The Next-Gen Brand Lab
Budgets and unit economics
How long should I run a campaign before deciding it does not work?
Last updated 2026-09-09
A full test window — roughly one month of controlled spend, and never less than the time it takes the platform to exit the learning phase. On Meta that means about 50 optimisation events per ad set per week, so at a ₹400 cost per lead you need roughly ₹20,000 per ad set per week before the delivery is even stable. Judging a campaign before that point is judging noise, and the most expensive mistake in this category is killing something that was working during its learning phase.
Why early judgement is so costly
Restarting resets the learning. A campaign killed at day 10 and relaunched at day 20 has paid for the unstable phase twice and has no more information than it did at the start.
Compounding this, performance during learning is genuinely worse and genuinely unrepresentative, so the data most likely to make you kill a campaign is the data least worth acting on.
What you may act on early
Creative-level decisions, not campaign-level ones. If one concept out of five is clearly not producing after enough impressions, turning it off is a within-test decision and does not reset anything.
Structural changes — objective, audience, budget distribution — restart learning. Make them deliberately and rarely, and not on a Tuesday because the weekend was quiet.
Want this answered for your project specifically?
A Phase 0 call is a questionnaire, not a pitch. You get the market read before anyone spends a rupee on advertising.
Book a Phase 0 call ↗