How to monetize an AI agent, step by step.
A 30-day plan from zero to first revenue. No reliance on VC timing.
Day 0: pick your model.
The first decision is not pricing. It is which model to price against. Pick wrong and the next thirty days optimize a dead end. Pick right and every later step compounds. There are three paths, and the entry point is traffic shape.
If you have fewer than one thousand monthly active users and the intent behind each session is high — the user came on purpose, to do a specific task, and would notice if the product disappeared — go subscription. The math is simple. A thousand users paying $15 per month is $15,000 MRR. Two hundred fifty of them converting to paid at a 25% rate is $3,750 MRR. Either number is enough to tell you whether the category wants to pay. Subscription is also the cheapest to ship: Stripe Checkout plus a gate on the paid routes, and the billing problem is solved.
If you have more than ten thousand monthly active users and the intent is varied — some users come to experiment, some come for one answer, some come daily — go ads. At that scale, subscription is not going to convert high enough to cover the cost of carrying the free majority. Ads shift the question from “will they pay” to “will they show up,” and at ten thousand MAU, showing up is already solved. Surfacedd’s Surfaces pay a revenue share on disclosed sponsored content inside the agent’s output, so every visit contributes.
If the agent is transaction-native — it books travel, places orders, closes deals — go transaction fees. Take a percentage of the booking or a flat fee per closed purchase. The model lines up with the value the agent delivers, and the customer is paying for an outcome instead of access.
Do not mix models on day zero. One primary, one plan. You can layer later. Founders who launch with three revenue lines usually have zero working ones. For the longer treatment of which model fits which shape, read AI agent monetization.
Week 1: ship the simplest thing.
Week one is build-and-launch, not build-and-refine. The goal is the smallest payment surface that lets real money move. Everything else is noise until that exists.
For subscription: Stripe Checkout with two prices — monthly and annual — and a middleware gate on the paid routes. A user who is not signed in sees the pricing page. A user without an active subscription sees the upgrade prompt on the gated features. A user with an active subscription gets through. That is the whole system. Do not build a billing portal; Stripe ships one. Do not build seat management; add it when a customer asks. A week is enough for this if you stay disciplined.
For ads: install the Surfacedd SDK on the surface where the agent renders text. A Text Surface lives inside the output of the agent as a disclosed sponsor placement. It does not edit the organic answer. It does not interrupt the flow. It takes one API call to enable and roughly a day to review in staging. Once it is live, every agent response becomes a revenue opportunity; traffic you already had starts paying. See the developer guide to AI app monetization for the integration walk-through.
For transaction fees: do not build the integration yourself in week one. Partner with an existing affiliate program — Amazon Associates, Booking.com’s affiliate API, Shopify Collabs, a flight aggregator — and take the referral rate while you work on native plumbing. A 4% referral rate on real bookings beats a 15% take rate on vaporware. Native integration is a quarter of work; affiliate is an afternoon. Start with the afternoon version and collect data on which transactions actually close.
The rule for week one is that something billable has to exist by Friday. If your plan for Friday is “finish the onboarding,” you have picked the wrong thing. Onboarding optimizes a funnel that does not yet have a bottom.
Week 2-3: instrument.
Weeks two and three build the instruments that tell you whether the model is working. Without them, you are running on vibes, and vibes will tell you to keep going long after the numbers say stop.
Track five events. Conversion signals: sign-up, first paid action, upgrade, downgrade, cancel. Retention: a weekly active flag per user, a thirty-day retention cohort, a ninety-day retention cohort. Per-user revenue: gross revenue divided by monthly active users. Cost per task: API spend plus tool spend divided by the count of completed tasks. And a catch-all error rate so you know when a silent regression is eating conversion.
A dashboard is worth building if you plan to look at it daily. If you plan to look at it weekly, use an off-the-shelf analytics tool — PostHog, Mixpanel, Amplitude, or whatever your stack favors. The point is the data, not the dashboard. Founders who spend week two building a dashboard almost always ship week three’s decision one week late.
Set the thresholds up front, in writing, before you see the numbers. Write down what “this model is working” means. A sample set: 3% free-to-paid conversion within fourteen days, 70% month-one retention on paid, contribution margin above 60% per paid user, cost per task below 30% of per-task revenue. These are starting points, not universal truths; your category may demand different numbers. The discipline is the written-in-advance part, not the specific values. Without it, you will rationalize bad numbers into acceptable ones.
Add an annotation log. When you ship a change, note it. Pricing change, onboarding change, new model in production, SDK update — anything that could move a number. Three weeks from now you will look at a chart, see a dip, and need to know what caused it. The annotation log is how you answer the question without a forensic investigation.
Week 4: decide to scale or switch.
By the end of week four you have two numbers that matter: contribution margin per user and retention at the thirty-day mark. If both are above your thresholds, scale. If either is below, something has to change.
Scaling means pouring acquisition into the existing model. Paid ads if the math pays back within a reasonable window. SEO if the category has patient demand. Partnerships if there is a natural host product with the users you want. Do not rebuild the product in week five. A working model grows by adding users, not by adding features. Founders who answer “how do I double revenue” with “ship more features” are almost always wrong; the answer is usually “double the users.”
Switching means the model you picked is not the model that fits. This is the hardest part of the plan because the founder has now spent thirty days inside one model and is attached to it. The most common failure mode in this program is not bad numbers in week four. It is a founder who sees bad numbers in week four and refuses to act on them because switching feels like admitting the first three weeks were wasted. They were not wasted. They taught you which model does not fit. That lesson is worth the thirty days, and refusing to use it is the actual waste.
A switch is not a rewrite. Subscription to ads is an SDK install and a gate removal. Ads to transaction fees is a partner integration and a fee-take layer. Transaction fees to subscription is a Stripe plan and a gate. The plumbing is a week of work. The hard part is the decision, and the hard part is why most agents that should switch, do not.
Write the week four decision into week zero. Commit, in advance, to act on the numbers even if they hurt. See AI agent monetization platforms for which vendors make each switch easier.
When to switch models.
Switching before the thresholds trip is twitchy. Switching after the thresholds trip and nothing changes is the goal. Pick the triggers now, in the calm, before the week four stress.
Trigger one: thirty-day retention below 40% on paid users for two consecutive cohorts. Subscription requires retention, and at that level, the math does not close. Switch to ads if you have the traffic, or to transaction fees if the agent closes deals.
Trigger two: customer acquisition cost above sixty days of payback on blended channels for three consecutive weeks. If it costs more to acquire a user than the first two months of revenue, the model does not scale without venture subsidy. Switch to a model with lower acquisition requirements — ads scale with organic traffic, transaction fees scale with agent utility, subscription does not scale without paid acquisition.
Trigger three: API cost per task above 40% of per-task revenue. You will not earn your way out of that by optimizing prompts. Either raise price, reduce model tier on the hot path, or switch to a model that pays per task rather than per user.
Frequently asked questions.
How fast can a new AI agent reach first revenue?
Which model should I start with if I have no users?
Is it a mistake to launch a free tier on day one?
What metric proves a monetization model is working?
When should I bring in ads alongside a subscription?
Advertising for AI agents, built to be disclosed.
Join the waitlist. We are onboarding developers and advertisers in the order they sign up.