$YOU.L: is YouGov really an AI loser? | Jonathan Cohen, Zipperline Capital

Summary

• The core call: YouGov trades at roughly six to seven times EBITDA on a ~$500M EV, if that; Cohen said it was down ~80% over two years, while Walker described it as roughly halved over the past year, because the market has bucketed it as an AI loser — while Cohen and management argue the opposite. The first slide of the company’s H1 deck reads “AI is creating a powerful structural advantage for YouGov,” and Cohen’s contrarian frame is explicit: “I want to own businesses where the market thinks they are disrupted by AI, and I think they’re literally going to be worth more because of AI.”
• The moat is not surveys but a proprietary attitudinal data set — data from 30 million people, 60+ countries, and 20 years of daily-refreshed longitudinal history. Cohen says Kantar and Nielsen do not match YouGov’s magnitude of historical data or refresh frequency. The proof point is Dieselgate: VW called YouGov the day the scandal broke because YouGov had tracked VW brand perception for 20 years without VW even being a client; any alternative “would have basically taken months to set up and you’re flying blind.”
• Cohen’s rebuttal to the synthetic-data threat leans on evidence that AI-only panels are problematic: a Cornell study cited by the New York Times found synthetic results “wholly unreliable” and full of extreme bias — one company’s AI simulation of the 2024 election wrongly called a narrow Kamala Harris win. The key distinction is synthetic panels versus synthetic data: YouGov builds auditable synthetic data on top of real humans that can be verified by re-asking the panel, extending its $50–100k high-end product down-market at lower cost.
• The low-cost-of-capital-era competition has already struggled. Cint is down ~95% since its Swedish IPO after reporting fraud within its data set, Dynata filed for bankruptcy, Morning Consult has done layoffs, and GWI’s revenue growth has slowed while it is now in a debt position — because “data quality, accuracy is by far the most important thing” and fewer players can deliver it. The irony: Anthropic itself used YouGov for its Super Bowl campaign and its public-records series polling Americans on AI.
• Cohen’s activism-light already got YouGov to turn its £10M dividend into at least a buyback after “hundreds of emails and calls” with management and the board. He now spends “75% of my day” pushing UK companies on capital allocation, ~80% of his portfolio is repurchasing shares, and his inversion of the illiquidity objection is that a buyback capped at 25% of daily volume “adds a natural buyer of your stock in the market every single day.”
• His UK rules of the road: “you never ever ever ever compare to US or even frankly European multiples,” because, relaying a dictum from a mutual friend, there are “only three natural buyers of UK stocks — share buybacks, short covering, and takeovers.” He screens out anything without $1M+ of insider ownership and prefers heavily covered names that once traded at far higher caps: “betting on things that have happened already happening again.”
• On AI winners and losers broadly, Cohen’s lesson from running net short at Coltrane through 2020–22: “the market is a particularly poor predictor of medium-term to long-term winners and losers during periods of change.” He was short touted COVID winners Peloton and HelloFresh (down ~80%) and long Greencore and Marston’s; he said Marston’s and Greene King were up 75–100% as of January. Andrew Walker’s pushback — Chegg at $1.12 and Wix show AI can genuinely destroy terminal value — draws Cohen’s filters: B2B, mission-critical, no student-focused academic end market, and no share-based-compensation issue of the kind seen in some US SaaS businesses.
• The closing framework: own content, not distribution, because content moats survive channel shifts. Warner Music IPO’d at $2B in 2005, was taken out at $3B in 2011, and is worth ~$15B today even as record stores died — while in the last decade of media “the only thing that has worked is Spotify and Netflix,” and Netflix only became highly valuable after it invested in its own content.

Deep dive

1. Game selection: why a New York investor hunts in the UK

• Cohen’s self-portrait: “I think of myself as Ted Lasso, but like Ted Lasso season 3, not season 1 where I’m out of my depth” — a decade in UK/European small- and mid-cap long/short after finding the post-Tiger-Cub US game “incredibly competitive,” with areas of genuine edge “sparse and difficult to find.” His honest answer for why the UK: “it’s simply game selection” — less competition, less liquidity, thin analyst coverage, few dedicated long/short investors.
• The structural kicker: low-hanging fruit in capital allocation, market communication, and governance — “if you think about what Dan Loeb was doing back in the ’90s in the US, it’s just stuff that these businesses in UK PLC haven’t had to think about before.”

