Ethos Technologies Inc.

Summary

Ethos sells other companies' life insurance. It takes no insurance risk itself; it earns a commission for finding the customer, deciding within minutes whether that customer can be insured, and then administering the policy.

The company reported a large Q2 beat, roughly doubled its full-year revenue guidance, announced a $100M share buyback, and the stock went from $18.50 to $39.85 — up 115% in six weeks. The growth is real and the cash generation is real. But two things in the Q2 filings deserve attention and were not discussed on the earnings call:

• the profit margin that actually matters did not recover, and management now describes the Q1 markdown as an ongoing condition rather than a one-time event;
• the concentration on a handful of insurance partners got worse, not better; and

Revenue more than doubled year-over-year for a second straight quarter. The company generated $66.9M of operating cash in six months. It authorized its first buyback. It nearly doubled its own full-year forecast.

At the same time: the margin that measures whether a policy is worth selling did not recover, and the 10-Q now attributes that to an ongoing condition rather than a passed event. Dependence on one carrier increased to 43% of revenue and 67% of receivables. A bank facility that two carriers required expired unexplained. Almost half the balance sheet is an estimate of money that will arrive over the next decade. And management did not provide details about "persistency" on the earnings call.

None of that is evidence of anything wrong. It is a description of a company that is growing very fast, is genuinely cash-generative, and whose reported results depend to an unusual degree on judgments that only management can see — with a governance structure that leaves outside shareholders no way to press for a clearer view.

The stock has doubled since the last time I looked at it. My valuation suggests it is rightly priced to overvalued.

Part 1 — What the company actually is

It is not an insurance company

When you buy life insurance, someone has to promise to pay your family if you die. That promise is a financial liability. The company making it has to hold capital in reserve against it, is regulated as an insurer, and takes the risk that people die sooner than expected.

Ethos does none of that. Ethos is the shop, not the manufacturer. Ethos is the marketer, seller and partial administrator. It finds the customer, processes the application, and hands the finished policy to a real insurance company — Ameritas, TruStage, Banner Life, North American, Aflac, or Protective. Those companies, called carriers, hold 100% of the risk. Ethos gets paid a commission for the intermediation.

This is why Ethos can grow fast without raising much capital. It is also why its fortunes depend entirely on a small number of carriers continuing to want its business.

The three-sided arrangement

Ethos sits in the middle. It also acts as the third-party administrator (TPA) — the back office that handles billing, changes to policies, and record-keeping after the sale.

Two ways a policy gets sold

This distinction runs through everything that follows.

• Direct channel (DTC) — the customer finds ethos.com themselves, through a TV ad, a Google search, or Facebook. Ethos pays for advertising, but keeps the whole commission. Higher revenue per policy.
• Third-party channel — an independent agent sells the policy using Ethos's software. Ethos shares the commission with the agent. Lower revenue per policy, but Ethos didn't pay for the advertising.

In Q1'26 the direct channel was 76% of revenue. In Q2'26 it was 61%. That shift is behind several of the numbers below.

Why customers use it

Buying life insurance traditionally takes four to eight weeks and often involves a nurse visiting your home to draw blood. Ethos asks questions online, pulls data about you from third-party databases (prescription history, credit information, driving records, motor vehicle and medical-claims data), and gives 95% of applicants a decision immediately. The company reports a customer-satisfaction score (NPS) of 70 against an industry average of 14.

The technology is genuinely good. Whether it is defensible is a separate question.

Part 2 — The one Accounting that matters

Almost every risk in the company traces back to a single decision about when revenue gets recorded.

The problem

A life insurance policy pays Ethos a commission for years. A big payment in year one, then smaller payments each year the customer keeps paying premiums — typically modelled over about ten years.

So when a policy is sold, how much revenue should Ethos record today?

Ethos's answer: record it all today

At the moment a policy activates, Ethos books the first-year commission plus its estimate of every future renewal commission — all as revenue, immediately. The future payments are discounted by a persistency estimate: management's guess about what percentage of policies will still be alive in each future year.

The corresponding asset on the balance sheet is called commissions receivable — money earned on paper but not yet collected. At June 30, 2026 it stood at $324.3M ($30.9M current + $293.4M long-term), against total assets of $696.1M (10-Q Q2'26). Nearly half the company's balance sheet is a claim on commissions that have not yet arrived and depend on customers not cancelling.

Why this matters

Three consequences, and they are the reason this company is harder to read than its growth rate suggests:

C1. Revenue is partly an opinion. A large slice of any quarter's reported revenue is not cash received. It is an estimate of cash that will arrive over the next decade if customers behave as predicted. Management says so explicitly in the last 10-Q: recognizing revenue "requires significant judgment in applying a persistency estimate."

C2. The estimate can be revised, and revisions hit immediately. If customers cancel faster than assumed, the effect can be felt simultaneously in both Balance Sheet and Income Statement. Ethos must reduce both previously recognized assets and revenue — retroactively for the whole existing book, in a single quarter. This is not hypothetical. It happened in Q1'26.

C3. Profit and cash are structurally different numbers. Ethos records revenue years before it collects it, while paying agents within one day of a sale. It funds that gap itself.

Where a bad surprise shows up

A cancellation surprise does not appear in one place. It appears in four, and which one depends on the channel and the carrier:

The practical lesson: a persistency problem at Ethos does not necessarily look like falling revenue. In Q1'26 it looked like a marketing expense while revenue grew 104%. That is why the contribution margin, not the revenue line, is the number to watch.

