What an AI automation consultant costs, and what to ask before you sign

Most people asking what an automation consultant costs have already had one bad experience: a quote with no scope behind it, a build that stopped working the first time nobody was watching, or a login they still cannot get into. So the useful version of this guide is not a price list. It is the shape of the market, the questions that separate a real shop from a repackaged one, and an honest account of when you should not hire anyone at all.

We run an AI automation practice for small companies, so we have an obvious interest in you hiring someone. The questions below are written to be used against us too, and further down we answer several of them for ourselves.

What the market actually charges

Nobody publishes a real price list, because scope is unknown until someone looks at your workflow. What does exist is a set of published ranges from firms working in this market, and they agree with each other more than you might expect.

On hourly rates, Layer3 Labs publishes a card for the US market: freelance AI developers at $100 to $200 an hour, boutique AI firms of five to twenty people at $150 to $300, and enterprise consulting firms at $300 to $600. Zaps Studio, a workflow automation shop publishing US and UK figures, covers the bottom of the same market: $50 an hour for offshore Zapier work, up to $250 an hour for senior consultants. The two overlap where you would expect, around the independent and small-shop band.

On the work before the build, Layer3 Labs prices an AI readiness assessment at $5,000 to $12,000 and a proof of concept at $8,000 to $20,000. Assessment prices vary this widely for a real reason. Mapping one invoicing process that lives in one system is two days of interviews and a document. Mapping a practice running four systems that disagree with each other, where half the procedure exists only in one person's head, is a different job with a different price.

On builds, the two firms are describing different ends of the market. Zaps Studio puts a single workflow at $2,000 to $3,000 from an agency, and $400 to $5,000 from a freelancer. Layer3 Labs, working further upmarket, prices a full implementation at $15,000 to $75,000, with published benchmarks for individual workflows: $10,000 to $25,000 for lead intake into a CRM, $12,000 to $28,000 for customer support triage, $15,000 to $35,000 for document extraction. The gap between the two firms is mostly a disagreement about what counts as finished, which is the single most useful question to put to anyone quoting you. If invoicing is the workflow you have in mind, we wrote a longer piece on automating invoicing in a small business, including the cases where it is not worth doing.

Five things move a build from the low end of those ranges to the high end:

• How many systems have to talk to each other, and whether they have real interfaces or someone has to write software that drives a web page like a human would.
• How clean your data is. Two spellings of the same customer name is an afternoon. Fifteen years of a spreadsheet nobody owned is its own project, and it usually comes before the automation rather than after it.
• Exception handling. The path where everything is normal takes a day or two. The version that behaves correctly when the invoice is a photograph, the address is wrong, or the other system is down for an hour is most of the real cost.
• Compliance. Health, legal and financial data brings review, audit logging and access control that a marketing workflow never needs.
• Who owns the accounts afterwards. Building inside your accounts, documented well enough that a stranger could take it over, costs more than building inside theirs. It is worth the difference.

Running costs are the line most quotes leave out. Zaps Studio puts genuine business use of Zapier at $70 to $300 a month, Make's core plan at $0 to $30, n8n at $20 a month on their cloud, and n8n self-hosted at the price of a small server. On top of that sit model costs, which scale with how much text you push through them and belong in the quote as a monthly figure rather than a footnote. On retainers, Zaps Studio lists $500 to $3,500 a month and Layer3 Labs lists ongoing advisory at $2,000 to $8,000. Those are not the same product, so ask what a retainer actually buys before you compare two of them.

The two ways you will be quoted

Almost every proposal you receive is one of two shapes, and the difference decides who carries the risk of a bad estimate.

Hourly, or time and materials. You pay for time at a rate. This is honest in the narrow sense that you pay for what actually happens. It is also the model where every estimation error lands on you. If the integration turns out to need three weeks instead of one, the consultant bills three weeks, and their week was fully paid either way. There are situations where it is the right structure, mostly genuine research where nobody could scope the work in advance. A routine integration is not one of them.

