XZO — Exzeo Group, Inc.

Exzeo Group, Inc. (NYSE: XZO) — deep-value analysis, REFRESH revisit

September 7, 2026 · reference price … · 90,092,830 shares · market cap … · enterprise value … · sector: IaaS for P&C insurance

Why the analysis is being revisited. The mechanical triage deep_delta.py flagged REFRESH on September 6, 2026, with a single falsifier triggered out of seven: crestere_venit_qoq (sequential revenue growth) came out positive for two consecutive quarters (… in Q1 2026, … in Q2 2026), which, mechanically, contradicts the reference thesis’ claim that “all four leading indicators declined sequentially simultaneously in Q2 2026.” Diff package: delta-XZO-20260906.md.

What I explicitly inherit from the August 16, 2026 report (not re-derived, verified that nothing moved in the filings): the business structure and intra-group contracts, the capital-allocation history (the 2024 TTIC transfer, the IPO, the buybacks, the treasury portfolio), the CEO profile and Thorndike grid, the accounting red-flag list, the architecture of the five valuation models, and the FCF bridge. There is no new SEC filing after August 16: the last document is the Q2 2026 10-Q (period ended 06/30/2026), and the …lion buyback program from August 14 is a press release, not a filing.

What I re-derive in this report: (1) verification of the broken falsifier, on seasonally comparable data — the “What changed” chapter; (2) the growth assumption g1, lowered from 5.0% to 4.0%, with the arithmetic reason; (3) all five valuation models and the Monte Carlo simulation, at the current price; (4) the quarterly attribution of the intra-group receivables jump, which changes the nuance of the O’Glove chapter; (5) the comparison with the GBL score, which didn’t exist in the tracker in August and does now.

Correction to the diff package. The “Market: then vs now” table in delta-XZO-20260906.md reports the price “at analysis” as … and a move of …. This is wrong: data-pack-XZO-20260816.md and the reference report both use …. The … figure is a snapshot taken on August 30, when teza-XZO-20260816.json was written, not the price at the analysis date. The real move from thesis to today is …, not … — and it matters, because that’s exactly what consumes the safety margin.

Primary sources: 10-K FY2025 (26.02.2026, CIK 0001873951), 10-Q Q3 2025, 10-Q Q1 2026, 10-Q/A Q1 2026, 10-Q Q2 2026, 8-K 26.05.2026, 8-K 09.06.2026, 8-K 06.08.2026, DEF 14A 2026 — all in SEC-Filings/XZO/. Market data: data-pack-XZO-20260907.md. Context: research-XZO-20260907.md (brief treated as a deposition; corrections to it are marked in the text). Simulation: mc-XZO-20260907.json.

Executive summary

The thesis, unchanged in substance. Exzeo isn’t a software company sold to insurers; it’s HCI Group’s fee arm, listed in November 2025 with a free float of ~17%, and its economics are determined entirely by four intercompany contracts signed by the same person on both sides of the table. In Q2 2026, … of … thousand revenue — 93.8% — came from related parties (10-Q Q2, note (1) to the income statement); two individual clients make up 85.6%. The …% operating margin isn’t evidence of superior efficiency, but the arithmetic consequence of contractual rates (21.5% of TTIC premium, 8.5% of HCPCI premium) set by a CEO who has more capital on the paying side than on the collecting side.

The broken falsifier: verdict — wrong proxy, the substantive claim holds, but with a nuance that matters. Revenue really did grow sequentially two quarters running. But the thesis’ claim wasn’t about revenue, it was about the four snapshot-type leading indicators, and those fell and remain down: managed premium … → … thousand (…), policies 327,233 → 321,141 (…), premium per policy … → … (…, the third consecutive quarter of decline), ARR … → … thousand (…). The check the mechanics doesn’t do is the seasonal one: revenue always grows sequentially in Q1 and Q2 at a Florida homeowners MGA, where renewals bunch up ahead of hurricane season. Last year the same window produced … and …; this year … and …. Cumulative, Q4→Q2: … in 2025 versus … in 2026. On a seasonally comparable basis, growth decelerated by two-thirds, it didn’t recover. The falsifier was poorly specified, not the thesis disproven — and I flag it as such for future runs.

What is nonetheless new information in its breaking. Q2 2026 revenue (… thousand) is an all-time high, above the previous Q2 2025 peak (… thousand), and the trough was Q4 2025 (… thousand). “Revenue is rolling over” would have been wrong and I don’t use it. The reconciling mechanism is that revenue is earned on premium written during the quarter, while managed premium is a point-in-time balance; the decline in the balance as of June 30 will only show up in revenue in Q3-Q4 2026. The reporting due ~November 5, 2026 is the decisive test of the thesis, not August’s.

The growth assumption, lowered from 5% to 4% — the only opinion change in this report. The reason is arithmetic, not sentiment: the company guidance that the 5% rested on is on pre-tax profit, and in the first half of 2026, of the … thousand increase in pre-tax profit, … thousand (72.8%) came from investment income, i.e. interest on the …lion of treasuries. Operating profit — the only flow I discount, since cash enters separately as negative net debt — grew only 3.1% (from … to … thousand). The FY2026 guidance of …-125 million, cleaned of investment income estimated at ~…lion in H2, implies H2 2026 operating profit growth between … and …, with the midpoint at …. Four percent is already generous against that.

The balance sheet remains impeccable, and that’s real. Zero financial debt since July 1, 2024. … thousand cash plus … thousand US treasuries AFS = … thousand of liquid assets, minus … thousand of operating leases = net debt −…lion, i.e. …/share and 21.9% of market cap (it was …% at … — the price move ate into the cushion). Capex of … thousand per half-year, 0.5% of revenue. No goodwill, no acquisitions, no provisions. Current ratio 1.85.

Estimated value. Five triangulated models at … give a range of … – … per share, with a median of … The Monte Carlo simulation over 20,000 scenarios gives a median intrinsic value of … and an undervaluation probability of …% (it was …% in August, at … and g1 = 5%). Median margin of safety: …. The two paths — deterministic and stochastic — converge within 6 cents of each other, which says the range isn’t held together by a calculation artifact.

Verdict: AVOID at … In August the verdict was NEUTRAL leaning AVOID, at a price 9.1% lower and with a growth assumption a point more generous. Nothing in the business has worsened in the meantime; only the price-to-flow ratio has worsened. It isn’t a value trap — it has no debt, doesn’t burn cash, generates …lion of free flow a year, and buys back its stock at lows. But a …% probability of undervaluation doesn’t justify any position, and the main risk remains structural, not operational: 93.8% of revenue can be terminated by the parent with 120 days’ notice at the end of any quarter, without cause (DEF 14A, “Transactions with Related Persons”). Rational entry point: below … — the level below which three-quarters of the simulated scenarios are profitable — with a full position only below … the earnings-power value.