The audit backlog — a mechanism the Q2 filing describes more fully

The Q2 10-Q expands on something worth understanding. After a policy is issued, Ethos performs post-issue audits — checking whether the applicant's answers hold up. When they don't, Ethos recommends the carrier cancel the policy.

The carrier decides when, or whether, to act. The 10-Q says (emphasis added):

From the filing

"If these carrier actions following our post-issue audits are delayed for any reason... we have at times experienced and in the future may experience an outsized amount of policy terminations concentrated in certain periods when such carriers address their backlog of policy audits."

So cancellations can arrive in lumps, on a schedule Ethos does not control and has "limited to no visibility" into. This is a structural source of unpredictable quarters, independent of how the business is actually performing.

Part 3 — Analyzing the Financial Statements

All figures from the 10-K, the two 10-Qs, and the Q2 release. Dollars in millions.

Guidance, as revised on Aug 3:

Diving deeper in Q2

The Q2 beat was enormous and it is not a rounding error. Management guided Q2 revenue to $114–118M in May. Actual was $189.6M — 61% above the top of their own three-month-old forecast. They then raised the full-year number by $166M, a 29% increase, in a single quarter.

But three lines in that table are moving the wrong way, and they are connected.

• Revenue per policy fell 8% year-over-year ($1,920 → $1,758). It fell 20% sequentially, but that comparison needs a seasonal control — see below.
• The mix swung hard toward the lower-priced agent channel. Third-party revenue went from $47.1M to $73.1M (+55%) while direct went from $146.0M to $116.5M. Direct fell from 76% to 61% of revenue — though 61% is close to where FY2025 ended (63%), so it is Q1's 76% that stands out, not Q2's 61%.
• Contribution margin did not recover. More on this below, because it is the central finding.

Is the sequential move seasonal? Partly, and the control matters. Q1 is structurally the strongest quarter — the 10-Q says demand is "elevated by factors such as annual financial planning cycles that typically occur post-New Year," when digital marketing is also cheapest. Comparing the same two quarters a year earlier:

Three conclusions from the sequential comparison.

• The direct channel's decline is ordinary seasonality — it fell about as much last year, and is not a signal.
• The abnormal event is the third-party surge: +55% sequentially against a +16% seasonal norm, with activations rising 22% where they normally fall.
• The sequential ARPU drop is mostly arithmetic — agent-sold policies carry lower revenue per unit, so a channel surge of that size mechanically drags the average.

Sequential comparisons in this business are dominated by seasonality and by whichever channel happened to surge. Year-over-year controls for both — and it points the opposite way from management's explanation:

Year-over-year, the higher-priced channel gained share — and ARPU still fell 8%. Channel mix was a tailwind to ARPU in that comparison, not a headwind. Whatever pushed revenue per policy down was strong enough to overcome it.

Worth setting against the longer arc, because it cuts both ways. Direct channel share fell steadily for three years — 75% (FY23) → 68% (FY24) → 63% (FY25) — as whole life and IUL sold through agents grew faster. 2026 has so far reversed that: 68.6% for the first half against 61.1% a year earlier. So the year-over-year gain is real and not a seasonal artefact, but it is a one-half reversal of a three-year secular drift toward the agent channel, not an established new trend. (FY23–FY24 channel split from the IPO prospectus; FY25 onward from the 10-K and 10-Qs.)

So the sequential drop is mostly channel mix and seasonality and should not alarm anyone; the year-over-year drop of 8% cannot be channel mix and remains unexplained by anything management has said. The candidates are

• product mix within channels (more whole life, which the 10-Q says carries "lower coverage amounts and commission rates"),
• lower negotiated commission rates, or
• lower persistency estimates feeding into the revenue booked per policy.

The filings do not disclose enough to tell which.

One mitigating figure: six-month ARPU was $1,950 against $1,946 a year earlier — flat. Q1'26's $2,185 was an outlier on an unusually high 76% direct mix. So the erosion is real in Q2 but has not yet shown up across a full half-year.

And the guidance implies deceleration. Q3 is guided to $160–164M — below Q2's actual $189.6M. The full-year figure implies roughly $184M in Q4. Growth stays high year-over-year (roughly +73% in Q3, +67% in Q4) but the sequential trend flattens.

What "contribution margin" means and why it is the number to watch

Gross margin at Ethos is 98% and it is close to meaningless. Ethos counts almost nothing as "cost of revenue" — only the data-lookup fees on applications that actually became policies ($4.0M in Q2 on $189.6M of revenue). Everything that actually costs money — advertising, agent payments, the data fees on applications that were rejected — sits below that line.

Contribution margin strips out the cost of getting the policy, which gross margin ignores. But the company's definition is not simply "gross profit minus sales and marketing." Per the 10-Q, Contribution Profit is gross profit less S&M plus four add-backs for costs that sit inside the S&M line but are not per-policy acquisition cost:

That matters, because the two bases give different answers — and the stricter one tells a harsher story:

Two things only the strict basis shows.

FY2025 margin expansion was entirely the add-backs. On gross profit less S&M, contribution margin was 39.1% in FY2024 and 39.1% in FY2025 — flat. The reported improvement from 41% to 42% came from allocated overhead and stock compensation more than doubling, $4.8M to $10.5M. There was no underlying expansion.

Q1'26's reported 30% includes adding back $10.4M of IPO-triggered stock compensation that landed in S&M. On a basis that expenses it, Q1'26 was 23.7%.

