Variant Perception

Where We Disagree With the Market

The market has accepted a "Yellow Pages moment" frame for AUTO that does not survive contact with the dealer-side data. Consensus targets sit at 629–698p (a mean ~30% above spot), yet the marginal action is uniformly bearish — Deutsche Bank, Jefferies, Morgan Stanley and JPMorgan have all cut targets or downgraded since March, and the price (493.55p) trades below even the low end of the consensus range (470p). The implied assumption is that AUTO is a single-sided discovery layer about to be intermediated by AI agents, with pricing power already broken. We disagree on three specific points: the dealer-side software embedding has matured into switching cost the market has not yet priced; the buyback is retiring 6–7% of the float annually at trough multiples in a way consensus EPS targets do not model; and the FY25 ARPR walk has been read as a structural pricing-power break when the lever decomposition shows a guidance failure on one product SKU, not on the pricing engine. The single disclosure that resolves all three is the FY26 results pack on 21 May 2026 — twenty days from now.

Variant Perception Scorecard

The score reflects an unusually clean setup: consensus is loud and recent, the dominant narrative ("Yellow Pages moment"; "easy growth days over") is observable in print, and the central disagreement resolves on a hard-dated event (FY26 results, 21 May 2026). The reason the score is 72 rather than higher is that two of the four disagreements share the same resolution window — if 21 May disappoints, we lose three claims at once. That correlation is the binding risk, not the evidence.

Consensus Map

The Disagreement Ledger

#1 — Wrong frame: the dealer side is not Yellow Pages

The consensus comparison set is CARS, CarGurus, and the original 1990s classifieds. A consensus analyst would say: car-buyer search behaviour is the moat, AI agents are the threat, and 18% of AUTO traffic already comes via organic search where Google's SGE / ChatGPT have direct line of sight. Our evidence disagrees: the dealer side has compounded into infrastructure while the market kept watching the consumer side. Co-Driver hit 85% adoption in its first twelve months; retailers log into the platform 1.8 million times each month; the API serves 91 million calls; 75%+ of dealers run Auto Trader Connect. None of those numbers existed when "auto classifieds" became the consensus mental model. If we are right, the market would have to concede that AUTO has earned a meaningful re-rate toward dealer-software multiples in a worst-case buyer-side outcome — not the Cars.com-style 8× EV/EBITDA implied by the bear thesis. The cleanest disconfirming signal is a single quarter of declining retailer logins or API call volume; if dealer engagement bends down, the two-sided defence collapses.

#2 — Wrong denominator: float compression is not modelled

A consensus analyst would say: AUTO trades at 14.6× TTM earnings on a stalling growth profile, and the buyback is shareholder-friendly window dressing. The arithmetic disagrees. At ~£275m of annual capital return against a £4.06bn cap, the program retires 6.8% of equity capital each year. At a 4.5% FCF yield and 60%+ ROCE, the company is functionally issuing equity to itself at a 22-year payback — a positive-NPV trade with the entire float. JPM and UBS Sell ratings imply targets near 470p (low end of consensus range); both appear to hold share count near constant in their FY27 EPS forecasts. If the buyback runs three more years at this pace, ~17-20% of the FY27 EPS denominator simply disappears — enough by itself to lift implied EPS from consensus 33.5p to ~40p without any operating recovery. The cleanest disconfirming signal would be a buyback pause or scale-down at the 21 May print; if management opts to conserve cash, our positive-carry case weakens materially.

