"A theorem, not a proof." Dan Gilbert on testing your way to growth.

This issue first went out to subscribers of The Ad Podcast newsletter. Get the next one in your inbox, free: adpodcast.beehiiv.com

In 2011, Dan Gilbert joined Google as a graduate. He admits he didn't know at the time that the company even employed people. To him it was the box on the internet you typed things into.

Working with Google's advertisers, he noticed something they didn't: this new kind of advertising was as much a math and engineering problem as a marketing one. You could bid across millions of keywords, split-test everything, and read the data in completely new ways. His clients would nod politely and carry on. So he left and started Brainlabs (the name says it: smart people experimenting in a lab), hiring engineers, mathematicians and scientists and teaching them marketing.

Thirteen years later, Brainlabs is around 1,000 people, still independent, and the US is now more than half its business. Dan moved here to run it. I caught him over video from the NAMM Show floor, and he did not hold back.

The big idea: every model is a theory, so test it

Dan's sharpest point was about attribution. Every attribution model or marketing mix model is a theory, not a proof. You can't know it's right just by looking backward. So Brainlabs runs one model against another in the real world, for example with geo-split tests, and keeps the one that actually drives more revenue. Sometimes they don't fully know why the winning mix wins. They figure that out after.

He compared it to bridges. A brilliant engineer didn't sit down and solve bridge-building. People built a lot of bridges, some fell down, and they studied why the good ones stayed up.

Here's the sell-side version: your playbook is a theory too. The talk track, the cadence, the deck you swear by. Most sales teams never test them. They just repeat what the top rep did three years ago.

3 things to do about it

1. Run two versions, not one guess

Dan's old paid search rule: you'd never tell a client which ad would win. You'd run "free delivery" against "fast shipping" and let the results surprise you. One of those tests taught him buyers cared more about reliability than about free.

This week: pick one part of your outreach (a subject line, an opening question, a call to action) and send two versions to similar prospects. Change one variable only. Keep the winner, then test the next thing.

2. Make sure your incentives match your client's

Dan built Brainlabs around one question: given $100, how do we maximize revenue for this client? He argues you can't honestly answer that if your business makes money by marking up the media you're recommending. He was candid that many clients know about those conflicts and accept them as a kind of tax, often because procurement wants to show "no fees." His conclusion: those aren't Brainlabs' clients, and that's fine.

Sellers face the same test. If your comp plan rewards you for the bigger package rather than the right one, the buyer can feel it.

Next proposal: say out loud how you make money on the deal and why the recommendation is still right for them. Transparency closes faster than you'd think.

3. Kill the brand-versus-performance silo

Dan called brand versus performance an industry invention. Both drive revenue; they just pay back over different time windows. Splitting teams into "brand" and "performance" creates a fake silo. Look at short-term and long-term returns together.

The same goes for marketing and sales. Pipeline doesn't care which team "owns" it.

This quarter: sit down with your marketing counterpart and agree on one shared number and two time windows (this quarter and next year). Report on both together.

Bonus: the chat is the new homepage

Brainlabs' fastest-growing line is helping brands show up in AI chat answers. It has tripled in two years, and Dan expects it to be more than half the business within a couple of years. His thesis is that the chat window becomes the front door to the internet, the way Google's first page made companies like Expedia and Zillow. They also run hundreds of AI agents doing junior-level work like research, basic optimization and scheduling. Because they're growing, the junior staff haven't been cut. They now coordinate the agents.

His warning to marketers: if you've never set up an agent, you're running out of time to learn. And the next level is "context engineering," giving the AI your meetings, your notes and your best practices so it gives you your answer, not the average one.

Best line of the episode

"A theorem is a theorem, not a proof." (Dan Gilbert, co-founder and CEO of Brainlabs)

Watch the full conversation

We also get into why measurable doesn't just mean Google and Meta, how wearables and your inbox could decide which AI you use, and Dan's very candid take on ads coming to ChatGPT.

Watch the full episode on Spotify. Also on Apple Podcasts and YouTube.

Want more from Dan? His first book, Don't Be A Midwit, is out in early 2027. Pre-order it here.

Want the next one? Every issue takes one conversation from The Ad Podcast and turns it into ideas you can use. Subscribe free at adpodcast.beehiiv.com.

Talk soon,Dylan