Fixed price, off a scoping phase. Someone spends defined time understanding the problem, then quotes a number for the build with the scope written down beside it. Now the estimation risk sits with the consultant, which is where it belongs, since they are the one who has done this before. The cost to you is that scope becomes a real object: change requests are events with prices attached, and you have to be willing to say what finished means.

A free quote produced in twenty minutes on a first call is neither of these. It is a guess with a number attached, and it will be revisited. For what it is worth, we quote a fixed price off a scoping sprint and never bill hourly. How that works is on our pricing page.

Questions to ask any consultant, including us

Ask all of them, of everyone you talk to. A firm that has thought about this will answer quickly and specifically. A firm that gets uncomfortable has told you something useful for free.

1. Who owns the code and the repositories when this is finished? Good: you do, in writing, in a repository you can log into today. Bad: a warm sentence about owning your data, which says nothing about the code.
2. Whose name is on the cloud, API and tool accounts? Good: yours, or theirs with a written path to move them and a reason why theirs makes sense for now. Bad: theirs, and nobody had thought about it until you asked.
3. What happens if we stop working together in a year? Good: a specific handover made of documented code, a runbook, and a call to walk your people through it. Bad: reassurance that it will not come to that.
4. What does the exit cost? Good: a number, or a formula that produces one. Bad: no answer, which almost always means no plan rather than no charge.
5. Who actually does the work? Good: the person on this call, or a named person you meet before you sign. Bad: our team, our engineers, a bench you will never be introduced to.
6. What happens when the model or the API you built on changes? Good: an account of what they watch, how they test against it, and who pays when something has to be rewritten. Bad: silence, or an assumption that it will not happen. It will happen.
7. What will you refuse to automate? Good: a real list. Most careful shops will not automate anything where a wrong answer is expensive and hard to notice, which usually covers final pricing, medical or legal advice, and anything that moves money on its own. Bad: nothing. A shop with no refusals has not thought about the failure cases.
8. Where does our data go, and does it train anyone's model? Good: named providers, named settings, and a written answer you can hand to your lawyer. Bad: a general statement about taking security seriously.
9. What does this cost to run every month, and who pays the API bill? Good: a figure, plus a clear answer about whose account it bills to and what happens if usage doubles. Bad: a build price with no running cost attached to it.
10. What happens when the automation gets something wrong? Good: a description of how an error surfaces, who sees it, and what the manual fallback is that morning. Bad: an accuracy percentage with no process around it.
11. How are change requests priced? Good: written change orders, priced and approved before anyone starts building. Bad: a promise to be flexible, which prices the change after the work.
12. What would make you tell us not to do this? Good: an actual answer, delivered without hesitation. Bad: none.

Three of those for us, since we said to use the list on us. We build and host in our own accounts by default, because a company without an engineering team rarely has anyone to administer a cloud account, and we transfer everything into accounts in your name for a one-time fee whenever you ask: the code, a written runbook, and a call to walk your team through running it. The person on your intro call is the person who builds the thing, because there are two of us and no bench. And we quote a fixed price, never an hourly rate.

Red flags

• A quote before anyone has looked at the workflow. Nobody can price work they have not seen. A number offered that early is a hook, and it will move.
• Per-seat pricing for something built once, for you. Per-seat makes sense for software sold to thousands of companies. On a bespoke build it is a way to charge more later for work already delivered.
• AI attached to something a form and a database would do. Ask what the model is actually deciding. If the answer is thin, you are paying an AI premium for ordinary software.
• No written scope. Without one there is no way to tell a change request from a correction, and every disagreement becomes a matter of memory.
• Refusing to name the stack. A proprietary platform they will not describe usually means the work cannot go anywhere else.
• Accounts in their name with no transfer path. This is the most common form of lock-in in this market and the easiest one to check before you sign anything.
• A discount for signing this week. A price that expires on Friday was never really a price.

When to hire nobody

A decent share of the time, the right answer is that this is not a project. Five cases where you should keep your money.