#3 — Wrong horizon: one product-SKU miss is being extrapolated to a structural break

A consensus analyst would say: ARPR growth has collapsed from low-double-digits to mid-single, the product lever printed at 60% of the bottom of guidance, and management's lever-level forecasts are no longer reliable. The composition disagrees. Of the three levers, the price lever delivered at the top of guided range (£100 of the £90-100 range) — the pricing engine works. The stock lever inverted, but that was cyclical (cars sold faster, dealers carried less inventory) and was disclosed honestly. The product lever miss was concentrated in a single experimental SKU — the per-transaction Deal Builder fee — that has been folded into the core advertising bundle. Total ARPR is up 54% over five years through COVID, the chip shortage, and the arrival of generative AI. If we are right, the market would have to concede that lever-level forecasting has been noisy, but the pricing engine has not broken — and that the FY26 walk should look more normal as the experimental SKU gets absorbed. The cleanest disconfirming signal is a second consecutive sub-£100 product lever print on 21 May; one period of mix is recoverable, two is structural.

#4 — Asymmetric forensic risk: a Vanarama impairment is mis-read as bad news

A consensus analyst would say: AUTO is forensically clean, intangibles are large but uneventful, and the Vanarama useful-life cut in October 2023 was a routine accounting tightening. Our evidence disagrees on the asymmetry. Cutting useful life from 15 to 5 years without an associated impairment charge is a textbook setup for a future write-down — and the FY26 annual report (late June) is the most likely venue. If a £80-150m non-cash charge lands, the consensus reaction will likely be punitive, but the underlying FCF base is untouched. The variant view is that this is a one-direction surprise: an impairment is misread as a thesis breaker but actually clears the air, while no impairment is a cleaner-than-expected outcome. The cleanest disconfirming signal would be a separate goodwill review with no charge taken, indicating the £148m Autorama price tag remains supportable on management's models.

Evidence That Changes the Odds

How This Gets Resolved

The 21 May 2026 print is the load-bearing event for three of four disagreements. That correlation is the binding risk in the variant view: a single bad disclosure could refute the dealer-SaaS, buyback-carry, and lever-decomposition claims simultaneously. The Vanarama impairment is the one event that resolves on a different clock (late June) and is structurally asymmetric — both outcomes are useful, neither falsifies the broader thesis.

What Would Make Us Wrong

The clearest path to a wrong call is that the dealer-side software penetration we are crediting as switching cost is actually a vanity metric. Co-Driver adoption at 85% in year one could mean 14,000 dealers signed in once; 1.8m monthly retailer logins could mean a small number of power users hit the portal hourly; 91m API calls could be system polling rather than transactional dependency. None of these metrics is audited against a peer benchmark. If H1 FY27 disclosure shows engagement metrics flat-to-down in dollar-weighted terms — meaning the largest dealers (top 10 = under 7% of revenue) drive most of the engagement and the long tail is disengaging — the dealer-infrastructure frame collapses, and we are left with a weakening single-sided ad business after all.

The buyback-carry case relies on the program continuing at current pace and price. If the FY26 outlook flags slower capital return — whether to fund acquisitions, defend against the CMA outcome, or simply because management decides the multiple is no longer "trough" — the mechanical EPS support disappears, and the float-compression argument moves from "load-bearing" to "decorative." The combination of a soft FY27 retailer revenue guide (4-6%) plus a buyback pause would be the cleanest combined refutation, because it would show management itself believes neither the operating recovery nor the trough-multiple thesis.

The lever-decomposition claim depends on a recovery print on 21 May that does not happen in three of four scenarios we can construct. A second consecutive product-lever miss under £100 — even if the price lever delivers — would force us to concede that the FY25 walk was not an outlier. We would then have to choose between a slower-recovery view (still bull on a 3-year horizon) and an outright pricing-power-broken view (bear). The current variant view has no defence against two consecutive sub-£100 product levers; that is the cleanest single number that breaks it.

Finally, the Vanarama angle could be wrong in either direction. A clean impairment review with no charge in late June would suggest management's models still support the carrying value — useful information, but not "thesis-clarifying" the way we have framed it. A larger-than-expected impairment (£200m+) would be read as confirming Coe's M&A judgment is questionable, which would compound rather than relieve the broader credibility deficit on lever-level guidance.

The first thing to watch is the FY26 ARPR walk in the 21 May 2026 results pack — specifically whether the product lever prints at £100 or above.