Every month Kavya re-derives the same budget maths in a spreadsheet, and every month she gets the blended cost-per-lead subtly wrong. She asks Claude for a planner. It takes 13 turns — because the rules that make ad maths true aren't the obvious ones. Then she writes them down, folds them into one prompt, and rebuilds the planner in a fresh chat in one message.
This is what she ended up with. Every rule below it — the slabs, the cap, the order of operations — had to be dragged out of her head one turn at a time. Change the inputs and watch them fire.
Click through it. Notice what each turn actually is: not Claude failing, but a rule arriving late. Every correction is a piece of context that was missing from turn 1.
At the end of the messy chat she asked Claude one question: “list every correction I gave you in this chat.” Claude has the whole conversation in front of it, so the log costs almost nothing to produce — and this log is the context she was carrying in her head all along.
Blended CPLBlended CPL = total spend ÷ total leads. It is NOT the average of the channel CPLs. A model will reach for the mean every time — say this explicitly or it will be wrong.
Minimum viable spendMeta ₹15,000 · Google ₹20,000 · LinkedIn ₹30,000.
Below the floorBelow its minimum a channel does not merely underperform — it produces nothing reliable. Zero its leads and warn. Counter-intuitive, and never inferred.
Split integrityIf the split does not total 100%, refuse to project. A warning nobody must act on is decoration.
Learning phaseLinkedIn only — first month at 60%, it needs 2 weeks to leave the learning phase. Meta and Google are fine from day one.
RoundingRound leads down. There is no 0.7 of a lead.
FloorUnder ₹50,000 total, say plainly the budget is too thin to split three ways.
TargetTake a target cost per acquisition and show whether the plan lands under or over it.
BehaviourRecalculate live on every change.
OutputOne screen, no login, prints for the client deck.
The log, folded into a single message: role, spec, the rules, the order they fire in, the edge cases, the output format. This file is the real deliverable of HO4 — not the tool. The tool is just the proof that the context is complete.
Build a single-file HTML monthly ad-budget planner for three channels: Meta, Google and LinkedIn. No login, no backend, one screen, prints cleanly. INPUTS - Total monthly budget (rupees) - A % split across the three channels - A cost-per-lead for each channel - A target cost per acquisition RULES — the non-obvious ones matter most 1. BLENDED CPL = total spend / total leads. It is NOT the average of the three channel CPLs. Do not take the arithmetic mean. This is the single most common error in ad maths and it looks correct, which is why it survives. 2. MINIMUM VIABLE SPEND per channel: Meta Rs 15,000, Google Rs 20,000, LinkedIn Rs 30,000. Below its minimum, a channel does not just underperform — it produces NOTHING reliable. Zero out its projected leads and warn loudly. (Yes, this is counter-intuitive. It is still the rule.) 3. If the % split does not add up to exactly 100, REFUSE to project anything. Don't warn and then calculate anyway. 4. LEARNING PHASE — LinkedIn ONLY: show its first-month leads at 60%, because it needs about two weeks to exit the learning phase. Meta and Google are at full rate from day one. Do not apply this to the other channels. 5. Round projected leads DOWN. There is no 0.7 of a lead. 6. If the total budget is under Rs 50,000, say plainly that it is too thin to split across three channels. 7. Show whether the projected cost per acquisition lands UNDER or OVER the target. Recalculate live as anything changes. Rupees, Indian digit grouping.
She opened a new chat, pasted that one prompt, and got the same tool — working, first try. Same features, none of the archaeology.
This is the whole lesson. Claude was never the bottleneck. The context was — it just arrived one turn at a time instead of all at once. Say it up front and the fourteen turns collapse into one.