An agency owner adds "AI Services" to their website. Three weeks later, a prospect asks what it costs to automate their client reporting. How to price AI services as an agency is not something they have ever thought through. So they say something vague about "custom scoping," panic internally, and follow up with a number they pulled from nowhere. $800 a month. The prospect says yes.
Six months later, that agency is quietly losing money on the account. The build took longer than estimated. API costs crept up as usage scaled. Maintenance requests kept coming in. And $800 a month does not cover any of it.
This is not a rare scenario. It is the default one.
Most agencies that add AI to their service line price it the way they price everything else: based on how much effort it feels like, divided by what they think a client will accept. That works for design hours and copywriting retainers, where your costs track roughly with the work involved. But AI services do not work that way. The build cost and the delivered value are not even in the same conversation. And if you use effort as your pricing anchor, you give away the majority of the margin before the project even starts.
Why Hourly Pricing Breaks Down for AI Work
When an agency builds a client onboarding automation, they might spend 40 hours on it. At $75 per hour, they quote $3,000 and feel like that is fair. But the client just received a system that saves their team 10 hours a week indefinitely. At a $35 per hour loaded labor rate, that is $18,200 in recovered productivity in the first year alone.
The agency charged $3,000 for a system worth $18,000 to the client in year one. And probably double that in year two, because the system keeps running.
The Wall Street Journal noted in early 2026 that agencies are actively moving away from time-based billing in favor of output and performance models, specifically because AI accelerates delivery in ways that make hourly billing structurally punishing for the agency. Zendesk's research puts the average return on AI investment at $3.50 for every $1 spent across customer-facing deployments. The value is not proportional to the hours. It never was.
So the first shift is conceptual, and it matters more than any pricing number in this post: stop anchoring your price to what the build cost you, and start anchoring it to what the problem costs your client right now.
The Three Models Worth Knowing
There is no single right way to price AI services. There are three formats, and the right one depends on what you are building and where the client is in the relationship.
Setup Fee (Project-Based)
A one-time charge for delivering a defined system. This is the correct format when the scope is clear, the deliverable is bounded, and the client needs a fixed number to get internal approval.
What qualifies: a client onboarding automation, an AI-powered reporting agent, a lead scoring pipeline, a custom AI tool with a user interface.
Realistic ranges for the US market in 2026:
- A focused, single-workflow automation: $3,000 to $10,000
- A multi-system integration with custom logic: $10,000 to $40,000
- A purpose-built AI app with a real user interface: $15,000 to $50,000
The setup fee covers the cost of getting the system built and running. It does not cover keeping it running. That is what the retainer is for.
Monthly Retainer (Where the Real Revenue Is)
A fixed monthly fee for monitoring, maintenance, model updates, and handling the edge cases that always surface after launch. This is how agencies build the kind of revenue that lets them stop chasing new clients every month just to make payroll.
Think of it like a maintenance contract on a piece of software the client now owns. The system is live, but models drift. APIs change. The client's data structure shifts. Something needs to be updated every quarter at minimum. And somebody has to own that.
Realistic retainer ranges:
- Simple automations (one or two integrations, stable data): $500 to $1,500 per month
- Mid-complexity systems (CRM integrations, multi-step logic): $1,200 to $3,000 per month
- Complex agents and multi-system pipelines: $2,500 to $6,000 per month
At the low end, a $1,000 per month retainer on a stable system might take one or two hours of actual work per month to maintain. That margin is justified because you are carrying the monitoring responsibility, the maintenance risk, and the accountability when something breaks at 2 AM. You are not selling hours. You are selling reliability and ownership of the outcome.
Value-Based Pricing
This model ties your price to what the system is worth to the client, not what it cost you to build. It is the most defensible approach if you can articulate the ROI clearly, and it is where the market is moving.
The framing looks like this: "Your team currently spends 14 hours a week pulling data into client reports. At $40 per hour, that is $560 a week, or about $29,000 a year. This system automates it entirely. The build is $10,000 and the monthly maintenance is $1,200. You are at full payback in under five months, and every month after that is net positive."
When you frame it that way, the client is not comparing $10,000 against your quote. They are comparing $10,000 against a known, ongoing cost that is already happening in their business.
And that is a very different conversation.
How to Calculate Your Margin Without Guessing
The biggest obstacle to confident AI pricing is not knowing your own cost going in. If you are building everything in-house, your cost estimate is a guess until the project finishes. Scope changes. Engineers hit unexpected complexity. What started as a 40-hour build becomes 70 hours, and by then the margin is gone.
This is one of the structural advantages of working with a flat-rate engineering partner. When your build cost is a fixed number before the project starts, you can price with precision. You know exactly what you need to charge to hit a 2x margin, a 3x margin, or whatever your business model requires. No surprises on your end.
The math is not complicated. Say your engineering cost for a reporting automation is $5,000, flat. You price the delivery to your client at $12,000. That is a 2.4x margin on the build, before accounting for your project management and client communication time. Add a $1,200 per month maintenance retainer and that single client generates $14,400 in retainer revenue in year one alone, from work that takes a few hours per month to maintain once the system is stable.
This is also why getting a real engineering cost estimate before you quote a client matters. The free pilot program gives you exactly that: a full technical scoping of what a build would cost to engineer, before you put a number in front of anyone. You are not guessing when you price. You are working from a real baseline.
