How do you model CAC for usage-based pricing when you have no upfront contract value in 2027?

The outcome you should expect

The first thing to set is expectations, because the single biggest cause of bad decisions in a usage-based book is a leadership team measuring a consumption business with a seat-based ruler and then acting on the gap. When you switch from contract-value CAC to run-rate-ARR cohort CAC, the number you report will get *worse* before anyone understands why it got *better*. That is the expected outcome, and you should pre-brief it.

Concretely, here is what a correctly-built model produces. Land-CAC payback — the months to recover only the cost of winning the logo and its first dollar of usage — typically lands in the 6-to-12-month range for an enterprise consumption product with a real sales motion. Fully-loaded payback, which layers in the customer success managers, solutions architects, and expansion campaigns that actually drive consumption upward, lands in the 15-to-25-month range. Those two numbers will be four to eight months apart, and the gap is not an error. The gap is the expansion investment, and the expansion investment is the entire reason cohort revenue triples or quintuples between month one and month twenty-four.

You should also expect the model to reverse the sign of several judgments you currently hold. Accounts that look expensive at signature — long proofs-of-concept, heavy sales engineering, a six-figure Land-CAC — frequently turn out to be the best assets in the book because they ramp steepest. Accounts that look cheap, especially self-serve signups that converted on a credit card, frequently turn out to be worthless because they never cross an activation threshold. A blended CAC computed across both motions averages a great business and a bad one into a mediocre number that describes neither.

Expect the model to change *where you spend*, not just how you report. Once you can see that Land-CAC payback is eight months and Expansion-CAC payback is twenty-two months on a much larger revenue base, the headcount conversation stops being "do we hire another AE or another CSM" argued by whoever has the louder VP, and becomes an arithmetic comparison of two payback curves against the same capital. That is the practical deliverable. Everything else — the triangles, the curves, the crossover tables — exists to make that one comparison honest.

Finally, expect the model to give you an early-warning instrument you did not have before. When every cohort is tracked against a canonical maturation curve, an account or cohort tracking materially below curve at month nine is a customer-success trigger, not a renewal-quarter surprise. In a seat-based business you find out at renewal. In a consumption business you find out in the metering data within weeks, which is one of the underrated structural advantages of usage-based pricing: your revenue *is* your product-usage telemetry.

Why classic CAC breaks the moment the contract stops promising a number

The textbook formula every SaaS operator memorizes is S&M spend for the period, divided by new ARR for the period times gross margin, times twelve. Buried inside it is a load-bearing assumption: the deal you signed has a known annual value the instant ink hits paper. Fifty seats at $1,200 per seat per year is $60,000 of ARR on day one. You can divide by it. The denominator is real, knowable, and stable from the moment of signature.

Usage-based pricing detonates that assumption. A consumption contract may specify a price per credit, per gigabyte, per API call, per message, and it may or may not carry a capacity reservation — but the *realized* revenue depends entirely on how much the customer actually runs over the following year. In the purest case, a developer puts a credit card down and pays per unit with no commitment whatsoever. The new ARR of that deal at signature is somewhere between zero and "we will find out." You cannot divide by a number you do not have.

Operators who force usage-based deals into the seat-based formula commit one of two symmetrical errors. The optimist's error uses the capacity reservation or the rep's forecast as new ARR. Rep consumption forecasts run systematically hot — plan on actual first-year consumption arriving well below the happy-path number, because reps forecast the customer's stated ambition, not the customer's migration velocity. That inflates the denominator, makes payback look fast, and sets the board up for a brutal miss two quarters later. The pessimist's error annualizes month-one realized revenue. A customer who signed on the 25th and ran two test queries produces a denominator near zero, payback computes as infinite, finance panics, marketing budget gets cut, and a perfectly healthy land motion gets strangled by its own reporting.

Both errors trace to one root cause: in usage-based pricing, the value of a customer is a function of time, not a property of the contract. That single sentence is the whole reframe. Everything downstream — cohorts, run-rate ARR, crossover payback — is just bookkeeping that respects it.

