The published prices, checked at the source
Every figure here was read on the page it links to on September 1, 2026, and is quoted the way the source states it. Where a guide gives a range for a category rather than a price for its own work, the row says so. My own rows are what my pricing page has said all year, so this table is the one place a buyer can compare a fixed number against the ranges.
| Engagement | Who publishes it | Published price |
|---|---|---|
| Readiness assessment or audit | Hourback (own price) | $999 flat (source) |
| Readiness assessment or audit | The AI Consulting Network (own range) | $2,000 to $8,000 (source) |
| Readiness assessment or audit | ITECS, Dallas (own price) | $6,500, plus a $650 90-minute session (source) |
| Readiness assessment or audit | Five firms tabulated by Hourback | $1,500 to $15,000 across the five (source) |
| Readiness assessment | Walker AI Systems (own price) | $2,500 fixed, two weeks, written report you keep |
| Strategy and roadmap | The AI Consulting Network (own range) | $8,000 to $25,000 (source) |
| Strategy and roadmap | Kamyar Shah (worked example) | $25,000 (source) |
| Strategy and roadmap | Lazige guide (category ranges) | $10,000 to $30,000 freelance, $30,000 to $80,000 agency (source) |
| Strategy and roadmap | Walker AI Systems | Included in the $2,500 assessment: a ranked list of separately scoped, separately priced projects |
| Build or implementation | Starfish Solutions, Longview (own range) | $2,500 for quick-win automations to $15,000 and up for full implementation (source) |
| Build or implementation | The AI Consulting Network (own ranges) | Pilot $15,000 to $50,000; full implementation $50,000 to $150,000 and up (source) |
| Build or implementation | Lazige guide (category ranges) | Small $10,000 to $40,000; medium $40,000 to $150,000 (source) |
| Build or implementation | Wiss, for mid-market manufacturers | $75,000 to $250,000 consulting fees, breakeven in 12 to 18 months (source) |
| Build | Walker AI Systems (own price) | From $18,000, fixed against a written scope, six acceptance criteria signed at the end, you own everything |
| Ongoing: retainer or fractional | The AI Consulting Network (own range) | Fractional chief AI officer $2,000 to $8,000 a month (source) |
| Ongoing: retainer or fractional | Kamyar Shah (own range) | $12,000 to $15,000 a month for one day a week (source) |
| Ongoing: retainer or fractional | ITECS, Dallas (own price) | $3,375 for a prepaid 15-hour block (source) |
| Ongoing: retainer or fractional | Paul Okhrem (own price) | Fractional chief AI officer $30,000 a month, six-month minimum (source) |
| Ongoing: operate and maintain | Walker AI Systems (own price) | From $3,500 a month for one system, monitoring, incident response and a monthly accuracy report, cancel any time |
| Hourly | Kamyar Shah (category ranges) | Independents $150 to $250; boutiques $250 to $400; enterprise firms $400 to $500 and up (source) |
| Hourly | Paul Okhrem (own price) | $1,000 an hour, 80-hour minimum (source) |
| Hourly | Hourback, from Upwork medians | $50 an hour median for a freelance AI engineer, $100 for an ML engineer (source) |
| Hourly | Walker AI Systems | Not offered. No hourly billing, no free consultation |
| Small business, one workflow | Walker AI Systems (own price) | $1,500 automation sprint, fixed, one week, one job removed and shown working on your data |
Three things the table shows. Almost everyone quotes a range, and the ranges are wide enough to be useless for planning: the same guide can put a build anywhere from $10,000 to $150,000. The firms that print a real number are rare, and they are the ones you can hold to it. And the spread on assessments, from $999 to $15,000, is not a spread in quality so much as a spread in what "assessment" means, which is why the next section is about what the fixed scope buys.
What a fixed scope actually buys
My $2,500 assessment is two weeks and a written report. The report is yours whatever it concludes. Inside the scope: interviews with the people who do the work, not just the executive who signed off on it; a map of the processes and systems across the target functions, including the tools nobody documented; every opportunity quantified in hours and dollars with the assumptions shown so you can argue with them; a readiness score from one to five, where five means automation that independently verifies its own work; and a prioritized roadmap with honest build estimates, including what is not worth building. Your team's footprint is a kickoff call, one short interview per person, and one working session at the end.
A build starts at $18,000 because the model is the easy part. Integration into systems that were not designed to be integrated with, and the verification and monitoring layers that tell you whether the thing is still working, are most of the work, and they are exactly what gets cut to hit a lower number. The price is fixed against a written scope and a named date, half on signing and half on acceptance, and acceptance is six criteria we agree before work starts and test together at the end. Done is a test, not an opinion. You own the code, the prompts, the configuration and the documentation, built in your accounts and in your name.
Operate and maintain, from $3,500 a month for one system, exists because production AI rarely breaks loudly. It degrades quietly: a model update shifts behavior, an upstream schema changes, extraction accuracy slides and nothing throws an exception. The retainer buys monitoring that does not depend on a person being awake, incident response by the engineer who built the system, model migration against a regression suite, and a monthly accuracy report that always includes what the system got wrong. You can cancel any time and keep everything.
What is out of scope?
