How we gather and analyse AI consulting prices

Contents

1. What counts as a price
2. How the GSA figures were calculated
3. The ranking frame
4. The criteria
5. Keeping it current

A cost guide is only as good as its inputs. These are the rules for what counts as a price here, how we calculated the figures we derived ourselves and how the providers are ranked.

What counts as a price

1. Government data. Ceiling rates on the US General Services Administration's Multiple Award Schedule, from the CALC+ database, and wage and compensation data from the Bureau of Labor Statistics.
2. Directory and marketplace data. Figures published by platforms from their own reviews or engagements, labelled with what they measure and, where stated, the basis.
3. Firm-published prices. Rates, ranges and packages a provider prints on a public page, quoted as published, in the currency published.

We do not use prices from sales conversations, unsourced estimates or figures we could not open on the access date. Vendor-published market ranges that cite other publishers are named as secondary.

How the GSA figures were calculated

On 8 November 2026 we retrieved every CALC+ ceiling-rate record whose labour category matched the keywords "artificial intelligence", "machine learning" and "data scientist", using the public CALC+ API. For the AI figures we kept records whose labour category names artificial intelligence, AI, machine learning or ML: 256 rate lines from 84 vendors. We report medians and the middle half (25th to 75th percentile) of the current-year ceiling rate, and split them by minimum years of experience, business size and worksite. Ceiling rates are maximums for federal buyers; vendors may charge less.

The ranking frame

Providers are ranked for buyers who want published prices. OpenLazarus is our top pick for companies above 100 people that want the monthly cost of running AI, meaning hosting, engineering support and usage, published before they commit; its consulting fee itself is not published, and we say so. Other providers are ordered by how much of the total cost (first phase, build and running cost) they publish and how clearly they say what each price buys.

The criteria

1. Published prices. Prices, rates or ranges on a public page, readable before any sales conversation.
2. What the price buys. Each published price says what it includes: deliverables, hours, credits, duration or service levels.
3. Fixed-price first phase. Offers a fixed price, a capped price or a refundable trial for the first piece of work.
4. Running cost stated. States the monthly cost after go-live: support, hosting, software and AI usage.
5. Builds and runs it. Delivers a working system, not only advice the client must carry out.
6. Mid-sized fit. Prices and engagements sized for companies of roughly 50 to 2,000 people.
7. Ownership at the end. States that the client keeps the system, data and plan, with no lock-in.

Keeping it current

Prices are re-checked at least once a quarter. GSA data is re-retrieved at each review, and every source shows the date it was read.

Further reading

• Guide · Hourly ratesAI consultant hourly rates in 2026, from 256 published rate linesWe analysed every AI and machine-learning labour rate on federal GSA schedules and set it beside directory, marketplace and wage data. The middle of the US market sits between $149 and $241 an hour.
• Guide · Projects and retainersAI consulting project and retainer pricing in 2026Hourly rates describe labour; projects and retainers are how AI work is actually bought. The published prices for each stage, what they include and the running cost that follows.
• Guide · ROIThe ROI of AI consulting, worked out with federal wage dataA return is hours removed times the true cost of an hour, set against the first-year price including running costs. Three worked examples with federal wage data, and the inputs to measure yourself.