The number that matters: spend per day

The month is £69,650, £2,322/day average, forecast landing £215,657 = 83% of the £260,511 goal, banking £23,300 of September profit. Four budget changes, all on a Thursday. Days 1 and 2 are actuals (£1,382 and £1,429 against a £1,750 plan, because the budget was never raised on the 1st); the ladder below spends the remaining £66,839 over the 28 days left.

Set the Meta budget to SET figure, not the total (Meta under-delivers its configured budget ~6%; Google floats at ~£310/day and is auction-limited, a forecast not a decision). The first move is a 52% jump from the £1,110/day the account is on today, which will reset learning: do it now, on the cheapest days of the month, rather than mid-launch. Every later step is 22% or less, inside Meta's reset threshold, and each rung runs a full 7 days so it produces one clean 7-day CAC read before the next step. Step only if trailing 7-day CAC is under £50. Hold the current rung otherwise; the last rung forecasts £47.67, so there is under £2 of headroom by month end.

Where the money goes, which matters more than the shape

Last September the incremental budget was split between a new dedicated Autumn campaign (2.71x Meta ROAS, £27 CPA) and the ASC (1.89x, £45 CPA, down from 2.16x the week before). The Autumn campaign was the best thing in the account and was capped at £137/day. This year the same trap in a different shape: 15 of the 30 autumn ads sit inside the Evergreen CBO next to proven Tomato Vine winners where the algorithm starves them (Ads 843, 844, 846, 852 and 834 took £5 to £53 each across 1 to 2 Sept), and the dedicated Testing Autumn campaign is on £50/day.

£400/day clears roughly 70 conversions a week at the £40 CPA the autumn creative is currently running, enough for the ad set to exit learning before the 12 Sept drop rather than during it. Do not assume last year's 2.71x holds at scale - a £137/day campaign harvests the best inventory, and the early read on 1 to 2 Sept is autumn creative at 1.80x / £40 CPA against 2.64x / £28 for everything else. Ring-fence a measured budget, step it with the ladder, and cut it back if the gap is still there on the 10th.

The month day by day

Spend runs the ladder; revenue is shaped by her email calendar and the drop, which the ads do not drive. Both lines are shown together so the days where they deliberately diverge are visible.

What last September actually taught us

Last September ran £29,782 (£993/day, 86% Meta) and did ramp: up 43% from the first third of the month to the last. It ramped into the wrong thing.

The Autumn Pott drop lifted the list, not acquisition. Spend went up 26% into the drop week and new-customer revenue moved 13%, while returning and subscription revenue moved 71%. CAC went £42 to £49 and Meta CPM £15.28 to £17.83 to buy it.

Why the ramp is not shaped to a demand curve

The obvious next move is to point spend at the days that convert best. The data will not support it. Measuring demand strength as k in NC/day = k x spend^0.824 (which strips out the spend level, so windows at different budgets compare), a five-day window carries a standard error of ±0.70 in 2025 and ±1.39 in 2024, against a full between-window spread of only 0.86. The two years also disagree on where the good windows are, and their autumn drops fell on different dates (18 Sept 2024, 23 Sept 2025, 12 Sept this year), so neither a calendar nor a drop-relative profile replicates.

And it would not be worth much if it did. Holding the response curve constant, the entire spread between the best and worst possible September ramp shape is £440 of new-customer revenue on a £69,650 budget - 0.4%. That is why the ladder above is built for risk management and Q4 readiness (clean CAC reads, learning-safe steps, autumn creative out of learning before the drop) rather than for revenue optimisation, and why no step lands on 12 Sept.

The shape lesson is November, not September

Where ramp shape was worth real money last year, it went the wrong way. The heaviest week of the year ran at the worst CAC of the year, and then spend was cut by 31% just as demand doubled.

Reshaping that month on the same budget - first 20 days down to £2,300/day, the money into the 21-27 Nov run-up - models +£20,431 of new-customer revenue (+10.4%), about £11,543 of extra contribution. Two honest caveats: Only 7 days above £4,000/day and 4 above £5,000/day were ever actually run, so the target level is an extrapolation; and some of the BF-week demand is harvest of the earlier prospecting. This is the number to carry into the Q4 plan, not into September.

Version 2: the same ramp built to hit her forecast

Everything above spends the signed-off £69,650. This is the same machine with the budget solved from her number instead. Her September tab was revised down to £257,511 (from £260,511) after she saw the first two days: day 1 £7,000 to £5,000, day 2 £5,000 to £4,000. Her email calendar has also moved since the 31 Aug freeze - there is now no 3 Sept send and a Cook back-to-school email on the 24th. Her Ads column is still empty, so nothing in her sheet says what it costs to buy the revenue in it.

⚠️ The two spend figures are the same question answered by two curves we have both been using. This page's curve (k 6.21, anchored on the approved ramp) prices her target at £103,268; the Sept-to-Dec breakeven model's curve (k 5.41, which carries last September's seasonal ratio) prices it at £121,439. Treat £103,268 to £121,439 as the honest range and let the tracker adjudicate at month close. Note this also corrects the October-to-December table below on one point: on this page's curve her September target is reachable in profit (£13,332), where the colder curve said it was not.

