Episode 183: Ask Olivia Kory Whether AppLovin Ads Work & She Explains Attribution vs Incrementality

Full analysis

Olivia Kory, Chief Marketing Strategy Officer at Haus, came back on Marketecture and said the quiet part out loud: AppLovin's e-commerce ads work, but only for a specific kind of advertiser, and the reason they "work" is speed, not magic. That distinction, plus her broader point that whole channels are leaving money on the table because they refuse to be measured, is the real content here. For any operator selling media or measurement, this is a map of where advertiser dollars are stuck and why.

What's being decided: nothing binary. This is a briefing on a shift in how sophisticated advertisers separate "ads that correlate with sales" from "ads that cause sales," and which channels win or lose as that lens spreads. Type 2, easily reversible for any single advertiser. But the direction of travel matters for whoever sells affiliate, retail media, and podcast inventory. Forcing function: budget-allocation season and the AI-agent tools now recommending where dollars move.

The Market Analyst. Kory's cleanest signal is a commercial one for the channels that won't open up. She groups affiliate, Amazon search, and retail media networks together as places where advertisers are "holding back dollars" because nobody can prove lift. For an informed outsider: retailers sell ads against their own shoppers and mostly grade their own homework. AppLovin did the opposite. It let third-party incrementality testers in from the start of its e-commerce pilot in late 2024, and that openness is a big reason it earned trust fast. The read for retail media operators: the ones who let independent holdout tests run will pull budget away from the ones who stall. Ari Paparo's Google numbers are a side dish, and the summary itself flags the ~82% cloud growth figure as unverified. Don't repeat it.

The Skeptic. The load-bearing assumption is that AppLovin "works," and Kory is careful, but the framing still needs a squeeze. She confirms it works for low average order value, impulse categories like apparel and footwear, driven by fast short-term effects. That is exactly the pattern you'd expect from a model optimized to harvest people already close to buying. Short-term lift is real lift, but it is also the easiest kind to over-credit. And the affiliate takeaway rests on a "handful of tests" with a small sample she flags herself. One more thing: Haus sells incrementality testing. The claim that affiliate and retail media are suppressing their own advertiser market by not offering it is plausible and also precisely the argument that grows Haus's business. True and self-serving can both hold. Weight it accordingly.

The Operator. Try to act on this Tuesday morning and you hit the wall Kory describes. She won't recommend a CTV test under eight weeks, but AppLovin shows lift in days. So a growth marketer staring at a dashboard rewards the fast channel and starves the slow one, even when the slow one is building demand that shows up later. That is a measurement artifact driving budget, not performance. The CAPI point is the practical trap for platform teams: conversion APIs, the server-side pipes that send purchase data back to ad platforms, are being sold as measurement when they are really optimization fuel. Feed them your conversions and you make the platform's model smarter at taking credit. Kory's tell is that some advertisers now send only predicted-incremental conversions through CAPI. Smart, and almost nobody is set up to do it.

The Customer / End User. The advertiser here is genuinely stuck, and the podcast segment is the sharpest illustration. Paparo argued the move to video has broken podcast measurement outright: Spotify, Apple, and YouTube each use different video tech, none supports video ad insertion well, and you can't run a geo-holdout on a host-read spot. Haus's best workaround is a crude before-and-after forecast. In plain terms: buyers pouring money into video podcasts are flying blind, and they should stop treating podcast like streaming audio, which Paparo says is meaningfully more measurable. The customer isn't asking for more channels. They're asking for proof, and most sellers still can't hand it over.

The CFO. Follow the dollars and the story is about where budget flows when proof is unequal. If affiliate and retail media can't demonstrate causal lift, a disciplined finance team caps spend there and routes it to channels that can. That's the mechanism behind Kory's "limiting their advertising TAM" line, and it's a warning to any retail media network booking growth on last-click credit. The other CFO item is Google's capex: $195 to $205 billion guided for the year, with the quarter's first negative free cash flow in a long time and a stock that took a hit. For operators, that's the tell that even Google is spending faster than the ad business alone can comfortably fund. The cloud bill is now shaping the ad company's discipline.

Where the council splits. The Operator and the Market Analyst agree AppLovin's openness is the winning move; the Skeptic says a low-AOV impulse engine flatters itself on short-term lift and we shouldn't generalize it into "AppLovin beats everyone." Second fault line: Kory says the unmeasured channels are throttling their own budgets, while the Customer view is that advertisers keep paying into unmeasured podcast video anyway. Skepticism and spend can coexist for a while. Third: is CAPI a measurement tool or an optimization tool? Kory says optimization, and most of the market still sells it as measurement.

What it hinges on. Whether independent incrementality testing becomes table stakes for a channel to keep growing its budget. AppLovin bet yes and got trust. Retail media, affiliate, and Amazon search are betting they can grow on their own attribution a while longer. The council leans toward the openness thesis, with the Skeptic's caveat that short-term-effect channels will always look better under fast-read tests than demand-building ones. Before anyone reallocates on an agent's say-so, verify the holdout design and check whether the "lift" is just harvested intent.

Prediction: By the Q1 2027 earnings and budget-planning cycle, at least one major retail media network will publicly announce support for third-party incrementality or independent holdout testing, following the openness playbook AppLovin used to win e-commerce trust.

Confidence: Medium — competitive pressure and advertiser skepticism both push the same way.

Why: Kory laid out the mechanism directly: advertisers are holding budget back from retail media, affiliate, and Amazon search because those channels grade their own homework, and AppLovin won trust fast by letting outside testers in from day one. Retail media networks are chasing growth into a maturing category and can't keep booking on last-click credit while a rival dangles independent proof. The opposite outcome, everyone staying closed, gets harder each quarter that a competitor demonstrates lift a CFO can defend. The likeliest holdouts are the ones with the weakest real incrementality, which is exactly why at least one confident player breaks ranks first.

Revisit by 2027-04-30: We're right if a top retail media network (Walmart Connect, Roundel, Kroger Precision Marketing, or Amazon Ads) publicly supports third-party incrementality or holdout testing by then. We're wrong if the major RMNs all still restrict measurement to their own attribution.