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

Marketing Analytics and Reporting › Advanced

Forecasting Spend Scenarios for an Online Optician

Finance at an invented online optician wants next year's new customers forecast at three budget levels. You fit a simple curve to each channel's spend history, read marginal cost from it, build scenarios whose ranges widen where the data runs out, turn them into contribution finance can use and choose the incrementality test that would narrow the forecast most.
ExerciseAdvanced levelModule 2: Forecasting and incrementality for finance5 minUpdated

Why this matters

Finance does not want a single number from marketing; it wants to know what each budget level buys and how sure anyone can be. The marginal returns and forecasting guide explains why the next pound buys less than the average one and why scenarios beat one figure. The advanced step is fitting a curve that can be extended, being explicit about where the data stops supporting it, and spending test budget on the assumption that moves the forecast most.

The optician and its finance request

Iris and Lens is an invented online optician selling glasses and contact lenses. Finance has asked for next year's new customers and contribution at three budget levels, and will book the forecast into the plan alongside the marketing profit and loss view it already keeps.

Illustrative example — not a real result

Iris and Lens: weekly spend bands from last year

Non-promotional weeks only. First orders come from the order system, assigned to the channel of the last non-direct session.

ChannelWeekly spendWeeks at that levelAverage first orders a week
Paid search£2,0006160
Paid search£4,0009240
Paid search (current)£8,00014360
Paid social£1,000550
Paid social£2,000885
Paid social (current)£4,00012145
Online video (current)£1,000 / £2,0006 / 1020 / 32

Contribution per new customer over two years: £60. Finance's ceiling on marginal cost per first order: £45. Budget levels requested: £11,000, £14,000 and £18,000 a week (£572,000, £728,000 and £936,000 a year). Online video is a protected brand line at £2,000 a week in every scenario.

Range rule agreed with finance: ±8% on orders within the observed spend range; beyond the highest observed spend, count extra orders at between half and all of what the curve predicts.

From spend-response curves to scenarios

A power curve, orders equal to a constant times spend raised to a power below one, fits most channels with two or three points and has a useful property: marginal cost per order equals average cost divided by the power. Fit it from two bands, check it against the third, then split each budget level so the marginal cost per acquisition (CPA) is roughly equal across channels. Report every scenario as a range, and widen it wherever the split pushes a channel past the spend it has ever run at.

Task

  1. Fit a curve for each channel and check it against the band you did not use.
  2. Calculate marginal cost per first order at current spend, and estimate where each channel passes the ceiling.
  3. Split each budget level across channels and forecast weekly and annual first orders with ranges.
  4. Convert each scenario into contribution after media, with its range.
  5. Say what you would tell finance about the low and high steps.
  6. Specify the incrementality test that would most improve the forecast.

Worked answer

Illustrative example — not a real result

Curves and marginal costs

Power from two bands: ln(orders ratio) ÷ ln(spend ratio). Search: doubling spend multiplies orders by 1.5, so power = ln 1.5 ÷ ln 2 ≈ 0.585; the third band fits exactly. Social: doubling multiplies orders by about 1.7, power ≈ 0.77. Video: two bands only, power ≈ 0.68.

Marginal cost at current spend = average ÷ power. Search £8,000 ÷ 360 = £22.22 ÷ 0.585 ≈ £38. Social £4,000 ÷ 145 = £27.59 ÷ 0.77 ≈ £36. Video £2,000 ÷ 32 = £62.50 ÷ 0.68 ≈ £92 on first orders.

On the curves, search passes £45 at about £12,000 a week and social at about £10,500, both far beyond any spend observed.

Illustrative example — not a real result

Scenarios for finance

ScenarioWeekly split: search / social / videoWeekly first ordersAnnual first ordersContribution after media
Low£6,000 / £3,000 / £2,000452 (416 to 488)23,504 (21,632 to 25,376)£838,000 (£726,000 to £951,000)
Central£8,000 / £4,000 / £2,000537 (494 to 580)27,924 (25,688 to 30,160)£947,000 (£813,000 to £1,082,000)
High£9,250 / £6,750 / £2,000641 (546 to 684)33,332 (28,392 to 35,568)£1,064,000 (£768,000 to £1,198,000)

Low: marginal cost about £34 in both search and social. High: about £40, but 104 of the 641 weekly orders come from spend beyond the data (32 from search, 72 from social), so the low end counts them at half.

Step from central to low: saves £3,000 a week and loses 85 orders, about £35 per order lost. Step from central to high: £4,000 a week buys 104 orders at the central estimate (£38 each) and 52 at the low end (£77 each, above the ceiling).

What to tell finance. The cut gives up orders that cost less than the ceiling, so it lowers contribution unless cash is the binding constraint. The rise is profitable at the central estimate and loses money at the pessimistic end, because its extra orders all sit beyond observed spend. Recommend the central budget with a staged, tested rise rather than the full high scenario. Present return on investment (ROI) on contribution for each scenario, and keep video out of the first-order arithmetic: its effect builds over longer periods and needs its own evidence.

The test. Social carries most of the beyond-data orders and has the widest range, and its attributed orders have never been checked for Incrementality. Run the budget increase incrementality test: raise social to the high scenario's level in matched test regions, leave control regions alone, run at least one buying cycle plus a cool-down, and judge marginal cost per incremental first order against the ceiling, counting all-channel first orders from the order system. Guardrails: blended cost per first order and branded search. The result replaces the half-to-all assumption with a measured point, and the ratio of incremental to attributed orders adjusts the social curve, as the incrementality testing guide describes.

Common mistakes

  • Extending the curve without widening the range. Beyond observed spend the shape is an assumption.
  • Judging video on first orders. A brand line needs a longer evidence standard.
  • Fitting curves on promotional weeks. Spend and orders both rise for reasons unrelated to each other.
  • Testing the channel that is easiest. Test the assumption that moves the forecast most.

Check your judgement in the forecasting and incrementality checkpoint, or revisit finance reporting at the intermediate marketing analytics level.

Pages this lesson applies

Terms used on this page

Also related

Related