Ask this first, because a proposal that cannot answer it is a proposal that will grow. My out-of-scope list is on the pricing page and it is the financial control on every engagement: anything not in the written scope is a change order with its own price, not a favor and not a surprise. And there is a second list, of who I am not for: under about twenty employees and shopping the full build ladder, where the $1,500 sprint is the door and an $18,000 build usually is not; anyone looking for a $99-a-month tool they configure themselves, several of which are good and I will name them; anyone wanting a strategy deck with no intention of building; anyone needing a team of ten on site next month; anyone hoping to replace headcount outright rather than move people onto the exceptions; anyone unwilling to give a consultant access to real systems and real staff; and anything touching protected health information.
What happens if it is wrong after launch?
The honest answer from most vendors is that you will find out from a customer. Mine is a guarantee with two stated limits. Every system I build reports on itself independently and tells you when it is wrong; if one fails without telling you, that is my failure, not yours. It runs for thirty days after acceptance under the build's warranty, and after that for as long as I am actively monitoring the system. I cannot promise to catch a silence nobody is listening for, which is what the retainer is for, and the guarantee covers the telling: everything the system does tell you still needs an operator to act on it.
Who owns the code?
You do, outright, on delivery: code, configurations, prompts, workflows and documentation, built in your accounts under your name, with a team handoff. Ask any consultant whether the system lives in their accounts or yours, and whether you could run it the day they walked away. If the answer involves a platform fee to keep using what you paid to build, the build was a lease.
Have you ever told a client not to buy AI?
Ask this of everyone, and watch for the story. Mine is short because the practice is new enough that the answer is still ahead of me, and I would rather say that than invent a client. What I can show is that the engagement is built so that "do not build this" is a legitimate finding. The assessment report includes what is not worth building. The sprint's own terms say that if a week of my time will not pay for itself in your operation, I say so before you sign. And my field ticket guide does the arithmetic in public: at typical clerical inputs, the labor saving alone roughly breaks even on the retainer and never repays the build, which is why the case has to be the money that leaks, and why a shop at three minutes a ticket does not need me.
How is accuracy measured, and by whom?
Against an answer key written before the run, not against the system's own report of itself. My recorded demos publish the score spread across runs and the direction of the misses, and both demo pages say which numbers are measured and which are invented stand-ins. In a build, the six acceptance criteria are the measurement, tested together and signed. In operation, the monthly report samples extraction against source documents and verifies actions against the real system state, and it always includes what the system got wrong. A monthly report that never contains a failure is a report nobody is reading closely.
What does month six cost?
Ask for the number, not the model. With me it is one of two things. Nothing beyond your own model and infrastructure bills, because you own the system and can run it yourself. Or the retainer, from $3,500 a month for one system, rising with what is under management, re-earned every month. Third-party costs, model usage, infrastructure and licenses, are billed to your own accounts in your name, visible to you in full, with the run-rate estimated before you commit, so there is no margin hidden in someone else's invoice.
When you should not hire an AI consultant
- The process you want automated takes one person a few minutes a day. The arithmetic will not clear any of the prices in the table, including mine.
- A configurable tool already does it. Several are good, and the honest consultant names them in the first call.
- You want a strategy deck. The deck is the cheap part and it is worth roughly what it costs to produce.
- You cannot give the consultant access to real systems and real staff. Without both, every number in the proposal is a guess.
- Nobody in your company will own the exceptions. Every system routes something to a person, and if no one is assigned to that queue, the system is unfinished on day one.
Questions buyers ask
How much does an AI assessment cost?
Published prices for a readiness assessment or audit run from $999 to about $15,000, with most firms quoting a range rather than a number. Mine is $2,500, fixed scope, delivered in two weeks as a written report you keep, including the finding that you should not build anything yet if that is what the numbers say.
Why is a fixed price better than an hourly rate for AI work?
Because the buyer cannot see the hours and the seller controls them. An hourly engagement prices the seller's uncertainty onto the buyer. A fixed price against a written scope with acceptance criteria puts the risk of estimating badly on the person who knows the work, which is where it belongs. Published independent rates run from about $150 to $250 an hour and up to $1,000 an hour with an 80-hour minimum, and none of those numbers tells you what the finished thing costs.
Why does a build start at $18,000?
Because the model is the easy part. The majority of the work in a production AI system is integration into systems that were not designed to be integrated with, plus the verification and monitoring layers that tell you whether it is still working. Those layers are exactly what gets cut to hit a lower number, and cutting them is the single most common reason AI pilots never reach production, or reach it and quietly degrade. A build priced below this is either a prototype or a system nobody is watching.
What does an AI system cost in month six?
Either nothing beyond your own model and infrastructure bills, because you own the system outright and can run it yourself, or a monthly operate-and-maintain retainer, from $3,500 a month for one system, that buys monitoring, incident response by the engineer who built it, model migration against a regression suite, and a monthly accuracy report that always includes what the system got wrong. You can cancel any time and keep everything.
Is there a cheaper option for a small business?
Yes. A $1,500 fixed-price automation sprint: one job the office still does by hand, the smallest thing that removes it, built and shown working on your real data in one week, with a check you can run yourself. If a week of my time will not pay for itself in your operation, I say so before you sign. Below about twenty employees the full build ladder is usually the wrong tool, and the sprint is built for that size.