Can We Still Trust Attribution? Brian Quinn on AppsFlyer, Privacy, and Measurement

Full analysis

Brian Quinn's argument, stripped of the vendor gloss, is a real structural claim: attribution didn't die when Apple choked off the mobile device ID — it fragmented into a three-legged stool of real-time attribution, incrementality testing (holding some users back to measure true lift), and marketing mix modeling (statistical spend allocation). The tell worth chewing on is that walled gardens like Snap now ingest an outside vendor's attribution signal to feed their own bidding. That's a closed platform admitting its own grading isn't trusted. The question for operators: is independent measurement getting stronger as a moat, or is this the last good year before agentic AI and clean-room consolidation swallow the category?

The Market Analyst — Follow the ownership map, because that's where the money moved. Publicis bought LiveRamp's Habu; WPP bought InfoSum. The two biggest agency holding companies now own the plumbing that matches and measures data — while also spending client money on media. That's a conflict hiding in plain sight, and Quinn said the quiet part out loud. For an informed outsider: your scorekeeper now works for the team placing the bets. The winners here are the independent measurement players — AppsFlyer, and by extension anyone who can credibly say "we don't buy the media we grade." Snap feeding third-party signal into its own engine is the strongest bull case for independence I've seen articulated this year.

The Skeptic — The load-bearing assumption is that "triangulation" is a maturation rather than a retreat. Let's be honest about what happened: signal dropped from ~80% to ~25%, and the industry rebranded the workaround as sophistication. Incrementality plus mix-modeling plus probabilistic attribution isn't a better answer — it's three fuzzy answers you average and hope cancel out. And the Snap example proves less than claimed. A walled garden ingesting outside signal could just as easily be regulatory cover ("look, we use independent measurement") as genuine reliance. Quinn sells the guardrail; of course he thinks the car needs one. Ask whether any advertiser has actually reallocated real budget on triangulated numbers, or whether they nod at the dashboard and keep buying where they always did.

The Operator — Tuesday-morning reality: a performance team of three running "hundreds of agents" sounds efficient until the agents optimize toward corrupted signal at machine speed with no human to catch it. Quinn's fraud point is the actual news here — AI that emulates a full user journey, including a fake in-app purchase, defeats the behavioral checks that caught yesterday's bots. Now wire that fraudulent signal into an autonomous bidder. You don't lose money slowly; you lose it in an afternoon before anyone reads a report. The clean-room-as-infrastructure prediction (like cloud) is directionally right but slower than pitched — most brands still can't staff a clean room, let alone make it invisible.

The Customer / End User — The advertiser is the customer, and the advertiser is exhausted. The AdExchanger piece in the reading pile nails the real gap: programmatic optimized for who you reach, never where, and buyers still can't see what content their money ran against. Attribution maturing into three methods doesn't fix that — it adds two more dashboards to reconcile. And the internal 1:1 notes are a gut-check: advertisers being lost because of "poor campaign transparency" and "no reporting to show what ran or what worked." That's the actual demand signal. Buyers don't want a more elegant attribution philosophy. They want to know their spend wasn't wasted or defrauded, in language a CMO can repeat to a CFO.

The CFO — Real cost isn't the measurement license; it's the reconciliation tax. Three methods that disagree means a team arguing about which number to believe every quarter — that's headcount, not savings. Quinn's own hedge is telling: AppsFlyer stays out of full mix-modeling because it "requires deep customization per advertiser." Translation — the expensive, bespoke part is exactly where nobody wants to own the cost. The agentic pitch (shrink teams from eight to three) is the CFO's catnip, but the payback only lands if the guardrail actually prevents catastrophic misallocation. Pay for the fraud layer; be skeptical of paying premium for triangulation you can approximate with a good analyst and an incrementality test.

The tensions

Is independent measurement a strengthening moat or a managed decline? The Analyst sees Snap's adoption as proof independence is winning. The Skeptic sees a rebranded retreat with regulatory theater mixed in. Both can't be right — and the difference is whether advertisers move budget on these numbers or merely display them.

Does agentic AI make measurement more valuable or more dangerous? The Operator and CFO agree the fraud-into-autonomous-bidder loop is the genuinely new risk. But that cuts against the efficiency story: the same automation that shrinks teams removes the humans who'd catch bad data. The guardrail pitch only works if the guardrail is real.

Transparency vs. sophistication. The Customer wants to know where the money ran and whether it worked. The industry answered a different question — how to attribute conversions with less signal. Those aren't the same problem, and solving the second doesn't touch the first.

What it hinges on

Three beliefs do the work. First: whether triangulated measurement actually changes budget decisions or just decorates them — the whole "attribution isn't dead" thesis rests on advertisers acting on the numbers. Second: whether AI-driven fraud scales faster than detection, because if it does, the fraud-prevention layer becomes the most valuable thing a measurement vendor sells, full stop. Third: whether agency-owned clean rooms lose enough trust that independent infrastructure wins the measurement seat.

The council leans toward one durable conclusion: the fraud-and-guardrail story is the real, under-covered signal, and the "attribution matured" framing is partly vendor comfort. For operators, the move to pressure-test is your clean-room choice — if your scorekeeper is owned by your media buyer, get an independent read on anything that touches spend allocation.

The prediction

Prediction: By the end of the 2026 holiday quarter, at least one more major walled garden beyond Snap will publicly announce ingesting third-party mobile attribution signal (from AppsFlyer or a peer) into its own optimization — extending the pattern Quinn described.

Confidence: Medium — Snap set a template competitors imitate, and independent-signal adoption is a low-cost trust play.

Why: Walled gardens are under sustained advertiser and regulatory pressure over self-grading; adopting outside attribution is cheap credibility, and once one platform does it visibly, the others face pressure to match rather than look like they're hiding their own scorecard.

Revisit by 2026-12-31: We're right if a second walled garden (Meta, Pinterest, Reddit, TikTok, or similar) publicly confirms using independent third-party attribution in its optimization. We're wrong if Snap remains the sole named example and no comparable platform announcement appears.