What chasing her number would actually mean

Same discipline as the approved ladder: one deliberate learning reset on 3 Sept, then weekly steps capped at 20%. The arithmetic still forces an opening jump of 134% and a closing level of £4,531/day.

The off-peak column is the problem. The account has never run a single off-peak day at any of these levels. Its highest non-peak day ever is £2,454, on 9 Aug 2026; every day above that sits inside 15 Oct to 24 Dec 2025, when Christmas demand was carrying it. Her ramp would open on 3 Sept above that all-time high and finish 1.8x it. The response curve says the money still works; it has simply never been tested there in this season.

⚠️ And the ramp trips its own guardrail. The month averages £49.52 CAC, which reads as just inside the £50 stop line, but the average hides the shape: the last two rungs forecast £50.21 and £51.73. On its own rules this plan stops stepping on 17 Sept and never reaches the closing budget that makes her number work. That is the cleanest argument against committing to it: it is not that the spend is unaffordable, it is that our own stop rule halts it two thirds of the way in.

Her forecast, day by day

The same table as the approved plan, rebuilt on her £257,511: what to spend each day, the CAC that spend is forecast to buy, and how each day lands against her own daily target. Days 1 and 2 are actuals. Her calendar is her revised one, so the 3 Sept email is gone and Cook sits on the 24th.

The gap column reads her day shape, not the spend. Fully funded at £103,268 the month still runs £4,156 behind her over days 3 to 9 (-7.7%), then hands it back over days 24 to 30 (6.4% ahead). That is structural: in the first week the ramp cannot be at full height yet, whatever the budget, and the subscription ledger is at its lightest. So her early-September days are the ones to expect misses on even if the whole £103,268 is approved, and the recovery only arrives once the last rung is running.

The actual choice

Four points on the same curve. Every one is profitable in-month; the trade is September cash against customers bought for Q4.

The recommendation is to run the approved £69,650 ladder and re-decide on the 17th, not to commit to her number now. Going from £69,650 to £103,268 buys the last 16% of her target for £33,618 of extra spend and £9,968 of September profit - a real trade, and defensible for the extra £59,250 of 36-month customer value it buys, but not one to make blind on day 3 with zero off-peak evidence above £2,454/day. The middle option is the real one: the £84,998 stretch lands 91% of her target, still clears £18,932, and needs a £3,690/day close rather than £4,531. If the 3 to 16 Sept rungs hold CAC under £50, step onto that path on the 17th; if they do not, the approved plan was already the right size.

Where breakeven actually is in September

Three different breakevens, used for three different jobs. All include the audited constants: 56.5% contribution margin, £28,896/month overhead (verified against the KC dashboard 1 Sept), and £105,431 of September subscription + returning revenue that ads never touch.

September is profitable at a 2.50x blended, not 5x. The 5x figure divides the revenue goal by spend, but £105,431 of that revenue arrives with zero ad spend against it. As spend scales, blended ROAS falls by arithmetic while profit can still rise; that is why the guardrail is CAC, never blended ROAS.

Forecast vs what the account actually does

September forecast at the ramp, against the two most honest reference points: August 2026 (full closed month, Summer Sale included, so CAC runs flattered) and September 2025 (same season, last year's account).

How to read the month while it runs

Each pair separates a different problem, which is why both CAC and ROAS are tracked:

• CAC on target but NC ROAS below forecast → we are buying customers at the right price but they are spending less per order. AOV problem: check mix (Starter Pack vs singles), the offer, and the landing pages, not the ads.
• NC ROAS on target but blended below forecast → acquisition is fine; the subscription + returning base is soft. Check the Recharge ledger and the email calendar, not the ad account.
• CAC above £50 for 7 days → hold the ramp at the current rung. Above £74 the whole month goes loss-making at this spend.
• Everything on target → step the ramp on schedule. Every rung is pre-priced above.

October to December

Same machine, seasonalized: each month's response curve carries last year's seasonal shape anchored to the closed August actual, and each month's breakeven falls as the subscription + returning base grows against flat £28,896 overhead. The first two breakeven lines above (first-order £41.21 CPA / 1.77x, marginal £33.95 / 2.15x) hold all year; only the whole-month line moves.

Planned spend is the client sheet (target ÷ 5x): October £97,920, November £178,920, December £132,552 for the month. Every month of it is comfortably profitable (forecast profit £67,647 Oct, £124,237 Nov, £111,414 Dec), but on the measured curve it lands 66 to 73% of the targets, because the account delivers ~3.3 to 3.6x blended at those spends, not 5x. Landing the targets in full prices at 2 to 2.5x the sheet (October £220,835, November £420,824, December £269,665), spend far beyond anything the account has ever run (last November peaked at £3,427/day sustained). The honest plan: run the sheet as the floor, re-anchor this curve at each month close, and step toward the target-spend numbers only as the CAC guardrail keeps clearing.

Q4 efficiency genuinely improves (last year's Q4 NC ROAS ran 1.77 to 1.97 at 2 to 3.5x September's spend), which is why the curve supports much bigger budgets in Nov and Dec. The stop-CAC can loosen from £50 to the mid-£50s in Nov and Dec as the subsidy pool grows (£79,823 Oct → £134,004 Nov); it never loosens to the whole-month breakeven, because that line assumes the base shows up in full.