LTV Models: Where Founders Lie to Themselves

The LTV lifetime value formula founders use in SaaS is where the self-deception starts. The number on the pitch deck is not a measurement. It is a wish, dressed up as arithmetic, and every input has been quietly set to its most flattering value.

The mechanism is simple. Lifetime value runs on three inputs: how much a customer pays, how much of that you keep, and how long they stay. Founders systematically pick the optimistic version of all three. They use blended churn instead of early-cohort churn, revenue instead of gross profit, and a divide-by-a-tiny-number that turns a rounding error into a fortune.

A defensible LTV uses cohort churn, gross profit, and a capped horizon. It is always smaller than the deck’s number, often by half. This piece shows where each lie enters, what the honest version looks like, and how to build a model an investor cannot tear apart. The framing is analytical: how to think about the metric, not what to do about any company’s stock.

Key takeaways

• Every input is set to its rosiest value. The standard LTV = (ARPA × Gross Margin) ÷ Churn formula lets founders pick blended churn, revenue not gross profit, and an infinite horizon, each one inflating the result.
• Dividing by churn breaks at scale. David Skok notes the old formula “ceases to work properly” with long lifetimes or negative churn, where it can go infinite (ForEntrepreneurs, LTV analysis). A capped, discounted model is the fix.
• Revenue is not gross profit. A16z calls estimating LTV on revenue or gross margin “a common mistake”; it should be contribution margin (Andreessen Horowitz, “16 Startup Metrics”). Revenue-based LTV overstates by an estimated 20 to 40 percent.
• Retention is a moving target. Snowflake’s net revenue retention fell from 178 percent in FY2022 to 126 percent in FY2025 (Snowflake Forms 10-K). Using a peak cohort as a constant inflates every downstream number.
• The honest LTV is smaller, and that is the point. A model built on cohort churn, gross profit, and a finite horizon survives diligence. The deck number does not.

The LTV Honesty Checklist: a framework for the four lies

Most LTV arguments stay abstract. This one is built around a named, reusable asset: the LTV Honesty Checklist. Each row takes one input of the lifetime value formula, states the optimistic version a founder reaches for, the defensible version that survives diligence, and why the gap matters. Run any LTV model through these four rows and you find the self-deception in minutes.

The checklist is an original analytical framework. The figures used to illustrate it are sourced inline to the references cited throughout this piece.

LTV input	The optimistic version (the lie)	The defensible version	Why it matters
Churn rate	Blended churn across all customers, often a recent best month	Cohort churn measured over 12 to 24 months	Blended churn hides the months one-to-three spike; the curve is non-linear, not the constant the formula assumes
Margin basis	Revenue, or gross margin applied loosely	Gross profit, ideally contribution margin net of variable serving cost	Revenue-based LTV overstates by an estimated 20 to 40 percent; it counts dollars you never keep
Time horizon	Infinite (1 ÷ churn implies customers never fully leave)	Capped at three to five years	An infinite horizon and a tiny churn denominator manufacture lifetime value that no cohort will ever realize
Discounting	None; a dollar in year five counts as a dollar today	Future cash flows discounted at roughly 10 percent	Distant revenue is riskier and worth less now; ignoring the time value of money inflates the present value

Read top to bottom, the checklist explains why the deck number and the diligence number diverge. Each row independently inflates LTV. Stacked together, they can double or triple the figure a founder genuinely believes is conservative.

What is the standard LTV formula and why does it break?

The standard formula is LTV = (ARPA × Gross Margin) ÷ Churn Rate. It breaks because dividing by a small churn number produces unrealistic or infinite lifetimes, especially with negative churn or long horizons, and it assumes a constant decay that real cohorts do not show.

David Skok states the problem directly: “The old formula that everyone uses for customer lifetime value (LTV), average gross profit per customer divided by churn, ceases to work properly when you have very long customer lifetimes and negative churn. LTV can become infinite, which clearly doesn’t reflect reality” (ForEntrepreneurs, LTV analysis).

The math is the trap. If monthly churn is 1 percent, the implied lifetime is 1 ÷ 0.01, or 100 months. Shave churn to 0.5 percent and the implied lifetime doubles to 200 months without the business changing at all. A rounding error in the denominator becomes years of phantom revenue in the numerator. The same denominator sensitivity that distorts LTV also distorts the payback math covered in the CAC payback period, the SaaS metric that matters.

