Podcast episode

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

by Ken Rona ·July 10, 2026 ·Signal & Noise

ai-in-adtech attribution measurement programmatic walled-gardens

TL;DR

Brian Quinn, President & GM of AppsFlyer (mobile measurement and attribution platform), joins Signal & Noise to argue that attribution isn't dead — it's maturing into a triangulated methodology combining real-time attribution, incrementality testing, and marketing mix modeling (MMM). The episode is a practitioner-level walkthrough of how signal loss from Apple's App Tracking Transparency (ATT) framework reshaped mobile measurement, why clean rooms are becoming infrastructure rather than a distinct product category, and how agentic AI may collapse the measurement-optimization-execution workflow into a single loop.

What was covered

AppsFlyer's product evolution: Started 14 years ago solving mobile app install attribution; now covers web, CTV, PC/console, fraud prevention (product called P360), and data clean rooms. Revenue is entirely advertiser-side; 10,000 integrated partners.
Apple ATT impact: iOS 14 (launched during COVID) flipped the IDFA (mobile device identifier used for ad targeting) from opt-out to opt-in, cutting signal availability from roughly 80% to ~25%. Quinn says AppsFlyer deployed ~80 engineers on privacy response almost immediately and built probabilistic modeling and clean room infrastructure to compensate.
State of attribution: Quinn pushes back on the "attribution is dead" framing. Walled gardens including Snap now feed AppsFlyer's third-party attribution signal into their own optimization engines — a sign independent measurement retains value even inside closed platforms.
Multi-touch attribution (MTA): Described as "academically ideal but no longer feasible" given signal loss. The market has shifted toward triangulation: real-time attribution + incrementality experiments + MMM. Quinn mentions working with MMM vendors but keeps AppsFlyer out of full MMM build, believing it requires deep customization per advertiser.
Clean rooms: AppsFlyer entered the space post-ATT to recover lost Meta signal. Quinn sees clean rooms becoming invisible infrastructure — like cloud computing — rather than a standalone category. He flags the LiveRamp acquisition by Publicis and InfoSum acquisition by WPP and questions whether holding-company-owned clean rooms have fully aligned incentives with advertisers.
Fraud: P360 fraud product has been in market 8–9 years; described as one of AppsFlyer's most-adopted products. AI-powered fraud can now emulate full user journeys including in-app transactions, increasing detection difficulty. Quinn connects fraud prevention directly to agentic AI risk: bad data fed into autonomous agents runs unchecked without humans in the loop.
Agentic AI + measurement: Quinn predicts performance marketing teams shrink from 6–8 roles to 2–3 people running "hundreds of agents." AppsFlyer is positioning its attribution signal as the data guardrail for those autonomous workflows.

Notable claims & predictions

Quinn on walled garden self-grading: "Snap launched their universal attribution — these walled gardens are taking AppsFlyer's attribution signal to feed their own optimization. Think about all the data a walled garden has of their users, yet they're gonna take AppsFlyer's attribution signal as a priority in their own optimization. And then ask me if attribution is still a thing."
Quinn on ATT opt-in rates: "It's landed in like the mid-20s, like 25, 27%" of iOS users opted in to cross-app tracking after Apple's prompt defaulted to opt-out.
Quinn on the future of performance teams: "A performance marketing team that may have had six, seven, eight distinct roles a couple of years ago might now have a team of three running growth, with hundreds if not thousands of agents doing different things."
Quinn on clean rooms: "I believe you'll see clean rooms... I look at the analogy of cloud computing. There was a time where moving to the cloud was a strategic decision. Today it's how all software is developed. [Clean rooms are] just gonna be a critical piece of how brands implement and integrate with one another."
Quinn on MTA: "MTA was built for a world where you could track a single user across every touch point — and that doesn't exist, that no longer exists. So it's sort of an academically ideal state which isn't feasible."
Quinn on AI-driven fraud: "These AIs can emulate a user that downloads an app, behaves in the app, even can make a fraudulent transaction. The type of fraud is growing and harder to detect."

Fact check

Quinn's claim that ATT opt-in is "in the mid-20s, like 25, 27%": True but requires context. Industry estimates have ranged widely (roughly 20–46% globally depending on measurement window and app category), with earlier post-launch data from 2021 showing roughly 25% on iOS globally. Rates vary significantly by app vertical (gaming skews lower; utility apps skew higher) and by country (US opt-in tends to be lower than global average). The figure is consistent with widely cited early estimates but may understate current rates as app developers have refined consent prompts over four years. Not false, but listeners should understand this is a range, not a fixed number, and it has likely shifted since the initial rollout.

Quinn's claim that AppsFlyer has been in business "14 years": Consistent with public record. AppsFlyer was founded in 2011, making ~14 years accurate at time of recording.

Host's framing of the Uber/Facebook attribution anecdote (2018, turned off ads for three months, no measurable impact): Unverified as stated. A high-profile Uber attribution dispute is documented — Uber's former head of performance marketing, Kevin Frisch, publicly discussed discovering inflated attribution from mobile measurement vendors around 2017–2018. However, the specific detail of "three months" and the precise mechanism described by the host ("no measurable business impact whatsoever" on downloads) is a loose paraphrase of a more complex situation involving click injection fraud by ad networks rather than simply Facebook/Instagram self-attribution being wrong. The host's framing conflates the fraud/attribution-gaming story with a clean incrementality test — listeners should treat it as illustrative, not a precise case study.

Host's claim that "InfoSum got acquired by WPP": Consistent with reported news — WPP acquired InfoSum in 2023. And LiveRamp's Habu was acquired by Publicis. Both acquisitions are accurately characterized. No issue here.

Why this matters for ad-tech operators

Independent measurement as competitive moat: Quinn's description of Snap and other walled gardens ingesting AppsFlyer's attribution signal into their own bidding algorithms is a structural tell — even identity-rich closed platforms see value in third-party attribution credibility. For DSPs (demand-side platforms, software advertisers use to buy digital ads) and SSPs (supply-side platforms, software publishers use to sell ad inventory) building measurement products, the bar for trust is clearly what advertisers will act on, not just what the platform reports.
Clean room consolidation has a conflict-of-interest problem: The Publicis/LiveRamp-Habu and WPP/InfoSum deals concentrate clean room infrastructure inside agencies that also buy media on behalf of clients. Quinn's explicit concern about incentive alignment is one buyers and publishers should pressure-test when evaluating which clean room sits in their measurement stack — particularly as clean rooms increasingly carry measurement logic, not just data-matching utility.
Agentic marketing compresses the measurement vendor landscape: If autonomous agents consume attribution signals in real time to optimize spend without human review, the data quality and fraud-prevention layers of measurement platforms become more critical — not less. Vendors that are purely analytical dashboards without real-time signal output

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.

Listen to the episode →