Why it matters
A forecast is only useful if you know how far to trust it. WAPE gives that in one number: over a period, how much the forecast missed compared with how much really happened. A WAPE of 12% means the misses add up to 12% of the actual total.
It suits advertising data better than MAPE, the plain average of daily percentage errors. MAPE divides each day's miss by that day's actual, so a Sunday with one conversion and a forecast of three counts as a 200% error and swamps the week. WAPE divides the total miss by the total actual, so each day weighs what it is worth.
WAPE = Σ |actual − forecast| ÷ Σ actual
Sum over every day, or every forecast, you are scoring. Lower is better, and 0% is a perfect forecast.
How to compute it
- Google Ads does not report forecast error. You compute WAPE yourself, from a forecast you saved and the figures that came in.
- Export the daily actuals from the campaigns table, segmented by day, once the days are final.
- Wait for conversion lag to play out before scoring conversions, or the actuals read low and the error looks larger than it is.
- Score many past forecasts, not one: a single month can be lucky.
An example
A law firm forecast its daily leads for one week, then set the forecast against what arrived.
Monday
- Actual
- 20
- Forecast
- 18
- Miss
- 2
Tuesday
- Actual
- 22
- Forecast
- 20
- Miss
- 2
Wednesday
- Actual
- 19
- Forecast
- 21
- Miss
- 2
Thursday
- Actual
- 21
- Forecast
- 20
- Miss
- 1
Friday
- Actual
- 16
- Forecast
- 18
- Miss
- 2
Saturday
- Actual
- 3
- Forecast
- 5
- Miss
- 2
Sunday
- Actual
- 1
- Forecast
- 3
- Miss
- 2
Week
- Actual
- 102
- Forecast
- 105
- Miss
- 13
WAPE is 13 ÷ 102 = 12.7%. MAPE on the same week is 44.8%, almost all of it from Saturday and Sunday, the days that matter least to the firm. Both describe the same forecast. WAPE describes it the way the business feels it.
WAPE has no sign, so pair it with bias, the signed error over the actual total. Here the forecast ran 3 leads over on 102, a bias of about +3%: close to neutral.
In Optizads
Optizads scores its forecasts with WAPE. Each account's history is cut at up to 36 past dates, five days apart. The forecast made from each cut is set against what followed over the next 7 days, the next 28 days and the rest of that month. The forecast itself is the daily median of several simple methods, scored against a plain 14-day average on the same cuts. The likely range on the forecasting chart comes from those past misses. You do not see an accuracy score in the app.
Common mistakes
- Comparing WAPE across accounts of very different size: a small account's daily conversions are noisy, so its WAPE runs higher with the same method.
- Scoring a forecast on days whose conversions are still arriving.
- Reading WAPE alone: a forecast that always runs 10% high and one that swings both ways can share the same WAPE.
Related terms
- Run rate: The run rate is what a figure will reach by the end of a period if it keeps its pace so far.
- Budget pacing: Budget pacing is tracking whether spend is on course to land on the month's budget, and correcting the daily budgets when it is not.
- Conversion lag: The time between a click and the conversion it leads to.
- Seasonality adjustment: A note to Smart Bidding that conversion rates will jump for a short, known period, such as a flash sale of a few days.