A consistency note, since this dossier criticises Adjusted EBITDA for excluding stock compensation: Contribution Margin adds it back too, in smaller amounts ($1.2M in Q2 against $15.5M of total SBC). The same objection applies and should be applied evenly. The figures quoted below are the company's, for comparability with its filings and with the July note; the strict-basis figures above are the more honest read.

The previous July note framed the test precisely: "If it stays at 30–33%, the persistency / agent-comp compression is not a one-time. If it jumps back to ~38–40%, Q1'26 was a genuine one-time cohort cleanup."

It printed 33% as reported — 30.3% on the strict basis. The decline versus a year earlier was attributed to "the change to third-party agent compensation and persistency estimates implemented in the period ended March 31, 2026" — that is, the Q1 change is now described as an ongoing condition affecting current-period economics, not a one-time charge that has passed.

On the earnings call, CFO Christopher Capozzi framed the compression differently — as a deliberate consequence of channel mix: "That dynamic is just naturally causing margins to mix down with a significant acceleration of growth in third party." He added that every policy is "variable cash flow positive within 60 days."

The channel data lets us test the mix explanation directly, and it fails. The comparison the CFO's statement invites is Q1 to Q2, where agent business did surge. But the margin figure the 10-Q reports against is the year-ago quarter — and there the mix moved the other way:

Q2'26 had a smaller share of revenue from the lower-margin agent channel than Q2'25, and a contribution margin nine points worse as reported — 9.9 points worse on the strict basis. Channel mix was working in favor of the margin, not against it. It cannot be the explanation.

That leaves the 10-Q's own attribution — the Q1 agent-compensation and persistency change — as the account the data supports. Two qualifications, and both point somewhere worse than mix. Product mix within channels may have shifted toward whole life, which carries lower commission rates. And the 10-Q's risk factors note "increased pressure to offer higher agent compensation or incentives" as agency relationships expand — meaning the cost of agent business may be rising per policy, independent of how much of it there is.

Part 4 — The standing issues, and how they evolved

On the guidance beat

A 61% beat against a self-set three-month guide is obviously good news. Management set that $114–118M number in May, after the Q1 charge, with full knowledge of the dynamics. They then missed it by $72M in the right direction.

There are two readings. Either the business genuinely inflected in ways nobody could model — which the third-party channel's 55% sequential growth partly supports. Or the company cannot forecast its own revenue, in which case the raised full-year number carries the same uncertainty in both directions. The 10-Q itself supports the second reading in its risk factors: persistency estimates "have negatively impacted and may in the future negatively impact our ability to accurately predict future revenue and cash flows."

This is the second consecutive quarter of a very large beat against fresh guidance. That pattern was already flagged in the previous 10-Q analysis as a sandbagging question; it now has two data points.

On the buyback

The Stock Based Compensation is a determining piece of news for the multi-year shareholder math.

On August 2, the board authorized up to $100M of Class A repurchases.

What it is worth:

• At the Aug 3 price (~$22.83), $100M was about 7% of the market value.
• At today's price ($39.85), it is about 3.9%.
• The 2026 incentive plan reserves 18,962,000 shares for employee compensation. At $39.85 that reserve is worth roughly $756M. The buyback authorization offsets about 13% of it.
• It is an authorization, not a commitment. Management said on the call they will be "opportunistic... repurchasing shares only when we believe the market price does not reflect the underlying value of the business." No shares had been repurchased as of the filing.

A real and meaningful change in posture, and the first evidence that management thinks about per-share value at all. Not, by itself, an offset to the dilution program.

On the letter of credit

Two carriers had required Ethos to post a $30M unsecured letter of credit at Citizens Bank — effectively a bank guaranteeing Ethos's obligations to them. The Q2 10-Q says only: "The letter of credit expired in June 2026." No replacement is disclosed anywhere in the filing.

• Benign reading: the carriers no longer require the backing. Ethos's balance sheet is stronger post-IPO ($252.9M in cash and investments), so the guarantee became unnecessary. This is the most likely explanation.
• Less benign reading: it lapsed because the relationship or the terms changed, and the filing simply doesn't say.

On carrier concentration

This is a risk identified before that has deteriorated.

Ethos has six carriers. But revenue is not spread across six:

The FY24→FY25 improvement that was read as successful diversification has reversed. And on the balance sheet it is more extreme: one carrier accounts for 67% of all receivables (10-Q Q2'26).

Why this matters in plain terms: carrier contracts set commission rates and are reviewed annually. If the carrier representing 43% of revenue and 67% of receivables decides to cut Ethos's commission rate, build its own online channel, or simply stop, there is no contractual protection and no practical substitute in the short term. Ethos's defence is that it delivers so much volume that the carriers won't — it was the #1 source of new life premiums for three of its carriers in 2025. That is a real defence, but it is a commercial one, not a legal one.

Nothing in the Q2 filings suggests this is happening. The point is that the exposure grew while the previous reports after 1Q described it as improving.

On persistency

What happened in Q1: Ethos revised its assumptions about how long agent-sold policies stay in force, and about what it can recover from agents who were paid commission on policies that lapsed. The result was a $16.5M charge to sales and marketing.

What Q2 tells us: no new charge. But the 10-Q attributes Q2's still-depressed contribution margin to that same Q1 change, meaning the effect is running through current economics rather than being behind the company.

