this anyway" at period-end makes the measurement meaningless. We explained how to build the forecast in the article how to make a sales forecast; this article focuses on measuring the success of that forecast.

Lesson 2: How do you build a weekly error table?

In short: For each week, write the forecast, actual, error, absolute error and absolute percentage error side by side.

The example product group's eight-week forecast and actual sales (units) are as follows:

WeekForecastActualError (forecast − actual)Absolute errorAbsolute error %
11.000950+5050%5,3
21.0001.100−100100%9,1
31.1001.000+100100%10,0
41.1001.250−150150%12,0
51.2001.20000%0,0
61.2001.000+200200%20,0
71.3001.400−100100%7,1
81.3001.200+100100%8,3
Total9.2009.100+100800

The absolute error percentage is found by dividing the absolute error by that week's actual sales: for example, week 6 is 200 ÷ 1,000 = 20.0%. The totals in the table should also be checked: the total of the forecasts is 1,000 + 1,000 + 1,100 + 1,100 + 1,200 + 1,200 + 1,300 + 1,300 = 9,200, and the total of actuals is 950 + 1,100 + 1,000 + 1,250 + 1,200 + 1,000 + 1,400 + 1,200 = 9,100 units. To aggregate sales data across channels, use the reports on your order management screen; for how to read the channel breakdown, see the article how to measure profitability by sales channel.

ChartWeekly absolute error percentage: in week 6 the forecast missed by more than twice the average
  1. Week 15.3%
  2. Week 29.1%
  3. Week 310.0%
  4. Week 412.0%
  5. Week 50.0%
  6. Week 620.0%
  7. Week 77.1%
  8. Week 88.3%

Prepared with example values. Each rate: absolute error ÷ actual sales.

Lesson 3: How are WAPE and MAPE calculated?

In short: WAPE divides total absolute error by total actual sales. In the example, 800 ÷ 9,100 = 8.8%; accuracy is 100 − 8.8 = 91.2%.

WAPE (weighted absolute percentage error) is the most useful measure for a business, because the error in a high-sales week carries more weight than the error in a low-sales week. Calculation: 50 + 100 + 100 + 150 + 0 + 200 + 100 + 100 = 800 units of absolute error; 800 ÷ 9,100 = 0.0879, that is 8.8%. Accuracy is 91.2%.

MAPE, on the other hand, is the average of the weekly percentages: (5.3 + 9.1 + 10.0 + 12.0 + 0.0 + 20.0 + 7.1 + 8.3) ÷ 8 = 71.8 ÷ 8 = 9.0%. The two come out close; however, MAPE can blow up in weeks when actual sales are very small (as sales approach zero, the percentage goes to infinity). That is why WAPE is preferred, especially for low-volume products. Whichever measure you choose, use the same one every period; if the measure changes, the comparison breaks.

Lesson 4: What is bias, and in which direction is the forecast wrong?

In short: Bias is the sum of errors taken with their sign. In the example, the total error is +100 units, so the forecast is indeed 1.1% too high.

The sum of the errors is +50 − 100 + 100 − 150 + 0 + 200 − 100 + 100 = +100 units. Divided by total actual sales, 100 ÷ 9,100 = 1.1%: on average the forecast is slightly high. The average weekly bias is 100 ÷ 8 = +12.5 units. Because this value is small, it can be said that the bias is not systematic and that most of the errors fluctuate randomly.

Error directionNumber of weeksWeeksTotal units
Over-forecast (+)41, 3, 6, 8450
Under-forecast (−)32, 4, 7350
Exact hit150

In the weeks that were over-forecast, the error is 50 + 100 + 200 + 100 = 450 units, and in the weeks that were under-forecast it is 100 + 150 + 100 = 350 units; 450 − 350 gives a net bias of 100. Even when bias is close to zero, absolute error can be high: the two errors cancel each other out, but both weeks are wrong. That is why both must be read together. The reason bias keeps drifting in one direction is mostly that targets are set optimistically; to set targets realistically, see the article how to set and track a sales target.

Lesson 5: What is the cost of over- and under-forecasting?

In short: Under-forecasting creates lost sales, while over-forecasting creates excess inventory and carrying cost. The unit cost of the two is often very different.

The example product has a purchase cost of ₺200 and a unit contribution of ₺90. Let the inventory carrying cost be assumed at 2% per month (for details, see the article how to calculate inventory cost). If the 450 units that were over-forecast sit on the shelf for a month, the carrying cost is 450 × ₺200 × 2% = ₺1,800. If the 350 units that were under-forecast could not be sold, the lost contribution is 350 × ₺90 = ₺31,500. The total cost of error is 1,800 + 31,500 = ₺33,300.

ChartThe cost of forecast error: under-forecasting is seventeen times more expensive than over-forecasting
  1. Over-forecast (450 units, carrying)₺1,800
  2. Under-forecast (350 units, lost contribution)₺31,500

Prepared with example values. Over-forecast: 450 × 200 × 2%. Under-forecast: 350 × ₺90.

