AI SaaS customer acquisition cost: the formula is only the beginning
Customer acquisition cost looks simple: divide acquisition spend by new customers. The difficult work is deciding which costs belong in the numerator, which records count as customers and whether the time periods are comparable.
Willan Theunissen · 12 November 2026
For AI SaaS, there is another important distinction. Acquiring a customer and delivering the service are separate cost questions. Inference, infrastructure, support and payment costs can change the contribution available to recover acquisition spend.
This guide uses invented examples to explain the arithmetic. They are not industry benchmarks, forecasts, accounting advice or recommended conversion targets. The accompanying calculator lets you change the assumptions locally in your browser.
Define the unit before counting customers
A signup can represent a person, an account, a company, a seat or an API key. Those units are not interchangeable. A team with ten users may still be one commercial customer. A developer can create several keys without bringing several organisations into production.
Choose the customer definition appropriate to the business and use it consistently. For example, an acquired customer might be an account with an accepted paid subscription under an approved policy. A signed contract and a collected first payment may be different events, and reporting should preserve that distinction where it matters.
Trials, demos and opportunities are useful leading indicators. They should not be renamed customers because the business has not yet observed the later outcome.
Use more than one cost measure when it helps
Media-only CAC = media spend / acquired customers. This is useful for a narrow acquisition comparison but excludes other work.
Marketing CAC = media spend plus other included marketing costs / acquired customers. State which agency fees, creative, website allocations, tools or internal marketing costs are included.
Fully loaded CAC = included marketing and sales acquisition costs / acquired customers. Define the allocation policy and avoid counting the same cost twice.
The phrase fully loaded is only useful when the inclusions are documented. A calculation that excludes the sales team but includes advertising should not quietly become fully loaded CAC in the presentation.
Keep delivery costs outside the acquisition numerator
Hosting, inference, support and payment costs can affect customer contribution. They generally belong in the defined delivery-cost calculation rather than being automatically mixed into CAC.
There can be boundary questions. Trial support or free-trial compute may be treated differently from ongoing paid-customer support under the business’s chosen policy. The important requirements are consistency, disclosure and no double counting.
A sales salary allocated to acquisition must not also be deducted as the same delivery cost. A support allocation should be labelled as an allocation rather than pretending to be a directly measured per-customer variable cost.
Follow the actual conversion stages
For a sales-led illustration:
Sales-led
For a product-led illustration:
Product-led
For a developer-led illustration:
Developer-led
These are deliberately simplified models. A real business may have additional stages, parallel routes and different timing. The model’s value is that each rate has an explicit denominator. Do not apply a win rate measured from attended demos to all raw enquiries.
A worked sales-led example
Assume the following fictional inputs for one aligned acquisition cohort:
| Input | Illustration |
|---|---|
| Media spend | R60,000 |
| Other marketing costs | R40,000 |
| Allocated sales acquisition costs | R50,000 |
| Enquiries | 400 |
| Enquiry-to-booking rate | 25% |
| Booking-to-attendance rate | 75% |
| Attended-demo-to-customer rate | 20% |
Four hundred enquiries produce 100 expected bookings, 75 attended demos and 15 customers. Media-only CAC is R4,000. Marketing CAC is approximately R6,667. Fully loaded CAC is R10,000.
If the attendance stage were silently omitted, the model would imply twenty customers and a fully loaded CAC of R7,500. That would not be an improvement in performance; it would be a change in the formula.
Expected counts can be fractional in a planning model. Keep the full precision internally rather than rounding each stage before division. Observed customer counts are whole records and should be reported as such.
Why spend divided by new MRR is not payback
Consider a simplified planning illustration: R100,000 of marketing spend, twenty customers and R2,500 of monthly subscription revenue per customer. That produces R5,000 marketing CAC and R50,000 new monthly recurring revenue. Spend divided by new MRR is 2.0 times.
The ratio does not establish a two-month financial payback period. It excludes sales costs, delivery costs, churn, payment timing and other factors. It compares an acquisition-spend amount with a monthly revenue amount under the stated assumptions; it does not measure cumulative contribution recovery.
This distinction matters because a familiar-looking ratio can be interpreted more strongly than the evidence supports. Label the calculation, not the conclusion you hope the reader will draw.
Separate committed recurring revenue from variable usage
A fixed subscription can be normalised to a monthly recurring measure. Variable usage may depend on future behaviour and should not automatically be presented as committed recurring revenue.
Stripe, for example, excludes variable metered billing from its MRR calculation. That is an explicit provider convention, not a universal rule that every internal report must adopt without discussion. Stripe’s metered-billing explanation
The report should distinguish committed subscription amounts, estimated usage revenue, invoices, collected cash and accounting revenue. Different values can all be useful if their meaning is clear. The Stripe MRR and ARR explanation provides one reference point for recurring normalisation.
