Contents
The Three Pricing Models, and When Each One Is Correct
Every AI agent engagement collapses into one of three shapes. Picking the wrong one is how builders end up doing $8,000 of work for $1,200.
Model 1: Setup fee plus monthly retainer. The default, and the right answer for most people reading this. The setup fee covers the build, the integration work, the testing, and the first two weeks of tuning. The retainer covers hosting, model costs, monitoring, and ongoing changes. Typical shape for a small business agent: $1,500 to $6,000 setup, $300 to $1,200 per month. For a multi-workflow build touching several systems: $6,000 to $15,000 setup, $1,000 to $2,500 per month. This model works because the setup fee funds your working time and the retainer turns one sale into a durable revenue line.
Model 2: Retainer only, no setup fee. You absorb the build cost and recover it over the contract term. Typical shape: $800 to $3,000 per month with a six or twelve month minimum. This lowers the barrier for a hesitant buyer and can close deals that a $5,000 upfront number would kill. It is also how you go broke if the client cancels in month two, so it requires a minimum term in writing and it requires you to have enough cash to float the build. Do not use this for your first three clients.
Model 3: Per-outcome or performance pricing. You get paid per booked appointment, per recovered cart, per qualified lead, or a share of measured savings. Typical shape: a small monthly base plus $15 to $75 per qualified outcome depending on what that outcome is worth to the business. This is the highest-upside model and the most dangerous one. It only works when the outcome is unambiguously attributable to your agent, when the client's own systems can measure it, and when the client's downstream process is good enough not to waste what you deliver. If you send fifty qualified leads and their sales team ignores forty, you get blamed for the result and paid for nothing. Ask to see their close rate before you agree to it.
A fourth thing people try, hourly billing, is not on this list. Hourly punishes you for being fast, invites the client to audit your time instead of your results, and caps your income at your calendar. If a client insists on hourly, quote a rate high enough that you would be happy to be wrong about the estimate.
What to Anchor the Price Against
The most common pricing mistake is anchoring on your own effort. You built the agent in eleven hours, so you charge for eleven hours, so you charge $900 for something that will save the client thousands a month. The client is delighted, you are broke, and you have now set a price expectation you cannot escape.
Anchor on one of these instead, in order of preference:
- The salary line. "You are hiring a scheduler at $42,000 plus payroll tax and benefits. The agent handles most of that volume for $4,500 to start and $600 a month." The comparison is not close, and the buyer does the math themselves. This is the strongest frame available and it is why our hiring versus agent calculator exists.
- The revenue currently being lost. "You miss roughly a quarter of your inbound calls, your average job is $480, and you told me most missed callers just call the next company. Put your own number on that and compare it to $700 a month."
- The outsourcing bill. If they already pay an answering service or an offshore VA team, that invoice is your ceiling reference and it is public knowledge to them.
- The software line, last and reluctantly. If the buyer's mental category is "software", your price ceiling collapses to what SaaS costs. Fight to stay out of that category.
Run the arithmetic before the call, not during it. Our ROI calculator produces a payback period you can state as a sentence, and the AI automation cost breakdown is a useful sanity check that your quote is inside the range buyers already see elsewhere. The agency pricing calculator turns your own cost base into a floor you should not go under.
Your Real Costs, So the Retainer Is Not a Slow Loss
A retainer that does not cover your costs is worse than no retainer, because you are contractually obliged to keep losing. Before you set the monthly number, write down what running the agent actually costs you every month:
- Model API usage. The one that surprises people. A moderately busy support agent with a decent context window can run anywhere from a few dollars to a few hundred dollars a month depending on volume, context size, and model choice. Measure it in your test period rather than guessing, and read what an AI agent costs to run monthly before you commit to a flat fee.
- Automation platform seats. n8n cloud, Make operations, or your own hosting.
- Voice and messaging. Per-minute and per-message costs are real and they scale with the client's success, not yours.
- Vector storage, logging, and monitoring. Small individually, not zero collectively.
- Your time. Budget two to four hours a month per client for monitoring, small fixes, and the check-in call. If your retainer divided by those hours is less than you would accept hourly, the retainer is too low.
Protect yourself against volume spikes. Write a usage band into the agreement: the retainer covers up to a stated number of conversations, minutes, or runs per month, and beyond that a stated overage rate applies. Without this clause, a client's marketing campaign becomes your unpaid API bill. State it plainly and buyers accept it without argument, because every phone system and messaging platform they already pay for works the same way.
Scoping So Revisions Do Not Destroy Your Margin
This is where AI agent projects actually lose money. Not on price, on scope. The build is done, the client says "can it also handle...", you say yes because you want them happy, and six weeks later you are doing your third rebuild for the original fee.
Scope by counting objects, not by describing outcomes. "An AI agent that handles customer inquiries" is not a scope, it is a wish. A scope is a count:
- Intents: the number of distinct question or request types the agent handles. Name them in the agreement. "Twenty intents, listed in Appendix A."
- Integrations: the number of external systems it reads from or writes to. Name each one and the direction. Two integrations is a very different build from five.
- Channels: web chat, SMS, email, voice. Each channel is its own testing surface. Adding voice to a text agent is not a tweak.
- Languages, if more than one.
- Revision rounds: "Two rounds of revisions within the defined scope, during the first thirty days." After that, change orders.
Write the change order path into the original agreement, not into an awkward later email. One line does it: "Work outside the scope above is quoted separately at $X per additional intent and $Y per additional integration." Now "can it also do..." becomes a sales opportunity instead of a confrontation. Clients rarely object, because the alternative they are imagining is you saying no.
