What scaling teams get wrong about automation

The former Tinder CPO on the unglamorous groundwork everyone wants to skip.

Everyone wants to talk about the model. Almost nobody wants to talk about the data underneath it. We sat down with Ravi Mehta, former Chief Product Officer at Tinder, about the gap between ‘we have tons of data’ and ‘we have data we can actually use,’ and why that gap is where most AI ambitions quietly stall.

‘We have data’ isn’t the same as ‘we’re ready’

Ravi’s point landed fast: most companies have plenty of data and almost no agreement on what any of it means. Two teams define a ‘customer’ three different ways. The dates are in four formats. The thing you most want to predict was never recorded consistently. None of that is a modelling problem. It’s a housekeeping problem wearing a lab coat.

Do the boring audit first

Start with one question, not one dataset. Ravi’s rule of thumb: pick the single decision you’d most like to improve, then trace backwards to the handful of fields that actually feed it. You usually discover you need far less data than you feared, and that the little you need is messier than you hoped.

Fix it once, properly. Cleaning data isn’t glamorous and it never demos well. But it’s the difference between a model that holds up in production and one that looks brilliant in a slide and falls over the first real week.

The takeaway

If a vendor promises results without asking hard questions about your data, that’s the tell. The unglamorous groundwork is the work. Skip it and you’re not buying AI. You’re buying a very expensive guess.