What is genuinely reassuring: the persistency reserve liability grew from $4.9M to $43.2M in six months. That is cash received from carriers that Ethos is declining to record as revenue. It is the opposite of aggressive accounting, and it is the specific early-warning signal previously identified — a shrinking reserve alongside rising agent-comp expense would have been the bad pattern. The reserve did the opposite.

What is unresolved: persistency was not mentioned once on the Q2 earnings call. For a company whose largest single accounting judgment is a persistency estimate, and which took a $16.5M charge on that estimate one quarter earlier, its complete absence from management's prepared remarks and the Q&A is a disclosure gap. Everything meaningful on the subject is in the 10-Q risk factors, which is where companies put things they must say but would rather not discuss.

The risk factors are also more pointed than before, warning that as the book grows, Ethos "may experience larger and more frequent in-period adjustments to revenue" and has "less historical data" on the fast-growing agent channel.

On SBC and Dilution

Ethos granted employees shares whose vesting depended on the company going public. Under accounting rules, you don't record the expense until that event is "probable." So through 2025, Ethos recorded almost nothing — $7.96M — while disclosing in a footnote that the true cost, had the trigger been met, was $175.8M. When the IPO happened in January 2026, $181.7M hit the income statement at once, producing the $166.4M Q1 loss.

The profitable 2025 and the loss-making Q1 2026 are substantially the same economic event, recorded in different periods.

Where it stands now:

• Q2 SBC was $15.6M — a clean, normalized quarterly run rate. Annualized, roughly $62M against FY25's recorded $10.6M.
• Remaining unrecognized: $32.1M on RSUs and $56.2M on performance shares.
• The structure repeats. On July 8, 2026, the CEO and President together received 1.8M time-based shares and 4.8M performance shares. The performance shares vest on Adjusted EBITDA growth targets for 2027 and 2028, paying out anywhere from 0% to 200% of target. Ethos expects to recognize only $6.6M of expense on all 6.6M shares in Q3.

Note what that means: because the payout is uncertain, the accounting again recognizes a small fraction of the potential cost up front. This is entirely legitimate — it is how performance awards work. But it is the same shape as the earlier situation: real compensation whose recorded cost lags well behind its potential value, with the difference disclosed in footnotes rather than in headline earnings.

Also note the grant is more shareholder-friendly. The Form 4 data available in July showed "900,000 options each" and the reports read it as a plain re-grant after a sale. The 10-Q reveals the larger part is performance-based, tied to multi-year EBITDA growth with a market condition — a considerably better alignment structure than a straight option grant.

On "Adjusted EBITDA": this is the measure the market watches, and it excludes stock compensation entirely. Q2 Adjusted EBITDA of $35.2M is calculated after adding back $15.5M of stock compensation. It is a useful measure of cash operating performance; it is not a measure of profit to shareholders, because shares issued to employees are a real cost borne through ownership dilution. The CEO and President's new awards vest on Adjusted EBITDA targets — a metric that excludes the cost of the awards themselves.

On the marketing treadmill

Sales and marketing was $128.1M in Q2 on $189.6M of revenue — 68%. That is down from 75% in Q1 but well above FY25's 59%.

The counterweight, and it is a real one: the company generates cash. First-half operating cash flow was $66.9M, up from $24.0M a year earlier, despite a $146.9M accounting loss. Q2 alone produced $35.7M. Management's claim that policies are "variable cash flow positive within 60 days" is consistent with this.

So the treadmill concern has softened. Ethos is spending heavily to grow, but it is not burning cash to do it.

One item in the cash flow statement is worth flagging: accounts payable rose $25.4M in the half, "due primarily to increase in expected commissions to be paid to carriers as a result of clawbacks" — money owed back to carriers for policies that lapsed. Clawbacks are rising with the book. Expected at this growth rate, but it is the cash-side echo of the persistency question.

On Governance

• Dual-class shares. Class B carries 20 votes per share. The two founders hold about 38.5% of voting power. Sequoia-affiliated entities hold about 35.7%, Accel about 21.2%. Together, well over 75%.
• The "Lead Independent Director" is Roelof Botha of Sequoia — the largest single voting bloc.
• No dividend. The buyback is the first capital return of any kind.

For a public shareholder this means: none of the concerns in this dossier can be changed by shareholder pressure. There is no realistic path by which outside holders influence carrier strategy, compensation design, or disclosure practice. You are a passenger.

On insider selling

The lockup expired July 27. Since July 24, Form 4 filings show approximately $29.7M of gross insider sales, at prices between $18.88 and $35.44:

Separately, Accel distributed roughly 2.5M shares to its limited partners in August, and Sequoia distributed several million more. These are not open-market sales — the shares transfer to the funds' investors — but they convert a locked-up institutional block into shares held by many parties who may sell independently. That is a supply overhang that does not appear as insider selling.

Reading it fairly: selling after a lockup expires, into a stock that has doubled, is normal and expected. Nobody sold their whole position; Colis still holds about 1.54M shares directly. The item that stands out is the Chief Accounting Officer, who reduced his holding from roughly 184,000 to 92,000 shares — about half — in the weeks after the print he prepared. That is not evidence of anything improper. It is simply the kind of fact worth noticing when the central question about a company is the quality of its accounting estimates.

Also worth noting on the other side: in August all seven outside directors received share grants, and Roelof Botha's holdings increased.