What is the difference between blended churn and cohort churn for LTV?

Blended churn averages new and tenured customers together, which hides the early-tenure spike. Cohort churn measured over 12 to 24 months reveals the truth: retention curves are non-linear, with high churn in months one to three, then stabilization. The standard formula assumes a constant linear decay that almost no cohort actually follows.

Christoph Janz of Point Nine Capital made this point years ago: your LTV “might be higher (or lower) than you think” precisely because simple formulas assume linear decay while real cohorts spike early and then flatten (Point Nine Capital, Medium). A founder who measures churn in month two of a cohort and annualizes it captures the worst part of the curve and projects it forever, or, more often, measures a flattering recent month and projects that.

ChartMogul’s recommended approach is cleaner than a single blended rate: ARPA divided by a trailing six-month average of customer churn rate (ChartMogul, LTV methodology). A worked example from their methodology: $100 ARPA divided by 5 percent churn yields a $2,000 LTV. That smoothing helps, but it still does not segment by cohort, and cohort segmentation is where the real distortion lives.

The honest move is to plot the retention curve by signup cohort and read the shape, not a single number. Two businesses with identical blended churn can have completely different lifetime values if one bleeds customers in the first quarter and the other loses them slowly across years. The retention-definition problem here is the same one that separates gross from net retention in public filings, dissected in gross retention vs net retention in SaaS IPOs.

Why is using revenue instead of gross profit in LTV calculations a founder lie?

Because revenue-based LTV ignores the cost of serving customers, inflating the result by an estimated 20 to 40 percent. LTV should reflect the gross profit dollars available to recoup acquisition cost, not every dollar the customer pays. Founders reach for revenue because it is the easy number to grab and the flattering one to show.

Andreessen Horowitz is blunt about it. In “16 Startup Metrics,” the firm warns that “a common mistake is estimating LTV as a present value of revenue or even gross margin, instead of calculating it based on net profit” of the customer relationship. Their recommended formula is contribution margin per month times the inverse of monthly churn (Andreessen Horowitz, “16 Startup Metrics”). Contribution margin nets out the variable cost of serving the account, not just cost of goods.

The margin basis is not a small adjustment, because gross margin itself varies enormously across SaaS. Adobe reported 89.3 percent gross margin in FY2025 (Adobe Form 10-K, FY2025), while Snowflake reported 67 percent in the same fiscal year (Snowflake Form 10-K, FY2025), a 22.3 point spread. Two companies with identical ARPA and identical churn but those two margin profiles have lifetime values that differ by roughly a third before any other input changes.

LTV component	Revenue-based (the lie)	Gross-profit-based (defensible)
ARPA	$100 / mo	$100 / mo
Margin applied	None (100% of revenue)	Gross profit only
At 89% margin (Adobe-like)	$100 counted	$89 counted
At 67% margin (Snowflake-like)	$100 counted	$67 counted

Margin figures: Adobe Inc. Form 10-K, FY2025; Snowflake Inc. Form 10-K, FY2025. ARPA is an illustrative round number, not from any filing.

The reason this lie does so much damage is the asymmetry of the cash flow. CAC is paid in full dollars and recovered only in margin dollars. Counting revenue you will spend on infrastructure as if it were lifetime value is double-counting. This is the same logic that makes margin the binding constraint on every downstream metric, the full argument in why gross margin is destiny in SaaS.

The horizon problem: why infinite lifespans inflate LTV

The third lie is the time horizon. The 1 ÷ churn term implies customers eventually all leave but contribute revenue across an effectively infinite tail, and it counts a dollar arriving in year eight as worth the same as a dollar today. Both assumptions inflate the present value.

Skok’s fix is a discounted cash flow model. Rather than dividing by churn, you project gross profit per cohort year by year, apply a growth or expansion rate, discount each future year at a rate that reflects risk and the time value of money (he recommends roughly 10 percent), and sum over a finite horizon rather than to infinity (ForEntrepreneurs, LTV analysis). The DCF accounts for “the risks associated with revenue that is far off in the future, and the time value of money.”