The software you already pay for does it. Most accounting, scheduling and CRM products have shipped automation features in the last two years that nobody switched on. Before paying anyone, send one message to your current vendor's support team describing what you want. That question is free and it is frequently the entire answer.

It is a $500-a-year problem. If the task takes twenty minutes a week, that is roughly seventeen hours a year. At any rate on this page, the payback is measured in decades. Write it down as an annoyance you have accepted and go do something else.

The process underneath is broken. Automating a bad process produces wrong output faster and in greater volume. If three people in your office would describe the current steps differently, that disagreement is the actual work, and a whiteboard fixes it for nothing.

You need a person, not software. Some of what feels like a software problem is one unfilled role. If the job needs judgment, a relationship, or someone answering a phone at seven in the morning, hire the person. Zaps Studio puts an in-house automation hire at $45,000 to $75,000 a year plus $5,000 to $20,000 to recruit them, which is a real comparison worth running.

The volume is too low. A build that pays for itself in eight weeks at 200 orders a month takes four years at 20. Do that arithmetic yourself, before anyone quotes you, using your own numbers.

This is not abstract caution. S&P Global Market Intelligence's Voice of the Enterprise survey of more than 1,000 companies across North America and Europe found that 42 percent had abandoned most of their AI initiatives during 2025, up from 17 percent the year before, with the average organization scrapping close to half of its proofs of concept before they ever reached production. Gartner predicted back in 2024 that at least 30 percent of generative AI projects would be abandoned after proof of concept by the end of 2025, naming poor data quality, weak risk controls, escalating costs and unclear business value as the causes. Those are companies with budgets and technical staff. A ten-person firm gets fewer attempts than that, which is a reason to be picky about the first one rather than a reason to avoid the whole category.

How to run the engagement once you have picked someone

Once the contract is signed, a surprising amount of the outcome is determined on your side of the table rather than theirs.

• Get the scope in writing, including what is excluded. The exclusions list is the more useful half of the document, and it is the half most proposals omit.
• Name one owner on your side. One person who can answer questions and make decisions without convening anyone. Projects routed through a committee stall in week three.
• Budget a couple of hours a week of that person's time. Not much more, and not zero. Every automation encodes decisions only your team can make, and a consultant guessing at them is how you end up with something nobody uses.
• Ask for a working demo on a fixed cadence. Weekly suits most builds. Something you can click is the only progress report that cannot be dressed up.
• Check three things at every milestone. That you can log into the accounts it runs on. That someone has written down how to operate it without them. That you have watched it handle a bad input, not only a clean one.
• Run it in parallel before you switch over. Two weeks of the old process and the new one side by side will surface the exceptions no scoping conversation ever finds.

Common questions

How much does an AI automation consultant cost?

Published ranges cluster into a few bands. Layer3 Labs lists boutique AI firms at $150 to $300 an hour and full implementations at $15,000 to $75,000, while Zaps Studio puts a single agency-built workflow at $2,000 to $3,000. Your own number depends mostly on how many systems have to talk to each other and how clean your data is.

Should I pay hourly or a fixed price?

Fixed price off a scoping phase is usually better for a small business, because it moves the estimation risk to the consultant. Hourly billing means every hour the work runs long is an hour you pay for, and nothing about the arrangement improves when the consultant works faster.

Is a paid audit worth it before a build?

Usually yes, if what it produces is a written scope and a fixed quote rather than a slide deck. Layer3 Labs prices an AI readiness assessment at $5,000 to $12,000. The test is whether another firm could take the output and build from it.

Who should own the code and the accounts?

You should, or there should be a written path to move everything into accounts in your name for a stated fee. Accounts held in a consultant name with no transfer plan is the most common form of lock-in in this market, and it is easy to check before you sign.

When should I not hire an automation consultant?

When the software you already pay for does the job, when the task costs you a few hundred dollars a year, when nobody agrees yet on what the current process is, or when your volume is too low for the build to pay back. Ask your existing vendor first.

If you want a second opinion

Bring us a workflow and we will tell you what we would do with it, including the version where the answer is that you should not build anything. The intro call is thirty minutes and there is no charge for it.