Part 5 — What to watch in Q3 (reporting ~November)

W1. Contribution margin — the single most important number. As reported: 42% → 30% → 33%. On gross profit less S&M: 39.1% → 23.7% → 30.3%. If Q3 prints 33% or below again (30% strict), the compression is permanent and FY25's 42% was the anomaly. If it recovers above 36%, the mix explanation was right. Watch the add-back line as well as the margin — the reported figure improved from FY24 to FY25 purely because allocated overhead and stock compensation added back to S&M more than doubled, while the underlying margin was flat at 39.1%. A reported recovery driven by bigger add-backs is not a recovery.

W2. Carrier A's share of revenue. 54% → 38% → 43%. If it rises again toward the mid-40s, "diversification" is not happening and the concentration risk is compounding. Also watch the receivable concentration — one carrier at 67% is the more extreme figure.

W3. The persistency reserve liability. $4.9M → $21.9M → $43.2M. Continued growth means Ethos keeps holding back cash it could recognize — conservative, and reassuring. A decline while agent-compensation expense rises is the pattern that would signal the cushion is being consumed to absorb losses. That remains the clearest early-warning indicator in the filings.

W4. Whether persistency is discussed at all. It was absent from the Q2 call. If Q3 passes with no management commentary on the estimate that governs revenue recognition, that is a disclosure choice, and a consistent one.

W5. Actual buyback execution. The authorization is $100M. Watch the cash flow statement's financing section for shares actually repurchased. An authorization that goes unused for two quarters at these prices tells you management does not consider the stock cheap.

W6. Revenue against the Q3 guide of $160–164M. Two consecutive 30%+ beats against fresh guidance is a pattern. A third would say the company genuinely cannot forecast itself — which cuts both ways.

W7. Direct-channel revenue against its own seasonal pattern. The Q1→Q2 decline was seasonal (−20% this year, −18% last). The test in Q3 is the year-over-year rate: direct grew 131% in Q2. If that decelerates sharply while third-party keeps compounding, the business is becoming an agent-distribution company rather than the direct-to-consumer platform the equity story describes — a lower-ARPU, lower-margin, more intermediated mix. Watch alongside W1.

W8. The commissions receivable balance versus activations. $324.3M and growing. If it grows materially slower than activations, Ethos is quietly reducing how much future revenue it books per policy — a persistency downgrade visible in the balance sheet before it appears anywhere else.

Things we don't know:

• Persistency is never quantified. No actual percentage, no cohort table, no estimate-versus-realized comparison. The most important assumption in the business is undisclosed.
• Carriers are unnamed in the concentration data. "Carrier A" at 43% of revenue is not identified. The six carriers are named in the business description, but not matched to revenue.
• No customer acquisition cost or payback period. There is no way to judge marketing efficiency directly; contribution margin is the only proxy.
• No product-level breakdown. Term versus whole life versus IUL revenue is not disclosed, so the "product mix" explanation for the ARPU decline cannot be verified.
• Agent productivity is not disclosed quarterly. The 15,000 agent count is as of December 31, 2025.
• No proxy statement exists yet. The first DEF 14A is expected ahead of the 2027 annual meeting. Compensation and governance detail currently comes only from the 10-K and the IPO prospectus.
• Only one annual report exists. There is no multi-year audited history to test any of this against, and no data covering a recession.

Part 6 — Valuation

The stock is trading near its all-time high, up 115% in six weeks, at roughly 19 times a forward EBITDA figure that excludes about $62M of annualized stock compensation.

Residual Operating Income valuation

Penman’s enterprise-side ReOI model, run off Net Operating Assets rather than earnings. Valuation date 30 June 2026. A trunk of shared facts and forecasts, then two branches that differ in one assumption.

Method: Stephen H. Penman, Financial Statement Analysis and Security Valuation — the Residual Operating Income (ReOI) model (Ch. 13–15), driven by Net Operating Assets and operating income after tax rather than by earnings.

Valuation date: 30 June 2026 (the latest balance sheet).

Prepared: 2026-09-05.

Sources: 10-Q Q2'26 (period ended 2026-06-30, filed 2026-08-05); FY2025 10-K; 8-K/earnings release 2026-08-03 (FY26 guidance); price $39.85 at 2026-09-04.

Why this model and not residual earnings. Penman's equity-side residual earnings model needs positive, stable earnings and book value. Ethos has neither: the six months to 30 June 2026 show a GAAP net loss of $146.9M, and the equity book value was negative at 31 December 2025 before the IPO converted the preferred stock. The enterprise-side ReOI model sidesteps both problems. It values the operations off Net Operating Assets — which are positive, stable, and unaffected by the capital structure — and then adds back the financial assets.

Summary

Everything in this valuation is shared right up to one question, and then it forks. The structure below follows that: a trunk of facts and forecasts common to both answers, then two branches that differ in a single assumption.

The fork: at the end of the forecast (FY2031) the model has Ethos earning a 40.5% return on operating assets against a 12% cost of capital — a 28.5 percentage-point spread. Does that spread persist, or does competition erode it?

Branch B is the more sensible because of erosion of abnormal returns due to competition and experience. Branch A is shown because it is a simplified construction and because seeing the two side by side is the clearest way to understand how much of any growth-company valuation is the terminal assumption rather than the analysis.