The practical effect is large. A customer paying for ten years contributes far less in present value than the simple formula credits, because years six through ten are both uncertain and discounted. Capping the horizon at three to five years, which is the realistic planning window for most software businesses, removes the phantom tail entirely.

Methodology: the defensible DCF-based LTV

For any computed LTV, the honest version uses this frame.

• Inputs: gross profit per customer per period (not revenue), a cohort-based retention curve measured over at least 12 months, an expansion or growth rate read from real net revenue retention, a discount rate, and a capped horizon.
• Assumptions: discount rate around 10 percent per Skok; horizon of three to five years; retention measured from actual cohorts, not a blended snapshot; gross margin from the most recent audited period.
• Sensitivity: LTV is most sensitive to the churn denominator and the horizon. A one-point change in monthly churn or a two-year change in horizon moves the result more than any plausible change in ARPA. Treat both as ranges.
• What this misses: a DCF still assumes the retention curve and margin observed today persist. If a cohort’s behavior shifts, or margin compresses as the product scales into heavier workloads, the model is stale. Re-run it per cohort, not once.

The deck LTV versus the honest LTV: a worked illustration

Here is the same hypothetical business modeled both ways. These numbers are illustrative and invented to show the mechanism; none are drawn from a filing. The point is the size of the gap, not the specific figures.

Input	Deck LTV (optimistic)	Honest LTV (defensible)
ARPA	$100 / mo	$100 / mo
Margin basis	Revenue (100%)	Gross profit (75%)
Monthly churn	1.5% blended (recent best month)	3% cohort (months 1 to 12 average)
Horizon	Infinite (1 ÷ churn)	5 years, capped
Discount rate	None	10% annual
Resulting LTV	~$6,667	~$2,300

The deck number is roughly $6,667: $100 divided by 1.5 percent monthly churn. The honest number is roughly $2,300 once you apply gross profit, real cohort churn, a five-year cap, and a 10 percent discount. Same business. The defensible LTV is about a third of the pitched one.

That ratio is not unusual. When a founder shows a clean, simple LTV, the diligence-grade version is commonly 40 to 50 percent smaller, sometimes more. The gap is not fraud. It is the accumulation of four reasonable-sounding optimistic choices, each defensible alone, devastating together.

Why the capped, gross-profit, cohort-churn LTV is the only one worth trusting

A capped-horizon, gross-profit, cohort-churn LTV is the only version worth trusting because it removes each of the four sources of inflation at once. It counts only the dollars you keep, only the customers who actually stay, only the years you can realistically forecast, and it values distant cash at what it is worth today.

Every other version is a strict overstatement. Revenue-based LTV overstates margin. Blended churn understates loss. An infinite horizon manufactures a tail. No discounting overvalues the future. Stack the four and you get the deck number. Remove the four and you get a figure an investor will accept and a budget you can actually run a business against.

The discipline matters most where LTV feeds another decision. LTV/CAC ratios, payback periods, and acquisition budgets all inherit the lie if the LTV is inflated. A 3:1 LTV/CAC built on a deck LTV can be a 1:1 in reality, which is the difference between a fundable business and one that burns cash on every customer. The way usage-based and seat-based pricing change the retention and expansion curve underneath these numbers is worked through in usage-based pricing vs seat-based pricing.

The bear case: when an optimistic LTV might still be justified

The strongest argument against this whole exercise is that conservatism can be its own distortion, and the skeptics get something real here. A defensible LTV that is too pessimistic talks a founder out of a genuinely strong business.

Negative churn is the clearest case. When a cohort’s expansion revenue exceeds the revenue lost to logo churn, net revenue retention runs above 100 percent and the cohort grows in value over time. A16z is explicit that gross churn “shows actual losses to the business, while net revenue churn understates the losses” by including expansion (Andreessen Horowitz, “16 Startup Metrics”). The flip side is that for a true expansion business, a model that uses gross churn and ignores expansion understates LTV. The optimistic founder who counts expansion is not always lying; sometimes the expansion is real and durable.

Platform and usage-based businesses complicate the horizon argument too. A customer embedded in an ecosystem with high switching costs may genuinely stay for a decade, in which case a five-year cap throws away real value. The economics of those embedded, hard-to-leave positions are the subject of Microsoft Copilot and enterprise lock-in, where the durability of the relationship is the entire thesis.