This calculation is a warning: even when the error quantity is equal, the cost is not equal. Here under-forecasting loses ₺90 per unit, while over-forecasting loses ₺4 per unit (200 × 2%). If the product's inventory holding cost is high (perishable, going out of fashion, short shelf life) or its contribution is low, the picture reverses. For a detailed calculation of lost sales, you can use the article how to calculate the cost of stockouts, and for what to do with excess inventory, the article how to do an inventory aging analysis.

Lesson 6: How does a safety margin lower the cost of forecast error?

In short: If under-forecasting is more expensive, adding a margin to the forecast lowers the total cost. In the example, a 5% margin lowers the cost from ₺33,300 to ₺19,160.

When 5% is added to the forecast, the weekly forecasts become 1,050, 1,050, 1,155, 1,155, 1,260, 1,260, 1,365, 1,365. Compared with actuals, the over-forecast weeks are 1 (100), 3 (155), 5 (60), 6 (260) and 8 (165), totaling 740 units; the under-forecast weeks are 2 (50), 4 (95) and 7 (35), totaling 180 units.

ScenarioOver (units)Under (units)Carrying costLost contributionTotal
Current forecast4503501.800 ₺31.500 ₺33.300 ₺
5% safety margin7401802.960 ₺16.200 ₺19.160 ₺
Chart5% safety margin: lost contribution falls by ₺15,300 while carrying cost rises by ₺1,160
  1. Error cost of the current forecast₺33,300
  2. Reduction in lost contribution (31,500 → 16,200)−₺15,300
  3. Increase in carrying cost (1,800 → 2,960)₺1,160
  4. Error cost with the safety margin₺19,160

Prepared with example values. Total: 33,300 − 15,300 + 1,160 = ₺19,160.

The cost has fallen by 33,300 − 19,160 = ₺14,140. This does not mean the margin can be increased without limit: as the margin grows, the cost of excess inventory grows too, and beyond a certain point the total cost starts to rise. The way to find the right margin is to try several values (0%, 5%, 10%) in the same table and choose the one at which the total cost is lowest. We explained how to calculate the margin by product in the article how to calculate safety stock.

Lesson 7: How does accuracy change by product group?

In short: Total accuracy hides errors at the product level. Measuring by product group shows for which products the forecasting method should change.

In the weekly total forecast, positive and negative errors partly cancel each other out; looking at the product or group level, the error grows. Let the example product group's eight-week sales be split into three subgroups:

SubgroupActual (8 weeks)Absolute errorWAPE
Fast-moving products6.000380%6,3
Medium-moving products2.400340%14,2
Slow-moving products700320%45,7
Total9.1001.040%11,4

6,000 + 2,400 + 700 = 9,100 in actuals, 380 + 340 + 320 = 1,040 in absolute error; 1,040 ÷ 9,100 = 11.4%. At the subgroup level, WAPE rises to 11.4%, because errors that cancel each other out at the total level become visible here. The lowest-volume group accounts for only 7.7% of sales (700 ÷ 9,100) yet carries 30.8% of the error (320 ÷ 1,040). For slow-moving products, using a simple rule instead of forecasting, for example "order when inventory falls below three months of sales," often gives better results. To group products this way, use the method in the article how to do an ABC analysis, and to set the first forecast for new products, the article how to determine the first order quantity for a new product.

For seasonal products, error is usually high at the start of the season, because there is no history or demand arrives at a different rhythm. To plan the calendar for these products separately, see the article how to plan inventory for seasonal products. If you notice that campaign weeks distort the forecast, use the baseline method in the article how to measure the post-campaign sales drop to separate out the campaign's real effect.

Recommended readingHow to Make a Sales Forecast?

Lesson 8: How do you monitor forecast accuracy?

In short: Monitor four indicators in the same table every month: WAPE, bias, the lost contribution from under-forecasting and the carrying cost of over-forecasting.

IndicatorHow it is calculatedSample value
WAPETotal absolute error ÷ Total actual%8,8
Accuracy100 − WAPE%91,2
BiasTotal error ÷ Total actual+%1,1
Lost contribution from under-forecastingUnder-forecast units × Unit contribution31.500 ₺
Carrying cost of over-forecastingOver-forecast units × Unit cost × Monthly rate1.800 ₺
Takeaways from the exampleThe four numbers of the forecast accuracy example
  • 8.8%error rate of the eight-week forecast (WAPE)
  • 91.2%forecast accuracy
  • ₺33,300total error cost of the current forecast
  • ₺14,140reduction achieved by the 5% safety margin

In monitoring, direction is read first, then size: if bias has the same sign three months in a row, the forecasting method or the target has drifted systematically and should be corrected. If WAPE is rising, the demand structure may have changed (a new campaign, a new channel, the season). To add the table to the weekly management report, it is enough to add one line to the layout in the article how to prepare a weekly management report. It is also important to know how reliable the forecast is before busy periods; for preparation, see the article how to prepare for a busy campaign period.