2. Rules of the road: no US comps, and only three natural buyers

• Walker’s confession frames the problem: he kept buying “obviously too cheap” UK stocks at six times earnings expecting 18% annualized, only to watch them go to four times — “the only way you get paid in the UK market is when a private equity firm kind of comes and puts you out of your misery.”
• Cohen’s first commandment, even for globally exposed businesses: “you never ever ever ever compare to US or even frankly European multiples” — he will “immediately dismiss” any pitch built on that arbitrage, citing higher UK taxes, suboptimal capital allocation and leverage, and different growth and regulatory conditions. He relays a mutual friend Ben’s dictum that “there’s only three natural buyers of UK stocks: share buybacks, short covering, and takeovers.”
• His screens, born of diagnostic post-mortems: nothing without $1M+ of insider ownership (UK managers “are just not paid enough to really think about shareholder equity over the long term”), and a preference for names with heavy existing coverage that once traded at far higher caps and multiples — “betting on things that have happened already happening again, rather than trying to bet on something that hasn’t happened before.”

3. Sell-side reality: analysts track the share price; brokers are the conduit

• The Greencore mechanics: when a £1bn business shrinks to a ~$250–300M cap, coverage nominally persists but banks stop caring because trading revenue collapses — analysts “wait till ultimate certainty on the business to have ultimate visibility, likely when I’m selling,” then upgrade.
• The corporate-broking distortion: UK brokers often have buy ratings because they do business with the company, so the setup is “probably more inefficient on the short side than on the long side.” What Cohen actually uses sell-side for: long-tenured industry knowledge, management access, and as his “conduit into the management teams and boards” for what he calls “activism light” — shareholder stewardship on capital allocation and communication.

4. The buyback conversion: from Woodford’s dividend culture to a daily natural buyer

• The shift is real: when Cohen started, not a single business he owned was doing a share buyback; now “75% of my day is focused on getting companies to do share buybacks” and roughly 80% of his book is repurchasing — helped by brokers being paid to execute them.
• His inversion of the illiquidity objection: UK rules cap buybacks at 25% of daily volume, so a £20M buyback in a £300M cap can take months — which “adds a natural buyer of your stock in the market every single day.” Walker’s parallel argument: no company has shown him the proof point of a buyback-induced liquidity spiral, and if it happened “your stock is going to go up… you can always stop.”
• Why the dividend culture existed at all: Neil Woodford and Mark Barnett “were like the kings of the UK market,” so starting a dividend was the easiest route to natural flow — but those income investors have massively dwindled. Cohen’s concrete win: after “hundreds of emails and calls,” YouGov agreed to turn its £10M dividend into at least a buyback, within a more flexible capital-allocation plan rather than a permanent commitment.
• On leverage, a change of posture worth noting: “I’ve kind of come to grips with the idea that the market doesn’t like leverage. Frankly, if that’s the worst thing in the world, that’s kind of okay” — 1x instead of 2–3x limits equity downside.

5. YouGov is a proprietary data business, not a survey shop

• Cohen’s definition: YouGov “screens at first as kind of a market research firm — in reality, it’s a proprietary data business,” with everything sold derived from attitudinal data on 30 million people across 60+ countries over 20 years, tracking awareness, perception, and purchase funnels. Live examples as told: BYD gauging European brand perception, Marks & Spencer outsourcing its tracking, a book publisher finding the right demographic and retailer, and a hedge-fund client tracking consumer tariff sensitivity when tariffs came out in April.
• Revenue quality by his estimate (the company doesn’t report it): ~60% of revenue and ~70% of operating profit is recurring or repeat — syndicated daily-refreshed trackers, custom surveys, and a growing AI-enabled self-service platform, versus big-project market research.
• His stated framework — “business plus situation”: a quality compounder with barriers to entry, plus a reason the opportunity exists today. Versus Kantar and Nielsen, YouGov has by far the longest-standing data set and a more frequent refresh — daily versus weekly, monthly, or quarterly — and “you’re looking for changes versus history.”
• The Dieselgate proof point: VW called YouGov the day the scandal broke because YouGov had tracked VW perception for 20 years without VW being a client, enabling real-time measurement of whether crisis response was working — “another solution would have basically taken months to set up and you’re flying blind in the heat of the scandal.”

6. The panel is the moat: brand, retention, and an AI bot asking “why”

• The brand does real work: YouGov is “the number one or two most cited research source in the world,” with Nate Silver ranking it among the top five for polling accuracy — and it makes “basically no money” on polling, which exists “to create a brand of accuracy” that’s trusted and verified.
• Panel economics: loyal panelists treat it “as like a second job, a side hustle” (Cohen has seen LinkedIn profiles listing YouGov panelist), paid mostly in gift cards; the high retention of panelists makes the data apples-to-apples over time, and the real cost is maintaining the panel technologically, not acquiring people.
• BrandIndex Voices is the AI product he calls “a game-changer”: a bot at the end of every panel asks “I see you changed your answer on this — why?” About a quarter of panelists engage, some staying on for an hour and discussing unrelated things — helping YouGov gather increasingly detailed data while turning the expensive, weeks- or months-long custom “why” survey into tens of thousands of minimal- to no-cost daily interactions.