Three findings that survive whichever branch you pick:

1. The market price is not reachable from this operating forecast under either branch. Justifying $39.85 requires a terminal after-tax operating margin of 14.2% under Branch A or 24.1% under Branch B, against 7.2% today. The disagreement with the market is about margins, not about the tail. 2. 81% of the market capitalization is a forecast. Book value is $7.25 per share. Of the $2.54bn market cap (63,759k shares outstanding × $39.85), $0.23bn is net financial assets and $0.24bn is net operating assets; the remaining $2.06bn is the present value of residual income that does not yet exist. 3. Stock compensation is worth $8–14 per share of the disagreement. Run the identical forecast with SBC added back — the "Adjusted EBITDA" worldview — and the value rises 43–48%.

PART I — THE TRUNK

1. Why this model

Residual earnings model needs positive, stable earnings and book value. Ethos has neither: the six months to 30 June 2026 show a GAAP net loss of $146.9M, and equity book value was negative at 31 December 2025 before the IPO converted the preferred stock.

The enterprise-side ReOI model sidesteps both problems. It values the operations of Net Operating Assets — positive, stable, and unaffected by capital structure — then adds back the financial assets.

2. Reformulating the balance sheet

Penman's first move: split the balance sheet into operating items (the business) and financial items (the financing).

Common equity ≡ NOA − NFO, or here, since Ethos has more financial assets than obligations, CSE = NOA + NFA.

At 30 June 2026 ($ thousands):

The same exercise at 31 December 2025 gives NOA 246,705 and NFA 133,176, tying to CSE + preferred of 379,881. Both reformulations reconcile exactly to the reported balance sheet.

Three things to notice.

A1. Net operating assets went down while revenue doubled. NOA fell from $246.7M to $241.5M over six months in which revenue grew 108% year over year. Operating liabilities grew faster than operating assets. This is a genuinely capital-light business, and the model will reward it.

A2. NOA is three-quarters commissions receivable. $324.3M of the $443.2M of operating assets is money Ethos has recorded as earned but not collected — and that figure is itself the output of the persistency estimate. The capital base in this valuation is an accounting estimate, not a factory.

A3. The persistency reserve liability is quietly flattering the returns. It grew from $4.9M to $43.2M in six months. Because it is an operating liability, it subtracts from NOA — so the faster Ethos defers revenue it has already been paid in cash, the smaller its capital base and the higher its measured return on it. Prudent accounting; mechanically flattering ratio.

3. Reformulating the income statement

RNOA = NOPAT ÷ NOA, and RNOA = PM × ATO (profit margin × asset turnover)

RNOA barely moved — about 22% in both periods. But the composition inverted. Margin halved; turnover doubled. Ethos is no longer earning its return by being profitable per dollar of revenue. It is earning it by needing almost no capital.

Two caveats:

• FY2025's 14.1% margin is not real. It reflects only $10.6M of recognised stock compensation, when the company's own footnote put the true FY25 RSU cost at $175.8M. The H1 2026 figure of 7.2% — which carries a normalised SBC load — is the first honest read of this company's operating margin.
• H1 2026 is normalised by adding back the $181.7M one-off IPO vesting charge. Ongoing SBC ($15.5M in Q2 alone) is left in as the expense it is.

4. The explicit forecast: H2 2026 – FY2031

The near-term forecast is the company's own guidance, not mine. FY26 revenue $727–731M and Adjusted EBITDA $119–123M. H1 actuals were revenue $382.7M and Adj. EBITDA $68.8M, so guidance implies for H2 2026: revenue $346.3M and Adjusted EBITDA $52.2M. Converting to operating income means subtracting what Adjusted EBITDA excludes — roughly $44M of ongoing stock compensation (Q2 ran $15.5M; the new founder awards add $6.6M in Q3 alone) and $3.1M of depreciation.

H2 2026 operating income, pre-tax: $5.1M on $346M of revenue. That is management's own guidance with the cost of paying employees in shares put back in. On a full-cost basis Ethos is presently running at roughly breakeven at the operating line — which is why the model's first forecast period produces negative residual operating income.

Assumptions: revenue growth +35/30/25/20/15%; after-tax operating margin rising 6.0% → 11.0% as marketing leverages and SBC grows slower than revenue; asset turnover 3.0× → 3.2×.

5. Common inputs

Cost of capital. r_F is the required return on operations. Ethos has no debt and net financial assets of $234.6M, which are 9.2% of market capitalization. Using market weights, r_F = 12% implies a cost of equity of 11.3% — internally consistent for a recently-listed, founder-controlled, single-country company with concentrated customers. (Using book weights would imply an 8.3% cost of equity, which would be wrong; the equity trades at 5.3× book, so book weights badly distort the leverage relation.)

Share count basis. Two different counts appear in this note, deliberately. Aggregate market figures — market capitalization, enterprise value, price/book — use the 63,759k shares outstanding at 30 June 2026, because that is what the company is actually capitalized at. Per-share valuation output uses the 65,645k diluted weighted-average count, the more conservative divisor, with basic and fully-loaded counts shown alongside it in Branch A. Per-share comparisons against the $39.85 market price are like-for-like on either basis.

Tax at 25%. Historic effective rates were 6.1% (FY25) and 9.5% (FY24), reflecting NOL carryforwards. Using 25% throughout is deliberately conservative and ignores the real, if temporary, value of those carryforwards.

