Notes from the lab · What it costs

What an FMCG campaign actually costs to run on AI, line by line

July 2026 · 6 min read · Wynngrid

Every deck about AI production says "10x cheaper" and then shows no numbers. This note shows the numbers. It is built from a real campaign shape we run for FMCG accounts: 1 hero film of about 45 seconds, 12 performance cutdowns, 30 statics, and variants in 4 languages. Roughly 60 finished assets.

Two things before the table. First, these are ranges, because retry rates move with how tight the brand system is. Second, we count everything, including the generations nobody ever sees. Most published numbers quietly leave those out.

The cost sheet

Line itemWhat it coversRange
Model inference, videoHero film and cutdowns, including every rejected take₹30,000 to ₹55,000
Model inference, imageStatics, key visuals, resizes, including retries₹6,000 to ₹12,000
Voice and musicLicensed VO in 4 languages, track licensing₹10,000 to ₹18,000
Creative directionHuman hours: brief, look development, edit decisions50 to 70 hours
QC and reviewHuman pass on every asset before the client sees it15 to 25 hours
ComplianceDisclosure labels, records of what was generated and how2 to 4 hours
Infra and deliveryHosting, versioning, handoffUnder ₹5,000

The pattern worth noticing: the machine lines are small and falling. The human lines are the real budget, and they do not go to zero. They move up the stack, from executing shots to directing a system.

The number everyone gets wrong

Generation is cheap. Keepers are not. On hero-film frames we see roughly 6 generations for every 1 that survives review. On statics it is closer to 2 to 1. So when someone quotes you a per-generation price, multiply it by the retry ratio before you compare it to anything.

This is also why "cost per asset" quotes vary so wildly between vendors. The honest metric is cost per approved asset. Everything in the table above is counted against approved output, waste included.

Against the traditional sheet

The same campaign shot conventionally carries a different set of lines: a shoot day or two, location, talent and usage fees, a post house, and 6 to 10 weeks of calendar. For the accounts we have moved over, the production line lands at roughly a third of what it was, and the calendar compresses from weeks to days.

One worked example we can publish: for Neeman's we delivered 12 films in 17 days at about a third of the previous production cost. The account's ROAS on that work ran 21.07x against a 3.4x benchmark. The production saving is real, but the speed is what changed how the brand plans campaigns.

The saving is not the headline. The headline is that campaign 12 costs less than campaign 1, because the system trains on your brand every time it ships. A shoot never gets cheaper the second time.

What this means for a brand your size

Want this sheet for your campaign shape?

Thirty minutes with both founders. You keep the map either way.

Book a discovery call