There is a third, quieter breakage worth naming: the customer count itself becomes unreliable. Seat-based SaaS has a clean unit. Consumption pricing fuzzes it. Is a developer who signed up, burned four dollars of usage, and disappeared a "customer acquired"? Technically yes; economically no. If you compute CAC as spend divided by logos, a freemium funnel converting thousands of hobbyists shows a gorgeous CAC and a worthless book. The discipline mature consumption companies adopt is a qualified-cohort definition: a customer enters the CAC denominator only after crossing an activation threshold — first production workload, first month above a revenue floor, first invoice that clears a minimum. Everything below that line is funnel, not revenue, and it belongs in a conversion-rate dashboard rather than a CAC calculation.

What drives that outcome

Three mechanisms drive the numbers above, and they compound. Understanding them in order is what lets you defend the model rather than merely produce it.

Mechanism one: the denominator swap. Run-rate ARR replaces contract value. The standard construction takes a trailing 90-day window of realized, invoiced usage revenue, averages it to a clean monthly figure, and annualizes: sum the metered revenue for months M-2 through M, divide by three, multiply by twelve. The 90-day window is the consensus default because it smooths the spikiness inherent in consumption — a customer who runs one enormous batch job and then goes quiet should not whipsaw your ARR. A 30-day window gives faster signal and catches a ramp early but is genuinely noisy; a six-month window is board-friendly and extremely smooth but slow to reflect a usage cliff or a churn event. Ninety days balances both and aligns cleanly to a fiscal quarter, which matters more than it sounds when you are reconciling to a close calendar.

Mechanism two: the spend split. In a seat-based world you can survive one blended CAC, because the acquisition motion dominates and the renewal motion is cheap. In a consumption world the expansion motion is where most of the revenue comes from, and it carries a real, large cost. If an account lands at modest run-rate ARR and matures to several multiples of that, the large majority of eventual revenue is expansion revenue. Blending the cost of winning the logo with the cost of the solutions architects and CSMs who drove that multiple gives you a single number that is true of nothing — it overstates the cost of landing and understates the cost of growing, in exactly the proportions that make budget allocation wrong.

Land-CAC covers new-business AEs (fully loaded, including the OTE portion tied to new logos), the SDR and BDR function, demand-generation marketing and the martech stack to the extent it produces new logos, pre-sales engineering hours burned on the proof-of-concept before signature, and the marketplace listing fee on the landing transaction. Expansion-CAC covers customer success managers — who in a consumption model are a revenue role, not a support role — post-sales solutions architects who light up new use cases, account managers owning commercial expansion, lifecycle and in-product growth marketing, and, if you are sophisticated about it, an allocated slice of the product investment that exists specifically to create more billable surface area. Most companies leave that last one in R&D; allocating a portion is defensible but you must disclose it and then never quietly change the allocation.

Mechanism three: the cohort lens. Every customer enters the model in a cohort defined by signature month. You stop looking at "all customers" as an undifferentiated mass and start watching cohorts mature in parallel. Per cohort, per month, you track: cohort size in qualified customers, Land-CAC attributed to that month's signings, cumulative run-rate ARR summed across members, cumulative gross profit, logo retention, and net revenue retention measured against a month-three base rather than month one (month one is too noisy to anchor to).

The artifact all three mechanisms produce is the cohort maturation curve: months-since-signature on the x-axis, run-rate ARR indexed to month one on the y-axis, averaged across your mature cohorts. That curve *is* your model. Once it exists, any new cohort's month-one run-rate ARR projects forward against it, which gives you forecast revenue, LTV, and payback in one move. Rebuild it quarterly, because product changes, pricing changes, and customer-mix shifts all bend it.

The companion display is the cohort triangle: rows are signature months, columns are months-since-signature, cells are run-rate ARR. Reading across a row shows one cohort's life. Reading *down a column* is the underrated move — it shows how a given maturity stage trends across vintages. If your month-six number is bigger for each successive cohort, your land motion is improving even if nothing else in the dashboard says so. If that column is flat or declining while total revenue grows, your top of funnel is quietly degrading and total revenue is masking it. That column read is the single most useful diagnostic in consumption-model RevOps, and almost nobody looks at it.

Benchmarks and realistic ranges

Benchmarks in this space are directional, not precise, and the honest framing is that they are ranges practitioners recognize rather than published constants. Use them to sanity-check, never to justify.