How to Price AI Services: Real Numbers by Service Type
These are realistic mid-market ranges for US agencies in 2026. UK and Australian pricing maps closely to USD equivalents. Your actual number should move up or down based on the client's industry, the volume of data the system handles, and how many tools are integrated.
| Service | Setup Fee | Monthly Retainer | |---|---|---| | Client onboarding automation | $5,000 – $12,000 | $800 – $1,500/mo | | AI reporting agent | $4,000 – $12,000 | $500 – $1,200/mo | | Lead scoring and enrichment pipeline | $6,000 – $15,000 | $1,000 – $2,000/mo | | CRM and workflow orchestration | $15,000 – $40,000 | $2,000 – $5,000/mo | | Custom AI app for client's end users | $12,000 – $35,000 | $1,500 – $4,000/mo |
These are not ceilings. They are starting points for an agency without an established track record in AI delivery. An agency with documented results and a clear case study can price materially above these ranges. And should.
One real-world data point that grounds this: a mid-sized operations firm that automated a manual data processing workflow paid a project fee of $14,000 and a $900 per month maintenance retainer. The same workflow, done manually, cost the business approximately $47,000 a year in staff time. That is the kind of ROI conversation that sells itself, and the kind of pricing that is fair to both sides.
How to Frame the Price So the Client Says Yes
You can have the right pricing model and still lose the sale if you sequence the conversation wrong.
Most agencies lead with the number. "Our onboarding automation starts at $8,000." The client immediately starts comparing that against other quotes, against their gut feeling about what sounds expensive, against their budget ceiling. They are anchoring to the cost.
But pricing is not the opening line. It is the punchline. The entire conversation before the number is the setup.
The sequence that works: establish the cost of the current problem first. Quantify it. Walk through what a manual or broken process is actually costing the business each month in staff time, error rates, or delayed delivery. Then introduce the system as the solution to a known, measurable cost. And when the price lands, the client is comparing your number against what they already know they are spending.
Think of it like a restaurant. The chef does not price a dish by calculating ingredient cost and adding 20%. The price reflects the cooking skill, the consistency, the dining experience, and the accountability if something goes wrong. A restaurant that prices on ingredients alone goes out of business.
You are not selling a configuration file. You are selling a production-grade system that works reliably, at scale, and that someone competent will maintain and monitor after it ships. That is worth significantly more than the sum of its parts.
The Retainer Is the Business. The Project Fee Just Gets You In.
One more point, because most agencies miss this entirely.
The setup fee is how you get paid for the build. The retainer is how you build a company.
A 10-person agency with 15 AI retainer clients at $1,500 per month each is generating $270,000 in predictable annual revenue before a single new deal closes. That is a fundamentally different business from one that needs to close five new projects every month just to keep the lights on. Deloitte's 2025 Automation Survey found that companies automating for three or more years spend 22% less per unit of output than peers who have not invested. That gap compounds. So does the retainer revenue.
But this only holds if you own what you built. If the system lives on a vendor's platform you are reselling, the retainer revenue is fragile. If the client leaves, or the vendor changes pricing, or the platform goes down, you are exposed. Full IP ownership means the code belongs to the agency. You can modify it, white-label it, transfer it to the client, or include it in a future acquisition conversation. It is an asset, not a subscription.
To see what this looks like in the context of adding AI as a revenue line, read how agencies are adding recurring revenue without hiring AI developers.
And if you want to understand what your agency could build and own through a white-label engineering arrangement, our custom AI app service covers exactly that.
Frequently Asked Questions
Should I charge by the hour for AI automation services?
Hourly billing is a poor fit for AI work because the value delivered scales independently of the time it took to build. A system that saves a client 10 hours a week is worth the same to them whether it took you 20 hours or 80 hours to engineer. Fixed project fees plus a monthly maintenance retainer is the model that captures that value correctly and protects your margin over time.
What if a client pushes back on the price?
Usually the problem is not the number. It is that the client has not yet connected the price to the cost of their current problem. If they push back, ask them what the manual version of this process is actually costing them in staff time per month. Once they calculate that number, the conversation changes. Clients who understand the ROI rarely argue about the price. Clients who do not understand it always will.
How do I protect my margin if API costs go up after the build?
Include a usage scope clause in your contract that defines the expected data volume and call frequency the system is priced for. If the client scales significantly beyond that, the contract gives you a documented basis to revisit pricing. Working with a flat-rate engineering partner removes your build cost variability entirely. Your only ongoing exposure is the client's API consumption at the infrastructure level, which a well-written maintenance agreement can account for.
Can I resell these systems under my own brand?
Yes, and this is one of the most important questions to ask any engineering partner before the project starts. The right arrangement gives you full IP ownership: all the source code, all the documentation, no lock-in, no royalties. You own the deliverable and can price the delivery and ongoing maintenance entirely on your own terms. Explore our approach to building custom AI tools for agencies to understand what that looks like in practice.
When should I introduce the retainer in the sales conversation?
From the very first proposal. The retainer is not an upsell after the build. It is the second line item in the original quote. "Here is what the build costs. Here is what the ongoing maintenance costs. Here is when you can expect full payback." When the client says yes to the project fee, they already know the retainer is coming. There is no surprise conversation six weeks after launch. And the retainer revenue starts from day one of deployment.
Pricing AI services is not complicated once you have a real cost baseline to work from. If you are not sure what a build would cost before you quote a client, that is exactly what the free pilot program solves. We scope the engineering, assess the complexity, and give you a concrete number to build your pricing around. No commitment required.