Get the discovery details in writing before you quote. A structured audit is the difference between a scope and a guess. Our automation audit template and the client discovery call script capture process, volume, systems, and owner in one pass, and the proposal template turns that into a document with a defensible number on it.
What the First Three Clients Realistically Pay
Setting expectations, because the numbers people post online are the exception rather than the pattern.
Client one is usually underpriced and you should let it be. Often $500 to $2,000 for a build you will spend far too long on, sometimes for someone you already know. You are not buying revenue here, you are buying a reference, a testimonial, a screenshot, and proof to yourself that you can deliver. The one thing worth protecting even at this price is the retainer: even $200 a month establishes that the agent is a service and not a one-time file.
Client two is where you charge what the work is worth for the first time, usually $2,000 to $5,000 setup plus $400 to $800 a month. You have a demo now, which changes the conversation completely. The build is also faster because you are reusing most of client one's architecture, which is the argument for staying in one vertical.
Client three is where the model starts working: $3,000 to $8,000 setup plus $600 to $1,500 a month, and increasingly they came to you rather than the other way around. Three retainers in the $600 to $1,200 range is a real monthly floor, and it changes how you negotiate, because you can now afford to lose a deal.
The pattern to watch for: if client three costs you as much delivery time as client one, you are taking on projects that are too different from each other. Repeatability is the whole business. Ten variations of one workflow beats ten bespoke builds at the same price every time.
The sales conversation that gets you to these numbers is covered in how to sell AI agents to businesses, and the first-client mechanics are in how to get your first AI agent client. When a prospect pushes back on the number, the specific responses are in AI agent sales objections and how to answer them.
Presenting the Price So It Gets a Yes
How you present a number changes whether it lands, independent of the number itself.
Offer three options, not one. A single price is a yes or no question. Three tiers turn it into a which question. Structure them so the middle one is what you actually want to sell: a reduced scope option, the recommended build, and an expanded option with an extra channel or workflow. The expensive tier does most of its work simply by existing, because it makes the middle look measured.
Put the payback on the first page. "Setup $4,500, $600 monthly. Based on the numbers you gave me, this pays for itself in roughly seven weeks." The buyer was going to do that arithmetic anyway. Doing it for them, using their own figures, removes the pause where deals die.
Phase the payment rather than discounting. Fifty percent to start and fifty percent on go-live costs you nothing and reads as confidence. Cutting the price to close reads as an admission the first number was invented.
Never discount without removing scope. If you drop the price, drop an integration or a channel with it. A discount for the same work teaches the client that your prices are negotiable fiction, and every renewal thereafter starts from the discounted number.
If you want the full agency-level version of this, including offer design and delivery systems, the AI Agency Bootcamp covers it end to end, and our own automation services page is a working example of how an offer like this reads to a buyer. For a broader look at packaging an agency's whole service menu rather than one engagement, see the AI automation agency pricing guide.
FAQ
- How much should I charge for an AI agent?
- For a small business agent handling one workflow, a common structure is $1,500 to $6,000 as a setup fee plus $300 to $1,200 per month for hosting, model costs, monitoring, and ongoing changes. For a multi-workflow build touching several systems, $6,000 to $15,000 setup plus $1,000 to $2,500 monthly is a normal range. Price against what the client currently pays for the problem, whether that is a salary, an outsourcing invoice, or lost revenue, rather than against the hours the build took you.
- Should I charge a setup fee or just a monthly retainer?
- Charge both unless you have a specific reason not to. The setup fee funds the build so you are not financing the client's project out of your own pocket, and the retainer turns one sale into recurring revenue. Retainer-only with no setup fee lowers the barrier for a hesitant buyer, but it requires a written minimum term of six or twelve months and enough cash to float the build. Do not use retainer-only for your first few clients.
- Is performance-based pricing a good idea for AI agents?
- It has the highest upside and the highest risk. It works only when the outcome is unambiguously attributable to your agent, the client's systems can measure it, and their downstream process is good enough not to waste what you deliver. If you generate qualified leads and their sales team does not follow up, you get blamed for a bad result and paid nothing. Ask to see their current close rate and response times before agreeing, and always pair it with a monthly base fee that covers your running costs.
- How do I stop clients asking for endless revisions?
- Scope by counting concrete objects rather than describing outcomes. Specify the number of intents the agent handles, the exact integrations by name and direction, the channels, and the number of revision rounds included within the first thirty days. Then write the change order path into the original agreement, for example a stated price per additional intent and per additional integration. Once the path exists, an out-of-scope request becomes a new quote rather than an argument.
- What if the client says the price is too high?
- Do not discount for the same scope. Either restate the comparison, since the agent is usually competing against a salary or lost revenue rather than software, or remove scope to match the budget by cutting an integration or a channel and phasing it into a later stage. Phasing the payment, typically half at kickoff and half at go-live, resolves a lot of price resistance that was really cash flow resistance. Discounting without removing scope teaches the client that every future number is negotiable.
- How do I handle API costs that vary with usage?
- Write a usage band into the agreement. The retainer covers up to a stated number of conversations, voice minutes, or workflow runs per month, with a stated overage rate above that. Measure the actual cost during your test period instead of estimating it. Without this clause a client's marketing campaign becomes your unpaid model bill, and buyers accept the structure without argument because their phone and messaging providers already bill them the same way.
If this resonated
Module 02 is the long version of this post.
Eval set, refusal pattern, kill switch. Build it in a weekend.
