How much does AI automation cost?

The three pricing models you will see

Fixed-price project

One workflow, one written scope, one price, usually with a deposit. This is the model we use, and it is the most common for single-workflow builds. Its strength is that you know the number before you commit. Its weakness is that it only works if the scope is defined properly, which is why the assessment comes first.

Monthly retainer

A fixed monthly fee for ongoing build and support. It suits businesses that want a steady stream of automation work over a year. It suits the agency more than the client when the first workflow is small, because you pay for capacity you may not use.

Hourly or day rate

Common with freelancers and consultancies. Fair when the work is genuinely open-ended. Risky for a defined workflow, because the cost is unknown until it is finished and the incentive runs the wrong way.

The six things that actually move the price

1. How many systems the workflow touches, and whether each has an API. One system with a good API is the simplest case. Every additional system adds a connection, and a system that only offers exports or screens can double the effort on its own.
2. How many steps need judgement versus a rule. A rule is cheap to write and test. A step that needs AI reading needs a confidence threshold, a review screen, and tuning on your real data. A step that needs a human decision needs an approval flow.
3. Volume. More items per month means more edge cases to design for and more usage cost after launch. It also means a faster payback, so volume cuts both ways.
4. How messy the inputs are. Structured forms and EDI are simple. Free-text emails, scans, photos, and handwriting are where AI is needed and where exceptions come from.
5. How many approval steps and exception paths exist. Each one is designed, built, and tested. A process with one happy path and two exceptions is a fraction of one with six.
6. Rollout approach. Running the old process alongside the new one for a while is safer for high-stakes work and takes more of your team's time and ours.

Our how we work page goes through each of these as it applies to a specific engagement.

What publicly listed prices look like

On 24 September 2026 we looked at agencies ranking for AI automation searches that publish prices. This is not an endorsement or a comparison of quality; it is what is written on their pages, so you have a reference point.

• One agency lists a one-time build fee of $7,500 plus $30 to $150 per month in AI usage for a standard automation, with a $500 pilot option (AutomateNexus, Texas page).
• Another lists a $3,500 starting point for a basic workflow and $8,000 to $25,000 or more for AI agents, with a two to eight week timeline (AI Automation Agency New York).
• A third describes projects from "low thousands" to "six figures" with a two to six week build (The Automators, New York City page).

The spread is the point. The same phrase, "AI automation," covers a simple inbox rule and a multi-system integration with document reading and approvals. The drivers above explain the gap.

Why we don't publish price bands

Because a band wide enough to be honest is too wide to be useful, and a band narrow enough to be useful would be wrong for most readers. What we do instead: a free assessment of one workflow, a written review of where the hours go, an honest answer on whether it is worth automating, and, if it is, a written price and timeline. Half the setup fee is a refundable deposit until you approve the detailed design. You can stop at any point before that and owe nothing.

Ongoing costs after launch

• AI usage. Models are metered per document or message. For most small-business workflows this is modest, and the proposal states the expected range for your volume.
• Software and hosting. Any platform or infrastructure the automation runs on. Where we can use systems you already pay for, we do.
• Support. Monitoring, tuning exception rules as edge cases appear, and adapting to changes in your process or systems. What it covers and costs is in the written proposal.

Estimating the other side: what the process costs you now

A build cost only means something next to the cost of the manual process. The process cost calculator gives a quick, deliberately conservative estimate from people, hours, and hourly cost, with a recoverable range that is an assumption, not a promise. If the recoverable number is small, automation is probably not the answer, and we will say so.

Six questions that get you a real quote

1. Which steps will be deterministic rules, and which will use an AI model?
2. What happens when the model is not confident, and what happens when a system is down?
3. Which systems will you connect to, by what method, and have you confirmed access?
4. What do you need from our team, and how many hours of our time?
5. Who owns the automation, the code, and the data after launch?
6. What does support cover, and what does a change to the process cost later?

A provider that can answer all six in writing is quoting a real scope. One that cannot is quoting a guess, however confident the number looks.