Payback by model shape. Seat-based SMB SaaS recovers fastest — revenue is flat from month one, so cost recovery is linear and short. Seat-based enterprise runs longer on bigger deals and longer cycles. Usage-based self-serve can look deceptively fast because Land-CAC is near zero, but churn drag eats the advantage. Usage-based enterprise consumption is the long case, and the long case is fine: the revenue is back-loaded by construction, so the recovery is too. Add an AI-inference or communications-carrier COGS profile and payback stretches further still, because payback is computed on *gross profit*, and a fifteen-point margin haircut arithmetically pushes a sixteen-month payback out several months without anything operationally changing.

The rule to internalize: a usage-based enterprise payback of twenty months is not worse than a seat-based payback of fourteen months. It is a different revenue shape. By month thirty-six the consumption cohort is throwing off multiples more gross profit because retention keeps compounding. Payback measures *speed of recovery*, not *quality of the asset*. Pair it with LTV/CAC and NRR before drawing any conclusion at all.

Gross margin by archetype. This is where consumption pricing diverges hardest from classic SaaS, and where naive models break. Data-warehouse and analytics platforms run in the seventies because cloud compute and storage sit in COGS. Observability and monitoring tend to sit at the healthier end of that band. Communications APIs — SMS, voice — carry carrier costs that compress gross margin dramatically, often into the fifties or low sixties. AI and LLM inference APIs land in a similar compressed band because GPU inference is genuinely, structurally expensive. CDN and infrastructure businesses sit in the seventies with margin that improves as they scale bandwidth commitments. Model your own COGS curve alongside your revenue curve; do not import a SaaS gross-margin assumption into a consumption P&L.

Retention ranges. Net revenue retention is not a vanity metric in this model — it is the return mechanism. In a seat-based business, NRR above 100% is a pleasant bonus. In a consumption business, NRR above 100% is the entire investment thesis, because a long payback is only acceptable if the cohort keeps expanding for years afterward. The consumption companies that scaled durably did so on sustained NRR meaningfully above 100%, and they built their entire investor narrative around that number precisely because it is what converts a slow payback into an excellent LTV. If your consumption book is running at or below 100% NRR, the cohort math does not work and no amount of denominator engineering will fix it — you have a product-adoption problem wearing a finance costume.

Quality bars. LTV over Land-CAC should be comfortably high — the land motion ought to be very efficient in a model where expansion does the heavy lifting. LTV over fully-loaded CAC should still clear the familiar 3x bar; the bar survives the transition to usage-based pricing, but the *time horizon* over which you earn it stretches. A seat-based company might hit 3x inside thirty months; a consumption enterprise company might need forty to fifty. Same destination, longer road, bigger cohort at the end. Burn multiple — net burn over net-new run-rate ARR — is the metric that ties all of this back to cash runway, and it is the one that keeps a beautiful long-payback story honest when the bank balance disagrees.

A word on the denominator windows. If you are comparing your numbers to anyone else's, confirm they use the same trailing window and the same qualification threshold. Two companies can report payback numbers six months apart purely from window choice and activation-floor definitions, with identical underlying economics. This is the most common apples-to-oranges failure in consumption benchmarking and it is almost never disclosed.

Risks, edge cases, and failure modes

The accounting layer distorts the number before you touch it. Under revenue-recognition rules for variable consideration, you recognize usage revenue as the usage occurs, which conveniently aligns with what the model wants — realized metered amounts. But commission capitalization cuts the other way. Rules requiring you to capitalize incremental costs of obtaining a contract and amortize them over the expected benefit period mean your GAAP S&M expense in any quarter is *amortized* commission cost, not *cash* commission cost. Build CAC off the GAAP P&L and you are using a smoothed number that runs faster-looking than cash reality in any fast-growing quarter. Both bases are legitimate; report which one you used, in a footnote, every single time. "Payback fourteen months, cash-commission basis" and "payback nineteen months, GAAP-amortized basis" can be simultaneously true of the same company. The consumption twist: the amortization period depends on expected customer life, and expected customer life comes from the retention and maturation curves you built — so your unit-economics model and your accounting policy are coupled, and a change to one should trigger a review of the other.

