How Much Should You Charge to Build an AI Agent for a Client?

Manuel Suarez runs a digital marketing agency called Attention Grabbing Media.

A few years before anyone used the word "agentic", he started building Facebook Messenger chatbots for online personalities. The bots asked visitors simple questions. What's your current revenue? Do you have a YouTube channel? How many staff do you have? Then they turned the answers into cheap, qualified leads.

When I wrote about Manuel in Chapter Five of The Subscription Playbook, his business was generating about $200,000 per month in recurring revenue.

Notice that last word. Recurring.

Manuel didn't build a bot, send an invoice and move on. He built a business around keeping bots running. That's the difference between an agency that builds AI agents and an agency that gets rich from them.

So, how much should you charge to build an AI agent for a client? And more importantly, how do you make sure the client's token bill doesn't eat your profit? Let me show you.

How Much Should I Charge for an AI Agent?

Quote in three parts. Always.

Most agencies quote part one, mumble something about part two, and forget part three entirely. Then the agent gets popular, the model bill arrives on the agency's credit card, and the most successful project becomes the least profitable one.

(That's the AI version of a gym member who actually turns up every day.)

The Four Ways to Charge, From the Consulting Chapter of Feed a Starving Crowd

In the final chapter of Feed a Starving Crowd, I laid out a ladder for pricing consulting work. It maps almost perfectly onto agent builds.

Step 1: The free (or cheap) trial. When you have no references, offer to test your skills on a real business. Treat it as if you're being paid $1 million. For agents, that's a tightly scoped pilot: one workflow, one channel, a capped number of conversations and a capped model spend.

Step 2: A fixed fee for a defined scope. This is my preferred structure. The client knows the price, you know the job. The catch is that scope creeps. One of my first consulting gigs was marketing "driver fatigue alarms" for $1,500 a month, part-time. Part-time turned out to be basically full-time, and we parted ways after a month. Agent builds creep the same way ("can it also handle refunds?"), so write down what's in and what's out.

Step 3: A lower fee plus a percentage of results. You take less up front and share the upside. In the book, I said this only works when the business already sells its product, execution is the real problem, and you have the authority to make decisions. The same rules apply to an agent. A per-booking or per-lead fee only makes sense if the client's funnel works and the agent controls the step you're paid on.

Step 4: Partnership. Equity or revenue share for your work. Only consider it once you've done enough projects to know what an agent is really worth, and only with people whose values you share.

Most agent agencies should live on Step 2 for the build and add a Step 3 element to the retainer once they have data.

Worked Example: Quoting a Booking Agent for a Clinic

Let's price a real-looking job. The numbers below are illustrative. Use your own.

Now check it against the market. $2,790 a month for 2,000 inquiries is about $1.40 per inquiry. Fin charges $0.99 per resolved outcome. HubSpot charges $0.50 per resolved conversation. Zendesk charges $1.50 to $2.00 per automated resolution.

Prices checked 27 October 2026; they change often. See the AI Pricing Index.

So expect the question: "Why don't we just buy one of those?"

You need a good answer. For this job it might be: none of them books directly into the clinic's practice-management system, handles SMS the way the clinic wants, or comes with someone accountable when it gets a booking wrong. If you can't give an answer like that, you're about to lose the deal, and you probably should.

Model the numbers for your own job in the AI Token Cost Calculator before you send the quote.

When Should You Not Build a Custom Agent?

Here's an honest one. Sometimes the best advice you can give a client is "don't hire me to build this".

If the job is standard customer support on a standard helpdesk, products like Fin, Zendesk and HubSpot already do it, with published per-outcome prices and years of tuning behind them. (See the full AI agent pricing comparison for current rates.)

In that case, recommend the product. Then charge for what the product doesn't do: setup, connecting it to the client's systems, writing and cleaning the knowledge base, testing, staff training and monthly reviews.

You'll win more trust, and more referrals, than an agency that rebuilds what already exists and marks it up.

Build custom when the agent needs to work inside the client's own systems, follow rules unique to their business, or do a job no product does yet.

Who Pays for the Tokens?

Decide this before you sign. There are three clean options.

Option 1: The client's own API key

The client opens an account with the model vendor and you connect the agent to it. They see every cent. You never carry the risk.

This is the cleanest option for bigger clients. Clay does something similar in its own product: customers can bring their own API keys, which skips Clay's data credits, but they still pay Clay for the "actions" its platform performs. Copy that logic. The client pays the model vendor for tokens and pays you for the work.

Option 2: Pass-through at cost plus a stated margin

You pay the vendor and bill the client monthly, with a margin for the admin and risk. Say what the margin is. Clients can look up model prices in seconds, so hidden markups get found.

