Shift 30% of a year's testing volume from panel and eye-tracking studies to Dragonfly AI, and on a typical enterprise contract that's £295,800 back in the marketing budget – 1.6 times the value of the contract itself. Shift a more modest 10%, and it's still £98,600. These aren't hypothetical numbers.
They're what we found when we modelled it against our own clients' actual testing volumes.
The reason the number is that large is the real hidden cost in this conversation. Most CPG testing budgets are concentrated on a small number of full-service panel and eye-tracking studies a year, while everything else – the long tail of packaging variants, social ads, e-commerce listings – ships on instinct. The budget isn't missing. It's just locked into a testing model built for a handful of assets a year, not the volume a modern portfolio actually produces.
We don't think Dragonfly AI should replace panel testing or eye-tracking outright, and most of our customers don't use it that way. The useful question isn't 'panel or Dragonfly AI' – it's what a modest shift from one to the other is actually worth. That's an efficiency argument, not an effectiveness one: it doesn't ask anyone to trust a new methodology before they've seen a result, only to look at what reallocating existing spend returns.
Most of that saving sits in one place. At a 10% shift, £72,800 of the £98,600 total comes from in-store and shelf testing alone – because eye-tracking is the most expensive method most brands still rely on – against £20,900 from advertising and £4,900 from e-commerce. The pattern holds at every level: shelf is where the biggest, slowest, costliest studies have always lived, so it's exactly where shifting even a small share of volume unlocks the most.
At the volumes our largest clients test at, the same logic scales further still – some see annual research costs fall by closer to 80%. The exact proportions differ by client size and testing mix, but the direction doesn't: the more volume that moves, the more budget frees up.
These are modelled estimates against list pricing, not a guarantee for every contract – actual savings depend on which studies move and how complex they are. But it's why the question we're asked most often now isn't whether Dragonfly AI is accurate. It's how many months before it pays for itself. On a 20% shift, that's £197,200 back in the same year the change is made.
None of this requires giving up the testing a team already trusts, or the expertise built around it. It requires treating the testing budget itself as something worth optimising, the same way marketing teams already optimise media spend. The hidden cost was never just the creative nobody tested. It's the budget still being spent as if the only responsible way to test anything properly was to test almost nothing at all.