Stock compensation is expensed, in both branches. This is where most of the disagreement with sell-side valuations lives, so it is worth isolating. Run the identical forecast with SBC added back:

Penman is unambiguous that adding it back is not legitimate: shares issued to employees are value transferred from existing shareholders. The relevant figures: ongoing SBC ran $15.5M in Q2 alone (~$62M annualized, against $10.6M recognised in all of FY2025), with $88.3M already unrecognized on existing awards and a 2026 plan reserve of 18,962,000 shares — worth roughly $756M at today's price — still to be granted against. The $100M buyback authorized on 2 August offsets about 13% of that reserve, and no shares had been repurchased as of the filing.

Other choices: all cash and investments treated as financial (no operating-cash carve-out); operating leases treated as operating, netting the ROU asset against the lease liability — immaterial at $1.8M either way; the H2 2026 stub discounted at half a year.

6. What the trunk settles

Before any terminal assumption is made:

$11.97 is the floor of the whole exercise — the actual balance sheet plus five and a half years of forecast, with nothing assumed about the world beyond 2031. Everything above that number is a claim about the terminal period.

PART II — THE BRANCH POINT

At the end of the forecast, the model has Ethos earning RNOA of 40.5% against a 12% cost of capital: a 28.5 percentage-point spread, on NOA of $600M, producing ReOI of $170.8M in FY2031.

The only remaining question is what happens to that spread. Because

ReOI = (RNOA − r_F) × NOA

the terminal assumption is not really about growth. It is about whether the spread survives.

This is worth stating carefully, because the conventional phrasing disguises it. A continuing-value growth rate of 4% can be described two ways, and they are the same assumption:

• a growth rate similar to general nationwide economic growth (3–5%), which is why it is used as a reference — it is hard to grow forever above the average;
• Ethos earning RNOA of 40.5% every year, forever, a spread of 28+ percentage points over its cost of capital, sustained in perpetuity.

The first sounds unobjectionable. The second is a strong claim about competitive advantage. Choosing 4% asserts both.

The formal fork. Let the spread erode at a constant proportional rate after 2031, while NOA keeps growing at 4%. The honest way to state that rate is the one thing an analyst can actually form a view about: the spread half-life — how many years it takes for the 28.5-point advantage to fall by half.

A half-life of H years means the spread halves every H years and never quite reaches zero. At Branch B's 6.6 years, starting from 28.5 points at the end of 2031:

The advantage decays but never formally dies. What dies is its contribution to value: a shrinking stream discounted at 12% has a finite present value, which is the whole point of the continuing-value formula.

Mechanically, a half-life of H years is an annual survival rate of φ = 0.5^1/H — Penman's persistence factor, which is how the textbook states the same assumption. ReOI then grows at φ × 1.04 − 1, and:

CV_2031 = ReOI_2031 × (1 + g_ReOI) / (r_F − g_ReOI)

• half-life = ∞ → the spread never fades → Branch A (φ = 1.00)
• half-life finite → the spread erodes → Branch B (φ < 1.00)
• half-life = 0 → the spread vanishes at the horizon → Penman's Case 1, CV = 0, and the answer is the trunk's $11.97

Why years rather than φ. They are the same assumption, but only one of them is a belief a person can hold. On the persistence factor, 0.90 and 0.98 look like neighbours — a rounding difference. They are a 6.6-year advantage and a 34-year one, and $19.07 a share against $26.08. The scale is severely non-linear: every hundredth closer to 1.00 buys disproportionately more years, so the entire plausible range of competitive durability is compressed into the second decimal place, where nobody has a calibrated view. Stated as a half-life, the question becomes "seven years, or twenty?" — which is answerable from what is known about carriers, agent economics and third-party data vendors. The arithmetic below is unchanged; only the dial is.

PART III — BRANCH A: the spread persists

Assumption: the spread never fades — an infinite half-life (φ = 1.00). Ethos sustains a 28.5-point spread over its cost of capital in perpetuity, on a capital base growing 4% a year. Penman's Case 3 with g = 4%.

68% of the value sits in the continuing value — the period beyond 2031, about which nothing is known.

Scenarios

Sensitivity ($/diluted share)

Even here, the market price is only reachable at a cost of capital of 10–11% — i.e. by assuming the operations of a two-year-old public company, with one carrier at 43% of revenue and revenue recognition resting on an undisclosed estimate, are about as risky as the market as a whole.

PART IV — BRANCH B: the spread fades

Assumption: a 6.6-year spread half-life (φ = 0.90). Half of the 28.5-point advantage is gone by 2038 and three-quarters of it by 2044, while NOA keeps growing 4%. Competition gradually erodes the advantage without eliminating it.

Read the third column first. It restates each half-life as the claim it actually makes about the business: the year Ethos’ return on net operating assets finally drops back through 25%, from the 40.5% the 2031 forecast lands at. Branch B says 2038 — seven years of above-25% returns after the explicit forecast ends, then a slow decline toward the 12% cost of capital. The 13.5-year row says 2046. Branch A says the number never arrives at all. Stated that way the rows stop being a column of coefficients and become a set of competitive forecasts, each of which can be argued with on the evidence.

The result is not sensitive in a comfortable way. Even an extremely generous fade — a 13.5-year half-life, the advantage still half intact in 2045 — cuts the value to $22.41. And the family has a hard ceiling: an advantage that never fades at all is Branch A's $30.11. No fade assumption of any duration reaches the market price, because the longest one there is still falls $9.74 short.

Scenarios, both branches side by side

Note what happens to the bull case. Under permanent persistence it is $69.31 — nearly double the market price, and the number that made the market look defensible. Under a fade it is $32.92, still below the market. The bull case does not survive the terminal assumption changing.