Calculate a contribution measure with explicit inclusions
Extend the fictional fifteen-customer example. Assume R2,500 monthly subscription revenue and R500 expected usage revenue per customer. Assume R600 inference, R100 hosting, R200 support allocation and R100 payment costs per customer per month.
Under these assumptions, monthly revenue is R3,000 and included delivery costs are R1,000. Contribution is R2,000 per customer, with a contribution margin of approximately 66.7% under this specific cost definition.
The fifteen customers therefore imply R37,500 committed new MRR and R7,500 estimated usage revenue. Combined illustrative monthly revenue is R45,000, but it should not all be labelled committed MRR.
This is not necessarily an accounting gross-margin definition. It is a planning contribution measure based on the costs included. Finance should approve how the business uses those labels.
Simple payback is a steady-state approximation
Simple payback
- Example: R10,000 ÷ R2,000 = 5 months
In the example, R10,000 divided by R2,000 produces five months. That assumes a stable monthly contribution and does not account for churn, expansion, ramp-up, timing or discounting.
If contribution is zero or negative, there is no finite positive simple payback under the assumptions. The correct output is a clear explanation, not zero months or a fabricated infinity displayed as a useful result.
A high-usage customer may generate more revenue and more delivery cost. The model should reflect both sides rather than assuming every extra usage rand has the same margin as a fixed subscription.
A retained-cohort view answers a different question
A simple cohort model can calculate the contribution remaining from the acquired group over time. In the reference calculator, the first month’s contribution uses the starting customer count; the chosen monthly churn rate then reduces the next month’s retained count. Expansion and reactivation are excluded.
With fifteen customers, R2,000 contribution each and 3% monthly churn, the cohort produces R30,000 in the first month and progressively less thereafter. Cumulative contribution first exceeds the R150,000 acquisition cost in month six, rather than the five-month no-churn approximation.
This still is not a financial forecast. It assumes a constant contribution per retained customer, regular monthly timing, no expansion and no changes in delivery cost. A more complex business may need actual cohort cash flows, variable usage and finance-led modelling.
If the chosen horizon ends before recovery, report that result honestly. Do not infer that recovery is impossible forever, and do not imply that an unobserved future recovery is certain.
Align the time windows
A monthly advertising bill and the customers acquired during that same calendar month may relate to different cohorts. Enterprise sales can close long after the first enquiry. Product-led customers may convert after a trial. Acquisition work can also affect renewals or expansion without being the sole cause.
Where possible, examine acquisition cohorts and their later outcomes. Report the cohort age and the proportion of records still unresolved. Separate a current-month operational view from a mature-cohort efficiency view.
When the data cannot support cohort attribution, disclose the limitation. A blended calendar-period ratio can still be useful, but it should not pretend to be a precise causal measurement of the latest campaign.
Be cautious with lifetime-value shortcuts
A formula based on stable margin and churn can be a useful approximation for a sufficiently stable business. It is much less dependable when the product is new, usage varies, pricing changes or customer cohorts behave differently.
Do not turn a zero observed churn rate in a small or young cohort into infinite lifetime value. Do not adopt a universal LTV-to-CAC target without considering the business model, cash requirements and uncertainty.
For early decisions, a clearly labelled contribution and retained-cohort view can be more informative than a highly precise lifetime-value figure built on weak assumptions. Bessemer’s AI pricing discussion is a useful practitioner reminder that pricing and delivery costs interact; its examples should not be copied as universal margin benchmarks. Bessemer AI pricing playbook
Turn the calculation into an operating decision
If CAC rises, identify which stage changed. Did the audience become less relevant, attendance fall, sales conversion weaken or attribution change? If contribution falls, inspect usage, infrastructure, support and pricing rather than asking marketing to solve every problem.
Use a shared metric dictionary and reconcile the source systems. A dashboard should show what is observed, what is estimated and what is missing. Retain ownership of finance definitions and avoid sending confidential product data to marketing platforms merely to make the dashboard easier to build.
Use the acquisition calculator to test the arithmetic, or explore a measurement foundation to connect the model to actual business records. The useful outcome is not the most attractive ratio. It is a decision supported by clear definitions and evidence.
Keep the right things connected.
Commercial analytics and revenue operations for AI companies
Make the commercial numbers agree before asking them to guide growth.
Paid acquisition for AI and SaaS companies
Use paid media to test a commercial hypothesis, not just buy activity.
AI SaaS acquisition and contribution calculator
Explore sales-led, product-led and developer-led acquisition assumptions, with separate CAC definitions, delivery costs and retained-cohort recovery.
Start a conversation
Tell us what you are working towards. We’ll help you find a practical next step.