Marketplace fees quietly eat margin. A large and growing share of consumption revenue flows through the cloud marketplaces, which take a percentage of the transaction. Treat the fee on the landing transaction as a Land-CAC component; treat the ongoing fee on expansion transactions as either Expansion-CAC or a reduction of revenue, depending on policy. Ignore them entirely and you overstate gross margin and understate CAC simultaneously — a double error in the same direction. On a book where a large fraction of revenue routes through marketplaces, this is worth real points of blended gross margin, and gross profit is the numerator of payback.

Mixing motions in one cohort. A company running both a self-serve funnel and an enterprise sales motion that lumps both into one monthly cohort produces a meaningless blended CAC and a meaningless blended curve. The two motions have Land-CAC figures that differ by orders of magnitude, completely different ramps, and different retention profiles. The diagnostic is a bimodal distribution of cohort run-rate ARR — if the histogram has two humps, you have two businesses. Segment cohorts by motion at minimum, and by segment on top of that where volume allows.

Letting the curve go stale. The canonical maturation curve is built from history, and history ages. A curve built eighteen months ago, before a pricing change and two new product lines, will misforecast today's cohorts confidently and quietly. Refresh quarterly and watch whether recent cohorts systematically beat or miss it; a persistent one-directional gap means the curve needs rebuilding, not that the cohorts are anomalous.

Confusing revenue growth with gross-profit growth. On an inference-heavy or carrier-cost-heavy book, an account's run-rate ARR can grow while gross profit stalls or shrinks, because the customer shifted usage toward a lower-margin product tier or you cut unit prices to drive volume. Since payback runs on gross profit, track a gross-profit maturation curve alongside the revenue curve. Teams that only chart revenue miss this until the margin line in the board deck does something inexplicable.

Hyper-volatile and seasonal usage. If customers' consumption swings with their own business cycles — a tax product that spikes in one quarter, a commerce infrastructure product that triples on a single shopping weekend — even a 90-day trailing window misrepresents the account. You need a seasonally-adjusted run-rate ARR or a trailing-twelve-month base, and you should be cautious comparing cohorts that landed in different seasons. The apparatus still applies; the denominator definition needs a seasonality correction first.

Tiny-n false precision. Cohort modeling needs enough cohorts with enough history. Fourteen customers and a four-month-old oldest cohort is not a maturation curve; it is a set of anecdotes wearing a chart. At that stage, track accounts individually, talk to customers, and wait for six-plus months of history across several cohorts before trusting anything the curve tells you.

The counter-case: committed-spend-forward contracts. If your "usage-based" contract is really a drawdown commitment — the customer commits to a spend figure and meters against the balance — you are much closer to a seat-based deal than a pure-consumption one. That commitment is a real contract value; you can divide by it. Apply classic CAC payback to the committed floor with a modest haircut for the portion of commitment that customers historically fail to consume, and reserve the full cohort machinery for the uncommitted overage. Similarly, if your product has genuinely flat consumption from day one — a utility-style API where customers hit steady state within a month — there is no ramp, so there is no maturation curve to build. Month-one run-rate ARR is a fine denominator and classic payback applies directly. The cohort apparatus exists to handle the ramp; no ramp, no apparatus.

The board-comparison failure. The most common senior-level error is comparing a consumption enterprise payback against a seat-based SMB comparable and concluding the consumption business is inferior. Benchmark like against like, and always present payback alongside NRR and LTV/CAC so the back-loaded shape is visible in the same eyeful. Lead the CAC section of the deck with the maturation curve and the cohort triangle, then show payback — presented in that order, the long payback reads as the natural consequence of a compounding revenue model rather than as a problem to be fixed.

A practical rollout plan

You do not need a data platform to start. You need a spreadsheet, a billing export, and the discipline to hold definitions constant. Here is the sequence.

Weeks one and two — define and qualify. Write down, in one document, the definition of a qualified customer (first invoice above a floor, or first production workload), the run-rate ARR window you will use, and exactly where the Land/Expansion boundary sits by role. This document is the most valuable artifact in the whole project and the one teams skip. Assign definitional authority to one person in strategic finance. Everything else can be distributed; definitions cannot be, because the fastest way to destroy board trust is to redefine run-rate ARR between meetings so trend lines stop meaning anything.

