Published AI consultant rates for 2026 run from under $100 an hour to several hundred. If you have ever been asked to pick a supplier for an ML feature, you already know the rate is the least useful number on the quote.
I work at Dev Station, an engineering team that builds AI and inspection software for clients in the UK and US. This is the comparison method we suggest to buyers, including when they end up choosing someone else. The full write-up is here: AI Consultant Hourly Rates: Why the Rate Never Predicts the Bill.
The bill is rate times hours
Two quotes for the same feature, round numbers:
quotes = {
"A": {"rate": 100, "hours": 600}, # learning the domain, no reviewer
"B": {"rate": 200, "hours": 250}, # has built the same thing before
}
for name, q in quotes.items():
print(name, q["rate"] * q["hours"])
# A 60000
# B 50000
Quote B is cheaper and ships about four months earlier. Rates only become comparable once every supplier has estimated hours against the same written scope.
Five line items most hourly quotes skip
- Data preparation. Labelling, cleaning and getting access to systems nobody has exported from before. Often bigger than the model work.
- Evaluation. Building the test set, agreeing thresholds and measuring against them.
- Integration. Wiring the model into the screen your staff actually use.
- Retraining. The first drift arrives sooner than expected.
- Handover. Documentation your own developers can maintain from.
Ask for each one as a separate number. A quote that omits them is pricing a demo.
Fix the price of phase one
A supplier who understands the work can put a fixed price on a bounded first phase. Hourly billing for everything moves the estimating risk onto you. That can be fair, as long as you agreed to it on purpose.
If you are pricing AI work now, the guide has a five-point checklist for comparing quotes: https://dev-station.tech/ai-consultant-hourly-rates/

Top comments (0)