Option 3: Included, with a cap

You include usage in the retainer up to a limit (for example, 3,000 conversations a month), then charge per extra conversation. Simple for the client, risky for you if the agent's usage per conversation grows. Only offer it once you have a few months of real cost data.

Whichever you choose, add two clauses:

1. A price-review clause. Model prices change. Most fall, but not all: Google's Gemini 4 Argon is scheduled to go from $2/$10 to $4/$20 per million tokens on 1 December 2026, and Gemini 3.8 Flash doubles on 1 January 2027.
2. A spending cap and alert. Set a monthly ceiling with an alert at 80%. An agent stuck in a loop at 2 a.m. should hit a cap, not your profit.

Should You Charge per Outcome?

It's tempting. "You only pay when the agent books an appointment" is a lovely sentence to put in a proposal.

But think carefully about Step 3 of the ladder. You'll carry the cost of every failed conversation. You'll argue about what counts as "booked" (was the no-show a booking?). And you'll depend on parts of the client's business you don't control.

My advice: keep the retainer as your floor, and add a modest per-outcome bonus on top once you have three months of data showing what the agent actually achieves. That gives you upside without betting your payroll on it.

I learned that lesson the hard way with my first business, a "results-guaranteed" marketing agency, and I tell that story in the guide to outcome-based pricing.

How Do You Turn an Agent Build Into Recurring Revenue?

In Chapter One of The Subscription Playbook, I described the feast-or-famine cycle of project work. You win a few projects, max out your team, the projects end, and you sit there wondering how to pay everyone.

Agent builds are project work. The retainer is your way out.

A few ways to make the retainer stick:

• Report the value every month. Conversations handled, appointments booked, hours of reception time saved, cost per conversation. If the client can see the job being done, they won't question the invoice.
• Sell improvements, not maintenance. "We'll add after-hours rescheduling next month" sounds better than "support and maintenance".
• Land and expand. Start with one workflow. Once it works, propose the second. (This is the land and expand strategy from Chapter Seven of the book.)
• Productize it. If you build the same kind of agent for ten clinics, turn it into a fixed monthly package. That's a productized service, and it's far easier to sell, and to value, than custom work.

If you run a services firm, the guide to retainers vs subscriptions goes deeper on structuring the monthly fee.

Free Subscription Profit Review

Australian business owners can book a free 30-minute review with a BPAG Accountants adviser. An honest conversation about:

• whether a subscription model could work for your business
• how recurring revenue could cover your fixed costs each month
• how it could lift what your business is worth

Book my free review Call 1300 272 422

No cost, no obligation. General information only until we understand your situation.

An Australian Note: GST and Currency

If you're an Australian agency, two things catch people out.

GST. Once your GST turnover reaches A$75,000 or more, you must register for GST within 21 days, according to the ATO. If you're registered, your build fees and retainers for Australian clients will generally include 10% GST, and so will any usage you re-bill. Quote prices clearly as GST-inclusive or exclusive.

Currency. Most model vendors bill in US dollars. If you pass usage through in A$, a falling Australian dollar quietly cuts your margin. Either bill usage at the exchange rate on the invoice date, or build a buffer into your margin.

This is general information only. Talk to a registered tax agent about your own situation, especially if you buy services from overseas.

What to Do Next

1. Work out your real day rate with the Consulting Rate Calculator. Include non-billable time.
2. Measure cost per conversation on 50 test conversations before you quote usage.
3. Write the quote in three parts: build, run and usage, with what's included and excluded.
4. Pick a token option (client key, pass-through or capped) and add a price-review clause and a spending cap.
5. Compare with off-the-shelf agents and write down why your build is worth more. If you can't, recommend the product.
6. Pitch the retainer as the main event, with a monthly value report.

For the bigger picture on how agent products themselves are priced, read the pillar guide to AI agent pricing, and if you price other consulting work too, see pricing your consulting work.

Frequently asked questions

Get both books free

The Subscription Playbook (ebook + audiobook) and Feed a Starving Crowd, instantly. In Australia? Claim the free hardcover.

Leave this empty

First name

Email

CountryAustraliaNew ZealandUnited StatesUnited KingdomCanadaOther

What best describes you?Business owner (services)Business owner (retail / e-commerce)SaaS or AI founderHealth or allied health practiceTrades or home servicesAgency or consultantAccountant or adviserEmployee / marketerStudent / other

Yes, I'd like a free 30-minute Subscription Profit Review with a BPAG Accountants adviser (Australian businesses).

Your details go to Subscription Playbook and our sponsor BPAG Accountants Pty Ltd, who may contact you about the book and, if you ticked the box, your review. Unsubscribe any time. Privacy policy.

Send me the free books

Robert Coorey, MBA

Robert is the author of The Subscription Playbook and Feed a Starving Crowd and a Director at BPAG Accountants in Sydney. More about Robert.