Here is the honest weighing. The four corrections are the right default because the errors they fix all run in the same direction, toward overstatement, and a number that survives diligence is worth more than one that wins a meeting. But the corrections are a discipline, not a verdict. A business with proven negative churn and a real moat can justify a longer horizon and an expansion-inclusive model, as long as the founder shows the cohort data that earns it. The lie is not optimism. The lie is optimism without the cohort evidence to back it.

What operators should take from this

If you build or read LTV models as a founder, operator, or analyst, the transferable discipline is not a single formula. It is the habit of distrusting every input and showing your work. Here is the playbook, six concrete moves.

1. Run the LTV Honesty Checklist on your own model first. Before an investor does it for you, check each of the four rows: cohort churn not blended, gross profit not revenue, capped horizon not infinite, discounted not nominal. Find your own lies before someone else finds them.
2. Show both numbers in the deck. Present the simple LTV and the DCF-based LTV side by side, and name the gap. A founder who volunteers that the defensible number is 40 to 50 percent smaller reads as more credible, not less.
3. Plot retention by cohort, never as a blend. Pull the curve for each signup cohort and look at the shape. The months one-to-three slope tells you more about the business than any annualized churn rate.
4. Use gross profit, then push toward contribution margin. Start with gross profit, then net out the variable cost of serving the account, support, infrastructure, payment processing, to get to the contribution margin A16z recommends. Count only dollars you keep.
5. Cap the horizon at the planning window. Three to five years is the honest range for most software. If you claim longer, attach the cohort data that proves customers actually stay that long.
6. Re-run the model per cohort, every quarter. LTV is not a constant you compute once. Snowflake’s NRR moving from 178 percent to 126 percent over three fiscal years (Snowflake Forms 10-K, FY2022 and FY2025) is the reminder that retention drifts. A model built on last year’s best cohort is already wrong.

The sixth move is the one most founders skip. LTV is a snapshot of a moving system, and the moment you treat it as fixed, it starts lying to you again.

Where this is vulnerable: expansion, negative churn, and the CAC payback trap

A framework this clean deserves its own counterexamples, because the honest LTV has blind spots too.

Expansion revenue is genuinely hard to model. Net revenue retention above 100 percent means the LTV is growing, but projecting how long expansion continues is guesswork. A model that caps the horizon at five years may undercount a business where seats and usage compound for a decade. The fix, more cohort history, is exactly what early-stage companies lack.

Negative churn can reverse the whole argument. For a true expansion business, the conservative gross-churn model understates value, sometimes badly. The same caution that makes blended churn a lie makes ignoring expansion a different kind of lie. The honest answer is to model both and disclose which cohorts actually expand.

The CAC payback trap hides inside the ratio. Even a perfect LTV is misleading if CAC is rising while LTV holds, because the LTV/CAC ratio can look stable while payback periods stretch and cash burns. LTV is a lifetime figure; payback is a cash-timing figure, and a business can have a healthy ratio and a dangerous payback at once. That timing distinction is the entire subject of the CAC payback period, the SaaS metric that matters.

Public-company retention is not your retention. It is tempting to anchor a model to a disclosed NRR like Snowflake’s, but those figures reflect mature, enterprise-heavy bases. An early-stage cohort churns differently, and borrowing a public benchmark is its own form of optimism.

None of this overturns the method. It bounds it. The capped, gross-profit, cohort-churn LTV is the right default because its errors are small and its honesty survives scrutiny. What it cannot do is replace the slow work of watching real cohorts behave over real time. That part does not scale, and it is the part that separates a model from a guess.

---

Analysis, not investment advice. Figures are drawn from public SEC filings cited inline by company and fiscal year (Adobe and Snowflake Forms 10-K) and from named methodology sources (David Skok / ForEntrepreneurs, Andreessen Horowitz, ChartMogul, Point Nine Capital). Illustrative LTV examples are labeled as hypothetical and use invented round numbers. Frameworks here are for understanding SaaS unit economics and tradeoffs, not for making buy or sell decisions.

Want the full toolkit for modeling unit economics like this, the LTV Honesty Checklist, the DCF-based LTV worksheet, and the cohort-churn template used above? It’s in the Tech Business Analysis Playbook.