Weeks three and four — instrument the denominator. Nothing above works if you cannot trust the usage data. You need event-level capture with a customer ID and timestamp on every billable unit, idempotency so retries and replays do not double-count, a rating engine that turns events into dollars per the customer's actual price plan including tiers and credits, and a monthly reconciliation that ties metered events to billed revenue with any variance over a small threshold investigated. If run-rate ARR is built on metered events that drift from what you invoice, every cohort number downstream is fiction dressed as precision. Early-stage companies commonly run metering-to-invoice on their payments platform's usage-billing features and graduate to a dedicated metering layer as plan complexity outgrows it.

Weeks five and six — attribute the numerator. Map new-business S&M to signature months and expansion S&M plus CS cost to the accounts (and therefore cohorts) those teams own. Marketing attribution here is imperfect and always will be; directional is sufficient and vastly better than a blend. Even a rough 70/30 split of a shared role's cost between Land and Expansion beats leaving it unallocated.

Weeks seven and eight — build the crossover. Assemble the per-cohort table: cumulative gross profit against cumulative Land-CAC, and separately against cumulative fully-loaded CAC. Find the month each crosses. Report both numbers, always paired, with the gap explained as expansion investment. For rigor, discount the monthly gross-profit stream at your cost of capital before finding the crossover — at meaningful rates this pushes payback out by a couple of months on a long-tail cohort. Most teams run undiscounted for the operating dashboard and discounted for strategic-finance reviews; either is fine, disclosed.

The ongoing monthly close. Pull metered revenue for the closed month. Recompute trailing-90-day run-rate ARR per customer and roll up per cohort. Add one new diagonal cell to the cohort triangle. Allocate the month's S&M and CS into Land and Expansion buckets. Advance every cohort's crossover table by one month and flag any cohort that crossed payback. Compare every active cohort against the canonical curve and flag anything tracking below tolerance as a customer-success intervention — this is the step that turns a finance report into an operating instrument. Publish the dashboard on frozen definitions.

The quarterly review. Rebuild the canonical curve from all mature cohorts. Read the triangle *columns* to see whether month-three, month-six, and month-twelve values are improving across vintages. Reconcile cash versus GAAP CAC and explain the gap out loud. Re-examine whether any role's mix has shifted enough to warrant reclassification across the Land/Expansion line — reclassify deliberately and disclose it, never drift silently. Then update board materials.

Ownership. Strategic finance owns the model, the definitions, and the narrative. RevOps owns spend attribution and the Land/Expansion split, which is genuinely the hardest recurring operational task in the whole system. Data engineering owns metering reconciliation and the cohort warehouse. Customer success leadership owns the off-curve intervention process. The CRO and CMO consume the split to allocate headcount and budget. The model fails when nobody owns the definitions — not when the math is wrong, which is rare, but when the meanings drift, which is constant.

The one-paragraph version for the board. We acquire customers at a low Land-CAC and a small initial run-rate ARR. Over the following twenty-four months each cohort's run-rate ARR grows several-fold as customers migrate workloads and adopt new use cases, driven by an expansion motion we fund deliberately. Land-CAC pays back around month eleven; fully-loaded CAC pays back in the mid-to-high teens; LTV-to-CAC clears three times by roughly month forty. The slower payback versus a seat-based comparable is a function of revenue shape, not efficiency — and that back-loaded shape is exactly what produces our net revenue retention well above one hundred percent and our compounding cohort value.

Related questions

Should we still report a single blended CAC to the board?

Yes, as an output — investors and comparables expect it. Compute it as total S&M plus CS spend over total net-new run-rate ARR from land and expansion combined. But make decisions off the split. The blended figure is for the scoreboard; the split is the playbook.

How do committed-spend contracts change the math?

A real drawdown commitment restores a usable contract value. Apply classic CAC payback to the committed floor, haircut for historical under-consumption, and reserve cohort machinery for the uncommitted overage. Hybrid books need both methods running side by side with a clearly disclosed boundary.

What if we have almost no sales team?

Pure product-led acquisition makes Land-CAC so small that measuring it precisely is a poor use of effort. Model free-to-paid conversion rate and cost-to-serve free users instead. The full apparatus returns the moment you layer an enterprise motion on top of the self-serve base.

