Skip to content
Faisal Mahmud TusherDigital marketing, SEO & analytics
with Faisal Tusher

Marketing Analytics and Reporting › Advanced

Checkpoint: Forecasting and Incrementality for Finance

Scenario questions set in an invented online wine merchant test whether you can forecast marketing outcomes in a form finance can use: scenarios with ranges, curves read only where the data supports them, seasonal coincidence kept out of the fit, test results turned into adjustments, the right assumption tested first and uncertainty explained without weakening the recommendation.
CheckpointAdvanced levelModule 2: Forecasting and incrementality for finance4 minUpdated

What this checkpoint covers

Corkplume Wines is an invented online wine merchant with a monthly wine club and a strong Christmas peak. The questions draw on the marginal returns and forecasting guide, the incrementality testing guide and the budget increase incrementality test.

Task

  • Write a decision and a reason for every question before reading on.
  • Count a right decision with a wrong reason as a miss, and reread the linked section for each miss.

Questions

  1. The finance director wants one figure for next year's new customers to put in the plan. What do you give her, and why?
  2. Using the example below, the agency extends the paid search curve to a weekly spend twice the highest Corkplume has run and forecasts a large gain. How far do you trust it?
  3. A regression across all of last year's weeks shows paid social returning far more than any other channel. Most of the high-spend weeks were in November and December. What is wrong?
  4. A geo test on paid social at current spend finds incremental orders well below the orders the platform attributed. How should finance use the result in forecasts and monthly reports?
  5. Three assumptions drive next year's forecast: the paid social curve at the higher spend proposed for spring, the brand search curve at current spend, and the wine club's renewal rate, which comes from the club's own records. Which do you test first?
  6. A budget-increase test finds a marginal cost per incremental order whose range spans the affordability threshold. What do you recommend?
  7. The finance director says the ranges make the forecast look like guesswork. How do you answer without weakening the recommendation?
Illustrative example — not a real result

Corkplume Wines: paid search and the social test

Paid search, non-seasonal weeks:

Weekly spendWeeksAverage new customers a week
£3,0007150
£6,00011225
£9,000 (highest run)6270

Target cost per new customer: £40.

Paid social geo test at current spend, six weeks plus a cool-down: incremental new customers 180 against 450 attributed by the platform in the same regions and weeks.

Answers explained

  1. Scenarios with ranges. Give outcomes for holding, cutting and raising spend, built from marginal costs that worsen as spend grows, with ranges that widen where data thins. One average-based figure overstates both the gain from a rise and the loss from a cut. See forecasting outcomes from spend scenarios.
  2. Only within the observed range. The marginal cost of each step is already rising fast and the last step is above target (worked below), so the curve is flattening inside the data. Beyond the highest spend run, the shape is an assumption: show it as a wide range, recommend a staged step and test it. See diminishing returns and saturation.
  3. Seasonality, not social. Spend was raised in the weeks when demand rose anyway, so spend and sales moved together by coincidence and the fit credits social with the season. Fit on comparable non-seasonal weeks, or control for season, and confirm with a test outside the peak. See testing the forecast against reality.
  4. As a correction factor at that spend level. Divide incremental by attributed orders and apply the factor to social's attributed orders in reports and to its curve in forecasts, labelled with the test's spend, period and regions, and retested on a schedule. See from lift to budget decisions.
  5. The spring social curve. Test the assumption carrying the most money and uncertainty. Renewal is measured directly and brand search at current spend is a smaller decision; the spring rise is large and beyond observed spend. See hypothesis.
  6. A smaller step and a retest. The decision must survive the pessimistic end of the range. Raise by less than tested, read the new level, and only then go further. Treat a marginal cost per acquisition (CPA) near the threshold as a stop sign, not a green light. See reading the results.
  7. Ranges are the evidence, and the recommendation is built to survive them. Show that the recommended budget earns a positive return on investment (ROI) even at the low end, that the risky step is staged behind a test, and which result would change the plan. Precision the data cannot support is what finance learns to distrust. See reading lift and uncertainty.
Illustrative example — not a real result

Questions two and four worked through

Search steps: £3,000 more bought 75 extra customers, £40 each, exactly on target; the next £3,000 bought 45, about £67 each. The agency's £18,000 a week sits £9,000 beyond the highest spend run, three times the size of the last observed step.

Social test: 180 ÷ 450 = 0.4, so 40% of attributed customers were caused by the spend. Applied to a month with 1,200 attributed customers, reporting shows about 480 estimated incremental customers. The Incrementality factor holds at the tested spend only. A learner with six or seven right moves on; four or five rereads the sections behind the misses.

Where to go next

If questions two, three or five caught you out, redo the spend scenarios exercise for an online optician. Otherwise move on to data quality with the campaign data audit for a coach tour operator, and keep the intermediate marketing analytics level to hand.

Pages this lesson applies

Terms used on this page

Also related

Related