Recommended ReadingHow to Calculate Cost per Order: Cost Items, Profit per Order, the Fallacy of an Incomplete Calculation, Fixed Items, Break-Even Basket and Cost Reduction Scenarios Training

What are the most common mistakes when measuring forecast accuracy?

  1. Writing or changing the forecast after the fact. The measurement loses its meaning; the forecast must be recorded at the start of the period.
  2. Adding up errors with their signs and calling it "zero error." Positive and negative errors cancel each other out; absolute error must be calculated separately.
  3. Measuring only at the total level. High error at the product and group level stays hidden.
  4. Applying the same method to every product. For slow-moving products, an inventory rule works better than a forecast.
  5. Leaving the direction of error without a cost. The unit cost of under- and over-forecasting differs; the safety margin is chosen accordingly.
  6. Not separating out campaign weeks. The campaign effect looks as if it were forecast error.
  7. Changing the measure often. Comparison becomes impossible; WAPE or MAPE should be chosen and the same one kept.

How does NetKasam make forecast accuracy tracking easier?

  • Net sales in one place. Store, marketplace and website orders come together in the same panel; with order management, you do not add up actual sales from channels by hand.
  • Saved and exportable reports. You save the sales report with its criteria, export it to Excel and paste it into your forecast table (reports).
  • Inventory movements and turnover. The inventory movements report shows stockout days, and the inventory turnover report shows a product's speed (inventory management).
  • Critical inventory level. By setting a critical inventory level for your product, you limit the effect of under-forecasting.
  • Campaign module. You define campaign periods by product and customer group and report the results.
  • Bookkeeping is free. e-Invoice, customer and supplier accounts, inventory, cash, bank and expense management are free in every plan; see the pricing page for details.

Mini quiz: Test what you have learned

In short: Answer the 5 questions below, then compare with the answer key. 4 or 5 correct shows you have completed the training.

  1. If the forecast is 1,100 and the actual is 1,250, what is the error and in which direction is it?
  2. If total absolute error is 800 and total actual is 9,100, what is WAPE in percent (one decimal)?
  3. If total error is +100 and total actual is 9,100, what is bias in percent?
  4. If 350 units were under-forecast and the unit contribution is ₺90, how many ₺ is the lost contribution?
  5. If 450 units were over-forecast, the unit cost is ₺200 and the monthly carrying rate is 2%, how many ₺ is the carrying cost?

Answer key:

  1. 1,100 − 1,250 = −150; under-forecast.
  2. 800 ÷ 9.100 = %8,8.
  3. 100 ÷ 9,100 = 1.1% (forecast is high).
  4. 350 × 90 = 31.500 ₺.
  5. 450 × 200 × %2 = 1.800 ₺.

Frequently Asked Questions About Sales Forecast Accuracy

How do you measure sales forecast accuracy?

You calculate the difference between the forecast and actual sales; the sum of the absolute differences is divided by total actual sales. This ratio is WAPE, and subtracting it from 100 gives accuracy.

What is the difference between WAPE and MAPE?

MAPE is the average of weekly percentages; WAPE divides the total error by total sales. WAPE does not blow up in low-volume periods, which is why it is generally more useful for businesses.

What is bias?

It is the sum of errors taken with their sign. If it is positive, the forecast is consistently too high; if negative, consistently too low. Absolute error can be high even when bias is close to zero.

What should good forecast accuracy be?

It depends on the product. Low error is expected for steady sellers, while high error is normal for new or volatile products. The best comparison is your own history.

Do over- and under-forecasting cost the same?

No. Under-forecasting leads to lost sales and lost contribution, while over-forecasting leads to inventory and carrying costs. Unit costs differ; in the example, under-forecasting loses ₺90 per unit and over-forecasting loses ₺4 per unit.

What is a safety margin?

It is a safety buffer added to the forecast. If under-forecasting is more expensive, adding a margin lowers the total cost; if the margin grows too large, the cost of excess inventory raises the total again.

At what level should I measure the forecast?

At both the total and the product group level. At the total level, positive and negative errors cancel each other out; at the group level the real error shows.

How often should forecast accuracy be measured?

Every month; if the forecast is made weekly, it should be tracked weekly and summarized monthly. If the forecast is not recorded in advance, it cannot be measured.

Conclusion: Write down the forecast, compare it with reality, know which direction you miss

The most useful thing you can do today is to write down and save a forecast for this week and calculate the difference from actuals at the end of the week. After a few periods, WAPE and bias will tell you how far and in which direction the forecast is off. For products where under-forecasting is expensive, adding a small safety margin can by itself noticeably lower the total error cost in the first month.

Bookkeeping Is Now Completely Free

Manage your bookkeeping and e-Invoice operations with NetKasam without paying a thing. Customer and supplier accounts, inventory, collections and more in one panel, completely free.

Start FreeCustomer and Supplier Account Tracking

Read similar articles

19 free tools

Commission, shipping, VAT and profit calculations in your browser, no sign-up needed.

All tools