How does this change sales compensation design?

Paying full commission on signature in a no-commitment model pays for optimism. Most consumption companies split commission between a signature component and consumption-triggered milestones, which also aligns the capitalized-commission amortization period with realized customer life.

Can we compute LTV before the curve matures?

Not credibly. LTV needs the maturation curve for the ramp years plus a steady-state estimate with churn and discounting for the tail. Using month-one run-rate ARR understates it badly; using mature run-rate ARR with no churn or discount overstates it just as badly.

FAQ

What is the single biggest mistake companies make modeling CAC for usage-based pricing?

Dividing total sales and marketing spend by logos or signed contracts. With no upfront commitment, revenue at signature is near zero, so the calculation returns either an absurd number or an infinite one. The fix is to wait for real consumption data and use trailing run-rate ARR as the denominator instead of anything the contract promised.

How exactly do you compute Land-CAC?

Total the new-business spend for the period — AE fully loaded cost including new-logo OTE, the SDR and BDR function, demand-generation marketing, pre-sales engineering for the proof-of-concept, and the marketplace fee on the landing transaction — then divide by the count of qualified new customers who signed that period. Divide that by month-one (or first-90-day average) run-rate ARR times gross margin to get payback in months.

Why 90 days for the trailing window rather than 30 or 180?

Thirty days catches a ramp fast but is noisy enough that one large batch job distorts the account's whole ARR. Six months is smooth and board-friendly but slow to reflect a usage cliff. Ninety days balances responsiveness against noise and aligns to the fiscal quarter, which simplifies reconciliation. Pick one and never change it mid-year.

Is a twenty-month CAC payback bad?

Not on a consumption book. It reflects back-loaded revenue, not inefficiency. Judge it alongside net revenue retention and LTV over fully-loaded CAC. A twenty-month payback with strongly negative churn beats a twelve-month payback with flat retention, because the first cohort keeps compounding for years and the second does not.

Do we need a dedicated metering platform to do this?

No, not to start. A billing export, a spreadsheet, and a firm qualified-customer definition are enough to build credible cohort tables. Dedicated metering and rating layers become necessary as plan complexity grows — tiers, credits, volume discounts, multiple meters — and as manual reconciliation stops being tractable.

Where does customer success cost belong?

In Expansion-CAC, not in cost of goods or general overhead. In a consumption model CSMs drive workload adoption, which is revenue creation, not support. Classifying them as overhead makes the land motion look worse than it is and hides the true cost of the expansion engine that carries the business.

Sources

• https://www.bvp.com/atlas — Bessemer Venture Partners Atlas, cloud and SaaS operating benchmarks including CAC payback
• https://www.growthunhinged.com/ — Kyle Poyar's writing on usage-based pricing and consumption economics
• https://openviewpartners.com/ — OpenView's product-led growth and usage-based pricing research library
• https://www.iconiqcapital.com/growth/insights — ICONIQ Growth operating benchmark reports for B2B software
• https://fasb.org/ — Financial Accounting Standards Board, source for ASC 606 and ASC 340-40 guidance
• https://investors.snowflake.com/ — Snowflake investor relations, reported net revenue retention on a consumption model
• https://investors.datadoghq.com/ — Datadog investor relations, land-and-expand consumption disclosures
• https://stripe.com/docs/billing/subscriptions/usage-based — Stripe documentation on usage-based and metered billing
• https://aws.amazon.com/marketplace/features/sellers — AWS Marketplace seller program and listing terms
• https://a16z.com/tag/enterprise/ — Andreessen Horowitz enterprise software economics writing

Related on PULSE

• [Explaining negative churn and expansion revenue to board auditors who think retention caps at 100%](/knowledge/q421)
• [What the Rule of 40 actually measures and when it misleads](/knowledge/q417)
• [Building a tiered partner program that rewards scale without collapsing margin](/knowledge/q429)
• [Deal-share compensation models that keep channel partners hungry](/knowledge/q430)
• [What specific metrics are B2B RevOps teams using to measure AI's impact on lead quality in the top-of-funnel?](/knowledge/q16717)
• [Why are longer sales cycles now correlating with a shift from pipeline velocity to deal value predictability?](/knowledge/q16718)