Why this matters
Budget usually flows towards the channel with the best average cost, because that is what reports show. The marginal returns and forecasting guide explains why the next share of spend almost always buys less than the average suggests, and the cross-channel performance planning guide explains why each channel needs a role and a measure that fits it. This exercise asks you to use both on real-looking data where the average and the margin point in opposite directions.
The school and its channels
Fennwick Language School is an invented school selling evening classes and online courses. It runs paid search for people already looking for a course, paid social to reach adults who might start one, and online video in its home region. Over the past year it has changed each channel's budget at least once in comparable seasons, which gives a rough curve for each.
Fennwick Language School: economics
- Average course fee: £420. Contribution margin after tutor, room and platform costs: 30%, so £126 per enrolment.
- The director wants £36 of profit per enrolment, so the target cost per enrolment is £126 − £36 = £90.
- Next quarter's paid media budget: £22,000, the same as this quarter.
- Enrolments are counted in the school's booking system, not taken from platform reports.
Fennwick Language School: spend and enrolments by quarter, similar seasons
| Channel | Spend level | Enrolments | Average cost per enrolment |
|---|---|---|---|
| Search | £7,500 | 162 | £46 |
| Search | £9,000 | 180 | £50 |
| Search | £12,000 (current) | 204 | £59 |
| Social | £4,000 | 56 | £71 |
| Social | £6,000 (current) | 80 | £75 |
| Video | £2,000 | 15 | £133 |
| Video | £4,000 (current) | 28 | £143 |
Notes from the team:
- Search loses almost no impression share to budget at the current level; the extra spend went to broader terms.
- Social has never run above £6,000 a quarter; frequency on its core audience is still moderate.
- Video enrolments are counted on last click plus a "where did you hear about us" question. The school has never tested whether video lifts search enquiries.
Average versus marginal returns
On averages, search looks cheapest and video looks like waste, so the obvious move is to cut video and add to search. The marginal view asks a different question: what did the last step of spend in each channel buy? Divide the change in spend by the change in enrolments between two levels, and compare that figure with the margin-based target rather than with the channel's own average. A channel can have the best average CPACost of each conversion. in the plan and still be wasting its last share.
Video needs a different lens again. Its role is to create demand that other channels capture, so judging it on last-click cost alone undervalues it, and only an IncrementalityConversions caused by marketing. test can say how much it adds.
Task
- Calculate the marginal cost per enrolment for each step in each channel.
- Say which channel's last share of spend is above target and which is below.
- Propose next quarter's split with the total held at the same budget, staging any increase beyond tested levels.
- Forecast enrolments and contribution for your split, stating the marginal cost you assumed for untested steps.
- Decide how video should be judged next quarter.
- Write the conditions under which you would reverse your proposal.
Worked answer
Fennwick's marginal costs and proposed split
Marginal cost per enrolment:
- Search, £7,500 to £9,000: £1,500 ÷ 18 = £83. Search, £9,000 to £12,000: £3,000 ÷ 24 = £125.
- Social, £4,000 to £6,000: £2,000 ÷ 24 = £83.
- Video, £2,000 to £4,000: £2,000 ÷ 13 ≈ £154 on last-click counting.
Contribution on search's last £3,000: 24 × £126 = £3,024, less £3,000, so about £24. It barely pays for itself.
Proposed split: search £9,000, social £9,000 (raised in two steps of £1,500), video £4,000. Total £22,000.
Forecast, assuming social's next steps cost about £95 per enrolment, rising from the last tested £83:
- Search: 180 enrolments. Social: 80 + (£3,000 ÷ £95 ≈ 32) = 112. Video: 28. Total about 320, against 312 now.
- Contribution moved: social's extra £3,000 buys about 32 × £126 = £4,032 of margin, around £1,000 more than search's last £3,000 produced.
Search keeps its best share and gives up the step that bought broader terms at a marginal cost well above target. Because it was not losing impression share to budget, the cut removes the weakest queries rather than unfunded demand.
Social gets the money, but in stages. Its last tested step came in under target, but nothing is known above the current level, so each step runs for a few weeks and the marginal cost is read before the next. The forecast is a range built on a stated assumption, not a promise.
Video is held, not cut. Its last-click cost is the worst in the plan, but its role is demand creation, and cutting it may show up later as fewer search enquiries. Run a regional holdout: switch video off in some matched areas and compare search and total enrolments with areas where it runs. That costs no extra budget and turns an argument into evidence about its true cost per new student, the CACCost to acquire one new customer. that matters.
Reverse the move if:
- social's marginal cost after the first step exceeds the target, in which case hold the second step;
- search enquiries fall in the weeks after the cut by more than the lost broad terms explain;
- the video holdout shows search enrolments drop where video is off, which would make video cheaper than it looks.
Record the split and the triggers on the media plan template, with the role and measure for each channel from the media planning framework.
Common mistakes
- Moving budget on averages. Search's average is the lowest in the plan, yet its last step is the most expensive.
- Comparing marginal cost with the channel's own average. The comparison is always with the margin-based target.
- Extrapolating social beyond tested levels. Untested steps need an assumption, a stage and a check.
- Cutting video on last-click cost. A demand creation channel is judged on what it adds, not on what it is credited with.
- No reversal conditions. A proposal without them gets defended rather than checked.
Next, test yourself with the cross-channel planning checkpoint, or return to the intermediate performance marketing level.
