{"id":"fpa-forecast-accuracy-lines","audience":"fpa","category":"finance-work","track":"finance","technique":"verification","level":"intermediate","title":"Score forecast accuracy with MAPE and bias, using restated actuals","free":false,"task":"You're reviewing how well the revenue forecast has done. Write a prompt for plain text with lines starting \"Jan:\", \"Feb:\", \"Mar:\", \"Apr:\", \"MAPE:\" and \"Bias:\". Month lines show |forecast - actual| / actual as a percent to one decimal, with the numbers. MAPE is the average of the four, and Bias is total forecast minus total actual with its percent of actual. Use restated actuals. Under 80 words.","materialLabel":"Forecast vs actual, in $ thousands (attached to your prompt automatically)","material":"Revenue forecast accuracy review - figures in $ thousands\n\nMonth | Forecast | Actual (as first reported)\nJan | 100 | 90\nFeb | 120 | 130\nMar | 110 | 110\nApr | 140 | 112\n\nRestatement note (controller, Nov 3): April actual revenue was restated to 100 after a 12 credit memo was applied to the month. January to March are unchanged.\n\nDefinitions used by FP&A:\n- Absolute percent error = |forecast - actual| / actual\n- MAPE = average of the monthly absolute percent errors\n- Bias = total forecast minus total actual; positive means the forecast was too high\n\nSales leadership commented that the forecast was 'pretty good except April'. Their view is not part of the metric.","rules":[{"type":"plainText"},{"type":"maxWords","max":80}],"promptHints":[{"label":"Names the six lines","patterns":["jan:","mape:","bias:"]},{"label":"Uses the restated figure","patterns":["restated","100"]},{"label":"Gives the formula","patterns":["forecast - actual","/ actual"]},{"label":"Defines MAPE and bias","patterns":["average","total forecast"]},{"label":"Caps length","patterns":["\\d+\\s*words","plain"]}],"mistakes":[{"title":"Using the first reported April","why":"The AI uses 112 and gets a 25.0% error.","fix":"Say: \"Use the restated April actual of 100.\"","checks":["r0"]},{"title":"Dividing by forecast","why":"The AI divides the error by the forecast instead of actual.","fix":"Add: \"Percent error = |forecast - actual| / actual.\"","checks":["r0"]},{"title":"Missing bias math","why":"The AI says 'over-forecast' without totals.","fix":"Say: \"Bias = total forecast minus total actual, with percent of actual.\"","checks":["r2"]},{"title":"Wrong MAPE","why":"The AI averages the dollar errors or skips the addition.","fix":"Add: \"MAPE: average of the four percentages, with the addition.\"","checks":["r1","plainText"]},{"title":"Trusting the AI to count words","why":"AI is bad at counting and runs past word limits.","fix":"Ask for fewer words than the real cap and keep each line short.","checks":["maxWords"]}],"hint":"The first-reported April actual is outdated. Give the formulas and which actual to use.","rubricLabels":["Month lines show Jan 10/90 = 11.1%, Feb 10/130 = 7.7%, Mar 0/110 = 0.0% and Apr 40/100 = 40.0% (using the restated April actual of 100)","MAPE line shows (11.1 + 7.7 + 0.0 + 40.0) / 4 = 14.7%","Bias line shows 470 - 430 = +40, which is 9.3% of actual, and the forecast was too high"]}