Thirty listings carry the same number. Of the 304 listings in this directory that are not seeded placeholders, 50 publish an hourly rate floor, and 30 of those 50 put it at exactly $50. The next most popular figure, $150, was chosen by 8. Anyone researching AI consultant cost by reading rate cards is therefore reading a distribution with one enormous spike in it, and the spike is not a market price.
It is the first item on a dropdown menu. The submission form on this site makes the minimum hourly rate a required field and offers it as a list of 12 fixed options, the lowest of which is $50. The maximum rate sits directly underneath and is optional. That single piece of form design explains most of the shape of the data: 30 listings at the floor of the menu, and 10 listings that publish a floor with no ceiling at all because nobody made them fill the second box in.
This is the whole problem with pricing an engagement off a published rate, and it is worth saying plainly before the table below. A rate card tells you what a firm is willing to put on a public page. It does not tell you how many hours the work takes, who inside the firm does it, or what you own at the end.
Thirty listings, one number
Here is the full set of published floors across those 50 listings, queried on 16 October 2026, with the highest ceiling that appears alongside each floor.
| Published floor | Listings | Highest ceiling shown with it |
|---|---|---|
| $50 | 30 | $500 |
| $75 | 3 | $200 |
| $100 | 3 | $200 |
| $125 | 3 | $150 |
| $150 | 8 | $250 |
| $200 | 1 | $250 |
| $250 | 1 | $500 |
| $400 | 1 | $500 |
The median published floor is $50 and the median published ceiling is $125. Two of the 50 listings advertise a band of $50 to $500, which is a tenfold spread on a single card and tells a buyer nothing a buyer can act on. Six publish a flat rate, where the floor and the ceiling are the same figure, and those six are the only listings in the set that have committed to a price.
The other 254 of the 304 publish nothing about price. That is the number to keep in mind, because it means the rate card is a minority behaviour on this directory and probably on every other one. Firms that quote per project, or that will not quote until they have seen the work, simply leave the field empty.
What a band of fifty to five hundred leaves out
An hourly rate is a unit price. The bill is the unit price times the number of units, and the second term is the one that varies by an order of magnitude between firms bidding on the same brief. Two things drive it.
The first is who does the work. A firm quoting $150 with four senior engineers on staff and a firm quoting $150 that subcontracts to a partner shop are selling different things at the same number. Of the 304 real listings here, 52 publish a team size, and the 4 that publish a typical project size are naming a budget band rather than a headcount, so for most of the directory you have to ask. The question that settles it is not what the rate is but who will be in the standups.
The second is scope discipline. An engagement that ends in a document costs less per hour and more in total than one that ends in something running, because the document has to be implemented by somebody afterwards. We have written about what survives the end of an engagement in what you actually own when an AI consulting engagement ends, and about the difference between a roadmap and a build in what the roadmap leaves out.
Is a higher rate ever the cheaper option?
Routinely, and this is the part that reads as counterintuitive until you have paid for it twice. A specialist at $300 who has built the same integration nine times will scope it in a day and ship it in a week. A generalist at $75 who has not will spend three weeks discovering the problem the specialist already knew about. The $75 firm is cheaper per hour and more expensive per outcome, and the published rate card ranks them in the wrong order.
This is why the single most useful thing a buyer can ask for is not a discount but a reference to a piece of work that resembles their own. Integration work has a shape, and a firm that has done yours before will describe it before you finish explaining it.
Where the money actually goes
Buyers new to this category often assume the model is the expensive part. It is the cheapest and most predictable line in the budget. Both OpenAI and Anthropic publish per-token prices on a public page and bill inference as metered usage, so that line can be estimated in advance from your own expected volume and it does not need a consultant at all.
What a consultant is actually selling is the work around the model. Getting your data into a state the model can use. Deciding which of the things you asked for should not be built. Wiring the output into the system where somebody will read it. Writing the evaluation that tells you when the thing has stopped working. None of that is metered, all of it is hours, and the last of the four is the one most often dropped from a fixed-price quote and then billed later.
What should a pilot cost?
Less than the thing it is testing, and it should have a written stopping condition. A pilot priced at a fifth of the full build, with a date and a defined output, buys you the right answer to the question of whether to proceed. A pilot with no stopping condition is a retainer with a friendlier name. Ask what happens if the pilot fails, and listen for whether the answer is a number or a feeling.
Does an agency cost more than hiring?
Per hour, almost always. Per year, it depends on whether you can keep the person busy and whether you can keep the person. The comparison people get wrong is the one that puts a contractor's rate next to a salary and stops there, leaving out recruitment, the months before the hire is productive, and the risk that one departure takes the whole capability with it. We have set that comparison out properly in an agency against an in-house team.
Why do two quotes for the same brief differ by 5x?
Because they are quoting different work, and usually the brief allowed them to. A quote is only comparable to another quote when both price the same deliverables, the same integrations and the same handover. When the spread is that wide, the fault is generally in the brief rather than in the firms. A first-time buyer's checklist is mostly a list of things to pin down before you ask for a number.
Is one person enough?
For advisory work and for a narrow build, often. For anything that has to stay running after launch, a single contractor is a concentration of risk you should price. The trade is set out in when one person is enough, and the tiered or subscription arrangements some firms now offer sit between the two options, which we covered in AI consulting sold as a service.
Four questions that price the work instead
These replace the rate question, and all four can be asked in a first call.
What is the smallest version of this that would still be useful? A firm that can answer this has scoped work before. A firm that cannot will sell you the largest version.
Who specifically does the work, and what else are they on? Names and allocation, not job titles. This is the question that exposes subcontracting.
What do I hold if we stop after the pilot? Code, data, documentation, model configuration, access. Get the list before signing, not after.
What does month 13 cost? Maintenance, retraining, monitoring and the inference bill. A build quote that does not mention month 13 is not a total cost.
Firms that price by deliverable rather than by hour are easier to compare on all four, and the machine learning end of this market has moved further in that direction than the advisory end. The custom build end is the furthest behind, which we looked at in what you are buying when you buy a custom AI solution.
What to do with a published rate
Use it as a filter and not as a forecast. A floor of $50 next to a ceiling of $500 means the firm has not decided, so ask. A flat rate means somebody has thought about it, and 6 of the 304 real listings here have. An empty field means the firm quotes by project, which is neither a good nor a bad sign on its own and is the most common case by a wide margin.
Then price the work rather than the hour. Write down the smallest useful version, ask four firms what it would take, and compare the scopes they come back with before you compare the numbers. The spread in the scopes will be wider than the spread in the rates, and it is the one that decides what you pay.
You can read the hourly rates, cities and services on every listing in the full directory, including the fields left blank. Describe what you need built on the project form and we will show you the listings carrying that category, sorted by listing tier. If you run an agency, add your listing, and if you choose to publish a rate, consider publishing a flat one.
Figures in this article were queried from this directory's own database on 16 October 2026 and describe the 304 listings that are not seeded placeholders. A bare count of every row returns different numbers, because 129 of the 179 seeded placeholder rows also carry a rate. Nothing on this site is vetted, reviewed or rated, and no listing is ranked by quality.