Issue 78  August 13th 2026

When last-touch beats MMM

I listened to a great recent Mobile Dev Memo podcast with Michael Kaminsky, co-founder of Recast, on open-source media mix modeling (MMM), and one particularly spicy line has been rattling around in my head ever since:

"Last-touch attribution has lots of problems, but for a lot of businesses, it's way better than anything else that they have, and an MMM would actually be worse for them than just following the last-touch attribution."

Coming from someone who literally builds MMM software, that's a remarkable statement.

It feels like the industry has come full circle. For years, the conversation has assumed that attribution is broken, privacy killed measurement, and MMM is the natural evolution. But Michael's point is that for plenty of businesses, last-touch is still one of the most practical and actionable ways to make decisions. An MMM isn’t automatically better just because it's more sophisticated — without the complexity to warrant it, it can introduce more uncertainty than clarity.

At Branch, we think about this a lot, since we're often associated with last-touch and understand its limits. That's why we've invested in things like Funnel Analysis, which surfaces the touchpoints and channels that influenced a conversion before the final interaction. To me, the natural evolution of attribution is about adding context, not necessarily replacing the last-touch model.

The other part of the conversation I found fascinating: the history lesson on MMM itself. Michael’s take is that Meta and Google didn't open-source Meridian and Robyn because they wanted marketers to abandon attribution, they did it because they believed traditional MMMs had become structurally biased toward channels like TV, thanks in part to decades of agency incentives and legacy media buying. An open-source methodology let them prove they were driving more value than those models suggested.

That’s a good lens for how much the measurement conversation has shifted. CTV is the clearest example: It used to be one of the biggest reasons marketers leaned on MMM, since it behaved much more like traditional TV than digital performance marketing. Today, it's increasingly measurable and accountable as a performance channel — even as entirely new channels (AI discovery, retail media, creators) get added to the mix.

Last-touch didn’t suddenly stop working; customer journeys just got more layered, which is why I think we've moved beyond the privacy era and into the confidence era.

For years, every conversation centered on what we lost to privacy, but marketers have accepted that perfect deterministic measurement isn't coming back. The question has shifted from, "How do we know exactly what happened?" to, "Do we have enough signal to make the best decision possible?"

There's no single measurement methodology that's always right. For many marketers, last-touch enriched with assist and influence data is enough. Sometimes you need incrementality testing. Sometimes you need MMM. The future of measurement is about blending approaches to maximize signal, increase confidence, and make better decisions.

Reddit has a Google problem

Reddit reported 61% revenue growth and a 64% increase in ad revenue — so why did its stock fall the next day?

CEO Steve Huffman pointed to it: Search referrals were "choppy" during the quarter, and U.S. daily active users declined sequentially. The two are closely linked.

For years, the flow went Google → Reddit → brand site → customer. AI is compressing that journey. Google increasingly answers questions directly, so fewer users ever click through. For a platform like Reddit that depends on search referrals to bring in new logged-out users, fewer clicks from Google means fewer new users, period.

But the same shift is also an opportunity. Reddit has what AI search engines and brands both want: millions of real people talking about what they like, what they hate, and what they actually buy. Reddit is starting to turn those conversations into discovery experiences of its own.

Earlier this year, Reddit began testing AI-powered shopping inside search, generating product pages built entirely from what Redditors said, with a brand's catalog (image, price, buy link) sitting below. Brands can supply the product catalog, but they can't touch the write-up. As Reddit expert and commentator Jonny Waite put it, the comments “people write about a product this quarter become that product's page next quarter.”

That’s the broader pattern: Someone might ask ChatGPT what to buy, search Reddit, watch TikTok reviews, or have Gemini compare two products before ever visiting a brand’s site. By the time they arrive, a meaningful part of the purchase decision has already happened. So whereas SEO taught us to think about what our brand publishes, AI discovery forces us to think about what the rest of the internet says about us instead — and both of those channels are growing.

Reddit’s earnings are another signal of just how much the path to discovery is changing.