What Penman's Case 1 assumes (the zero half-life endpoint)

Worth stating, because it is routinely misread. Case 1 requires one thing: RNOA = r_F after the horizon. Equivalently, the operations are worth exactly their book value at the horizon.

It does not assume the company shrinks, fails, or stops growing — NOA, revenue and NOPAT can compound indefinitely; growth simply becomes value-neutral, because each incremental dollar invested returns exactly one dollar of value. Nor is it the floor: ReOI can go negative if RNOA falls below the cost of capital. Case 1 is competitive equilibrium, not disaster.

PART V — CHOOSING BETWEEN THEM

Which branch

Branch B, for three independent reasons.

Penman puts the burden of proof on persistence. Case 1 is the default — abnormal returns attract entry, entry competes them away. Cases 2 and 3 require affirmative justification of an advantage that survives indefinitely. Branch A quietly treats permanence as the neutral choice; it is not.

The risk analysis in this folder already answered the question. My previous analysis in July concluded that Ethos is "more likely to gradually mean revert toward average economics than to remain a durable high-return compounder," because the protection is behavioural — carriers prefer to outsource distribution — rather than contractual, IP-locked, or regulatory. Branch A silently contradicts that conclusion.

The specific moat is narrow. A genuine underwriting and distribution platform, but built on third-party data (Milliman, TransUnion, MIB) that any competitor or carrier can buy, with six carriers and one at 43% of revenue and 67% of receivables. That profile does not support permanent persistence. Nor does it support immediate collapse. A half-life of roughly seven years is the defensible middle — long enough to respect a real platform, short enough to respect six carriers, buyable data and no contractual lock.

What the market must assume

Inverting the model to extract the forecast embedded in $39.85:

Under Branch A the market needs the operating margin to roughly double. Under Branch B it needs it to more than triple, to a pre-tax 32% — a level no insurance distributor earns after charging stock compensation.

There is a second way to see the same thing, and stating it in years makes it sharper. No fade of any length reaches $39.85. The best case the whole family allows is an advantage that never erodes at all, and that is Branch A's $30.11. Solving for what the market price actually requires returns a persistence factor of φ = 1.026 — which is not a half-life at all but a doubling time: a 28.5-point spread that widens 2.6% a year, reaching 57 points around 2058 and compounding from there. That is not a coherent steady state.

So the market price cannot be reached from this operating forecast under any economically sensible terminal assumption. The disagreement is therefore not about the tail at all. It is about whether the margin recovers — which is a forecastable, near-term, observable question, and a far better one to argue about than a perpetuity.

Where the model is weakest

M1. The capital base is itself an estimate. 73% of operating assets is commissions receivable — the present value of commissions Ethos expects to collect over a decade, discounted by an undisclosed persistency assumption. If that assumption is cut, NOA falls and NOPAT falls. Note the perverse mechanic: a lower NOA reduces the capital charge and raises ReOI, partly masking the damage in the arithmetic. A persistency downgrade would hurt the value more than it hurts the model's headline output.

M2. The terminal margin is the whole answer and it is unobserved. Even under Branch B, 46% of the value sits in continuing value, resting on an 11% after-tax operating margin the company has never earned on an honest cost basis. The best available evidence — contribution margin at 33% versus 42% a year earlier, a decline that persists even though the higher-margin direct channel gained 4.7% of share year-over-year — points the other way in the near term.

M3. The fade is applied only to the tail. The explicit forecast still has RNOA climbing from 3.2% to 40.5% between 2026 and 2031 on the strength of the assumed margin recovery. Fading RNOA inside the forecast period as well would lower both branches further.

M4. Guidance has been unreliable in both directions. Management beat its own three-month-old Q2 revenue guide by 61%. The FY26 anchor here is that same management's FY26 guidance. If they can miss by 61% on a quarter, the annual figure carries wide error bars — which cuts both ways, and is a reason to weight the range above the point estimate.

Conclusion

Ethos is worth about $19 per diluted share against a market price of $39.85 — a 52% discount — on a plausible range of $12 to $22. Branch A's $30.11 is the same analysis with a terminal assumption that Penman, and the risk work in this folder, both reject.

What the model establishes regardless of branch:

• The business genuinely earns about 22% on net operating assets, above any reasonable cost of capital. It creates value. That is not in question.
• But it earns that return through asset turnover, not margin — turnover doubled to 3.1× while margin halved to 7.2%. A very capital-light business running at a thin profit.
• $11.97 is what the balance sheet and 5.5 years of forecast are worth. Everything above that is a claim about a period nobody can observe.
• To justify $39.85, the market must believe the operating margin doubles (Branch A) or triples (Branch B) from 7.2%.
• About $8–14 of the price depends on treating stock compensation as free.

The most defensible summary is a condition, not a target. Ethos becomes worth today's price only if the operating margin recovers substantially from where it has actually been running for the last two quarters. Every piece of evidence in the Q2 filing on that specific question points down; the revenue evidence points emphatically up. The valuation is a bet on which resolves first — and the terminal assumption, which usually dominates arguments like this, turns out not to be where the disagreement lives.

Part 7 — What-If Financial Simulator

Penman Residual Operating Income (ReOI) interactive model · Enterprise side · Driven from contribution margin · Balance sheet as of 30 June 2026

Part 8 — Monte Carlo Simulation

Balance sheet 30 June 2026 · Residual Operating Income (ReOI) under uncertainty · Part 6 valuation run over probability distributions