7. The synthetic-data rebuttal: “pure fiction” versus auditable extrapolation

• The market’s knee-jerk — surveys, therefore AI roadkill — misreads the asset: unlike Qualtrics, “one literally can’t get the data from a YouGov survey without access to the proprietary data set” and the panel; applying AI to that data “only further enhances the value of it.”
• Against pure AI panels, Cohen cites the evidence: a UK study by Strat7 found synthetic data “lacked logical consistency,” and a Cornell study covered by the New York Times found results “wholly unreliable” and full of extreme bias — one company’s AI simulation of the 2024 election inaccurately reported a narrow Kamala Harris win. The Times article calls such results “pure fiction.”
• The structural point: YouGov’s product is measuring deviation from trend, and “by definition synthetic panel data is simply extrapolating from past data — it’s not a human changing.” The distinction that matters is synthetic panels versus synthetic data: YouGov builds synthetic data on real panelists that can be audited and verified by re-asking humans, expanding its use case below the $50–100k high-end at greater speed and lower cost.

8. The competition already struggled — and Anthropic uses YouGov

• The low-cost-of-capital-era cohort that stitched third-party data sets together and “claimed it’s one large proprietary data set” — GWI, Morning Consult, Cint, Dynata — raised at “unicorn-like valuations” with “no regard for price,” and is now struggling: Cint is down ~95% since its Swedish IPO after reporting fraud within its data set, Dynata filed for bankruptcy, Morning Consult has done layoffs, and GWI’s revenue growth slowed while it moved into a debt position. Cohen’s read: “data quality, accuracy is by far the most important thing, and there’s just fewer and fewer people that can do that.”
• The irony he loves: YouGov’s shares fall “basically anytime Anthropic launches something,” yet via his weekly LinkedIn primary-research sweeps he found Anthropic thanking YouGov for its Super Bowl campaign — and Anthropic’s public-records series surveying Americans about AI also uses YouGov. Walker’s complementary observation from the KPMG-hallucination story: in a world of AI slop, “the known brands actually kind of cut through” — trust in the data’s source may be the durable asset.

9. “The market is a particularly poor predictor” — and Walker’s Chegg counter

• Cohen’s formative experience running largely net short at Coltrane through 2020–22: hype-cycle experts “typically exit stage left as quickly as they appeared,” and “the market is a particularly poor predictor of medium-term to long-term winners and losers during periods of change.” His receipts: short touted structural winners Peloton and HelloFresh (down ~80%), long structural “losers” Greencore and Marston’s; he said Marston’s and Greene King were up 75–100% as of January.
• Walker’s pushback — worth keeping in full: Chegg’s management said AI would be great for them the day ChatGPT launched, bought back shares, insiders bought — and the stock is at $1.12; Wix looked “really obvious” to Walker because many websites could be made with ChatGPT rather than Wix, even as the company levered up for a tender and bought shares. So the AI-loser bucket sometimes contains actual zeros.
• Cohen’s concession-plus-filters: his relevant businesses are B2B-focused and mission-critical, and he avoids an academic end market where “your end consumer is a student” and open-source pressure is permanent. YouGov “hasn’t pushed price as much as they should,” and — the UK bonus — “you never have to worry about share-based comp. The multiples are the multiples.”

10. Content beats distribution: why AI could make YouGov worth more

• Cohen’s simple-man framework: content businesses (data, brands, IP, even experiential real estate like pubs) have deeper, more sustainable moats than distribution middlemen, because “inevitably you can move the content to a different form of distribution.” His thesis statement: “I want to own businesses where the market thinks they are disrupted by AI, and I think they’re literally going to be worth more because of AI” — versus fighting a compressing terminal multiple on merely AI-resilient names.
• The music case as told: Warner Music IPO’d at $2B in 2005, was taken out at $3B in 2011, is worth ~$15B today — while the record stores died and, per Walker, “the only thing that has worked in media is Spotify and Netflix,” Netflix only after it invested in its own content. Cohen’s refinement: brands versus retailers is the better analogy — “there’s only one Walmart today after the internet,” and one Target, but many long-standing brands survived.
• Walker’s back-catalog bridge to YouGov: UMG’s catalog became more valuable as distribution fractured — maybe 20-year attitudinal history compounds the same way — though he flags his own doubt: SpaceX has no 20-year history and is now among the world’s most valuable companies. Cohen’s reply: BYD proves you don’t need the history to become a client; “the longer you have it, the better,” and if a customer churns, “YouGov owns the data.”
• The honest hedge on timing: music labels “went through very difficult periods of time before they figured it out” — but AI is moving faster than streaming did, YouGov already has commercialized AI products, and “yeah, I appreciate it’s a super anti-consensus view — but I’d stick by that.”