AI SaaS Meta Ads: Set Acquisition Targets After Inference and Usage Costs
Two AI SaaS customers can pay you the same amount and leave completely different money behind. One generates a few outputs. The other runs expensive jobs all day and needs support. If you give them the same acquisition allowance because their revenue matches, your Meta ROAS can look healthy while one cohort loses money.
I would set the allowance from contribution after delivery costs, measured over an explicit recovery window. Then I would check whether your advertising is bringing in customers who actually meet it. Start with the product economics before changing the bidding strategy.
Build the cost stack per paying customer
Use net revenue for the period after discounts and refunds, excluding sales tax. Subtract inference or API usage, variable hosting, storage, payment fees, support and onboarding costs attributable to serving that cohort. Include failed jobs and retries if you still pay for them. Do not assume the successful output is the only billable work.
Keep refund deductions consistent. If net revenue already excludes a refund, do not subtract the refund again as a cost. Track payment fees that remain charged separately. Fixed salaries and overhead sit outside this contribution subtotal; the acquisition target must leave room to cover them.
Use your own invoices and metering, rather than a headline model price. A customer can consume several models, tools and storage services in one workflow. Allocate shared variable costs using a stated rule, and retain an unallocated bucket where the evidence is incomplete.
Compare equal revenue with different usage
Every number below is hypothetical. The example covers one month and assumes $100 net revenue per customer after refunds. It illustrates the cost difference; it does not describe a client or current API pricing.
| Monthly amount per customer | Light usage | Heavy usage |
|---|---|---|
| Net revenue | $100 | $100 |
| Inference and API costs | $5 | $45 |
| Variable hosting and storage | $2 | $12 |
| Variable support | $3 | $15 |
| Payment fees | $3 | $3 |
| Contribution before acquisition | $87 | $25 |
| Contribution after $30 acquisition cost | $57 | −$5 |
Both customers show $100 of revenue against $30 of acquisition spend: about 3.33 times revenue divided by spend. But the heavy-usage customer leaves a $5 loss before fixed overhead in this month. The light-usage customer leaves $57.
If you need to retain $20 per new customer during month one, the acquisition allowances are $67 and $5 respectively. Those amounts are contribution minus the retained amount. They are internal operating targets, not guaranteed prices at which Meta can acquire those customers.
Choose the recovery window explicitly
A month-one contribution calculation is not lifetime value. To allow a longer payback period, sum contribution from the original acquisition cohort at the same age, including customers who cancel or stop paying. Do not calculate only among survivors; that removes the customers whose acquisition costs still need recovery.
Annual upfront collections also need care. Receiving twelve months of cash does not remove the obligation to serve the customer for twelve months. Reserve expected delivery costs and refund exposure before treating the collection as available acquisition budget.
Use the SaaS cohort payback worksheet to separate expected contribution from observed recovery. Stress-test heavier usage, weaker renewal and a changed model mix. If the acquisition target only works under the most favorable assumptions, I would keep the next budget increase limited.
Connect the economics to account decisions
Compare acquisition source, offer, plan and cohort age in your own reporting. Heavy usage is not automatically bad: it can accompany higher retention or expansion revenue. The question is whether that cohort's contribution supports its acquisition cost over your chosen window.
Do not assume you can identify a prospect's eventual usage before acquisition or create a separate campaign for every cost bucket. First test whether observable offer and product choices attract a better economic mix. A pricing or allowance problem may need a product change, rather than another audience exclusion.
Keep the business target separate from the optimization event. A signup, an activated trial and a paying customer answer different questions. Use the optimization-event comparison before changing the signal. Sending an event alone does not establish that the campaign is optimizing for that outcome or for your internal contribution calculation.
Bring the costs and the account together
For my anonymous real estate AI SaaS client, monthly Meta spend moved from $17,852.93 in December to $148,907.37 in May. Purchase CPA finished $8.40 lower. I owned the structure, creative direction, tracking and scaling decisions.
If you already spend at least $15,000 a month on Meta, bring your acquisition data and usage costs to me. See my Meta ads service for SaaS and schedule a call. We can identify which assumption needs fixing before more budget goes through the account.