Allstate Is Building an Agent That Will Happily Sell You a Rival's Policy
A 39.7% ROE quarter, a combined ratio eight points better than target, and a CEO who says the margin give-back is entirely discretionary
ALL · Earnings Call · 2026-11-05
An 86.6 combined ratio, and a CEO who says he can end it
Allstate's third quarter was, on the page, a monster: $2.5 billion of adjusted net income, $9.76 per share, a 39.7% adjusted return on equity, and a Property-Liability combined ratio of 86.6. Auto was 83.2 — flattered by 9.1 points of prior-year reserve re-estimates — with an adjusted underlying ratio of 91.4 against a mid-90s target. Homeowners ran a 60.1 underlying ratio. That is not a normal quarter; that is a quarter with a cushion in it.
Josh Shanker of Bank of America put the uncomfortable question plainly: auto's underlying loss ratio deteriorated roughly 600 basis points year-over-year while earned pricing fell about 3%. Tom Wilson did not hide behind severity statistics. “You can extrapolate from this a year ago to next year and get the same kind of deterioration. This is in our hands. We choose what we choose to do, and we are happy with where we are at.” — Tom Wilson, Chair, President & Chief Executive Officer · 2026-11-05 Read that twice. The CEO is telling you the current margin is a policy choice made to buy physical damage growth and average premiums down 2.7% — and that the direction of travel from here is lower, on purpose.
Which makes the reserve question the load-bearing one. Andrew Kligerman of TD Cowen cited 9.1 points this quarter, 8.8 in Q1, 6.6 in Q2, 7.5 in Q4, and asked whether to model it. The answer was unambiguous:
I would not forecast in your earnings estimate reserve releases in the future. It's not the way we run the place.
That is now the third consecutive call on which this exact question has been asked — Kligerman raised the same "each of the last four prior quarters" pattern in August, and Shanker probed the 2023/2024 accident-year conservatism back in April. Each time management says the same thing. In August it was “every quarter, we think we got reserves as accurate as we can get them” — Thomas Wilson, Chairman, President & CEO · 2026-08-05 — followed by another quarter of favorable development. Consistent, credible, and still unexplained enough that analysts keep circling it under a benign label like underlying loss trend.
Marketing to machines, and to competitors' products
Buried under the underwriting noise is the genuinely new thing: Allstate's personal agentic assistant. Wilson's framing of the shift is the sharpest sentence in the deck — “Essentially, we're now marketing to machines and people, not just marketing to people.” — Tom Wilson, Chair, President & Chief Executive Officer · 2026-11-05 The company is building its own personal assistant for protection, launching early 2027, that will sit in the flow of commerce, do straight-through processing, and — this is the part that matters — quote multiple carriers, not just Allstate. Asked directly by Wolfe's Tracy Benguigui whether it would quote competitors, Wilson did not flinch: “Yes. Yeah, it does today. Answer Financial does today.” — Tom Wilson, Chair, President & Chief Executive Officer · 2026-11-05
That is a real strategy, not a slide. Answer Financial already writes over $500 million of premium across more than 30 insurers, and Wilson's argument is that a machine-to-machine broker with better pricing intelligence needs less compute per quote and can serve the general-purpose agents rather than fight them. The price comparison engine becomes the asset; the brand becomes an option on top of it.
This is not a lonely bet. Muse agent sits among the market's top global keywords this quarter, and personal agents surfaced across this week's reporters — Booking's horizontal personal agent, NICE's personal AI agents, Airbnb's general-purpose agent, Shopify's. Allstate is not early to the wave; it is arriving with a different ticket, and it feeds the same machinery: mobility intelligence from Arity plus the announced INRIX and Replica acquisitions, crash detection, 3 trillion miles of driving data.
What's changed versus prior quarters is specificity. In April, agentic AI was framed as internal: “I think the real benefit from this will come from agentic AI, where agents are talking to agents” — Thomas Wilson, Chairman, President & CEO · 2026-04-30 — a cost story. In August, Wilson declined to quantify the P&L impact. Now there is a product, a date, a distribution mechanism, and a disclosure promise on the cost side: “Maybe we will break it out separately. Maybe we will footnote it. I don't know.” — Tom Wilson, Chair, President & Chief Executive Officer · 2026-11-05 Investors should hold them to that. The AI build is inside an expense ratio that management is simultaneously promising to reduce.
The multiple is the argument
Here is the tension the quarter cannot resolve. The last filed period showed revenue of $18.6 billion, up 12% and net profit margin near 17.6% against a collapsing earnings multiple: price-to-net-income of 4.6x and price-to-free-cash-flow of 4.9x. That is the market's verdict on Wilson's own admission: these are peak auto margins, and the company is deliberately steering them toward the mid-90s.
The stock has already re-rated regardless — up 9.7% over the last 90 trading days, just 3.0% off its August 6 peak of $275.11, and +762.9% across a decade and a half. Capital return is doing heavy lifting: $4.4 billion returned over twelve months, 5.7% of shares repurchased, roughly $9 billion of deployable holdco capital with no requirement to push it into subsidiaries. The quarterly repurchase line and a 73.0% liabilities-to-assets ratio say the constraint is price, not capacity.
Growth itself is real, and it is coming from all three channels — exclusive agents down about 5% in headcount but more productive, direct, and independent agents all contributing to 3.3% policies-in-force growth. Advertising spend has scaled past $2 billion over the trailing year from $638 million in 2023. Retention, the perennial weakness, was described as stabilized in August and not relitigated this week.
What to watch
The regulatory variable is now state-level and freshly reshuffled: five governors and seven legislators changed hands, and Wilson was blunt that policy reform through legislation is the cheapest cost lever he has — Florida's tort reform let Allstate cut prices 6%, and New York's reelected governor has passed reform that has not yet shown up in claims costs. That is upside optionality sitting outside the P&L.
Strip it back and Allstate is asking you to believe two things at once: that a 39.7% ROE is a real earning power number being spent down on purpose, and that a 2027 AI shopping agent will convert a distribution cost line into a platform. At 4.6x earnings you are not paying much for the second. You are, however, being asked to trust the first — and the CEO has already told you he controls the slope.