DCF Research

AI Consulting Pricing: Hourly Rates and Project Costs

By DCF ResearchEdited by Peter KorpakPublished · Updated

Published US federal contract schedules put the middle half of hourly ceiling rates at about $138 to $205 for data scientist roles and $150 to $237 for AI and machine-learning roles. A scoped assessment, proof of concept, or build can be estimated as hours multiplied by a blended rate, so total cost depends on team size, duration, and seniority more than on any single rate. The estimates below show their arithmetic so you can change the assumptions. This guide is part of our AI consulting firms research.

How much does an AI consultant cost per hour?

There is no single market rate, but public data gives anchors. On US federal schedules the median hourly ceiling rate is $169 for data scientist roles and $187 for AI and machine-learning roles, and it rises with experience and with vendor size. These are ceilings for federal buyers, not commercial quotes, and they do not cover independent or freelance consultants.

The US General Services Administration (GSA) publishes labor ceiling rates in its CALC+ tool: the maximum hourly rates contractors may charge under Multiple Award Schedule contracts. The percentiles below use the current records from GSA's ceiling-rates API, retrieved on 23 October 2026, for labor categories whose titles match the terms in the table.

Labor-category title containsRecords (vendors)25th percentileMedian75th percentile90th percentile
Data scientist1,488 (481)$138$169$205$245
AI, artificial intelligence, machine learning, ML, or generative449 (133)$150$187$237$288
Data engineer838 (308)$125$149$183$217

Note: current-year ceiling rates per hour, all worksites and business sizes. One vendor can hold several records. Percentiles use linear interpolation between ranked values, as in the PERCENTILE.INC spreadsheet function.

Experience moves the number. Median ceilings for data scientist titles are $142 for categories requiring 0 to 4 years of experience, $177 for 5 to 9 years, and $205 for 10 or more (AI and machine-learning titles: $160, $200, and $237). Vendors classed as other than small business show higher medians than small businesses: $193 against $165 for data scientist titles (339 and 1,149 records) and $215 against $175 for AI and machine-learning titles (137 and 312 records).

Some firms on our list publish rates on the same schedule:

Schedule holderSelected labor categories and hourly ceiling rates (minimum experience)
BCGData Scientist, $153.12 (1 year); Lead Data Scientist, $239.35 (5 years); Principal Data Scientist, $368.24 (7 years, master's); Lead Knowledge Analyst, $380.73 (9 years); Executive Strategy Senior Associate, $478.56 (2 years)
McKinseyBusiness Analyst - Data Science, $354.13 (3 years); Business Analyst - Executive/Strategy, $367.15 (1 year); Capabilities & Insights Expert, $396.76 (5 years, master's)
Cognizant Government SolutionsAI Engineer, $141.32 (entry) to $336.47 (10 years); Data Scientist, $167.02 (entry) to $397.69 (10 years, master's)
SlalomData Scientist I, $201.91 (5 years); Data Scientist II, $221.00 (7 years)
Accenture Federal ServicesEngineering Data Engineer 1 to 3, $147.37 to $237.17 (1 to 5 years; on-site and off-site lines)

Each firm's schedule lists many other categories. Commercial rates are negotiated separately and can be lower or higher.

Rate differences by type of work are firm-specific. BCG's Executive Strategy Senior Associate ceiling is 3.1 times its Data Scientist ceiling ($478.56 against $153.12), while McKinsey's strategy analyst category is within 4% of its data science analyst category ($367.15 against $354.13). On Protiviti's UK government G-Cloud 14 rate card, advisory day rates run about 1.3 to 1.5 times delivery day rates at the same level of the SFIA (Skills Framework for the Information Age) scale, for example £560 against £400 at level 1 and £2,500 against £1,900 at level 7. We found no public contract schedule that reports offshore or nearshore rates for AI roles, so we give none; ask each firm for its delivery-location mix and rate card.

How do billing rates compare with salaries?

The CALC+ median ceiling of $169 is about 2.9 times the median data scientist wage and 2.0 times wage plus benefits. BLS reports a May 2025 median hourly wage of $57.80 for US data scientists (annual mean $126,800; 75th percentile $76.39 an hour). Private-industry benefits were 30.0% of employer compensation costs in June 2026, so wages of $57.80 imply total compensation of about $82.57 an hour (Estimate: $57.80 ÷ 0.700). The ratios are 169 ÷ 57.80 = 2.9 and 169 ÷ 82.57 = 2.0. The remainder has to cover unbilled time, overhead, and profit, and because ceiling rates are maximums, treat the ratio as an upper bound. BLS wage data are a salary floor, not a billing rate.

What pricing models do AI consulting firms use?

Four models cover most deals: time and materials, fixed fee, retainer, and outcome-based. Time and materials suits uncertain scope, fixed fee suits defined deliverables with clear acceptance tests, a retainer suits continuing work, and outcome-based pricing suits only results you can measure.

ModelHow you payWho carries the riskBest forWatch for
Time and materialsHours by role at agreed rates, ideally with a ceilingThe buyer carries overruns unless a ceiling is setExploration, evolving scopeNo built-in incentive to be efficient; ask for weekly burn reports
Fixed feeA price per defined deliverable or phaseThe firm carries cost riskAssessments, proofs of concept, builds with acceptance testsThin acceptance criteria, risk padding, change orders for anything new
RetainerA monthly fee for capacity or defined servicesShared, depending on what the retainer definesMonitoring, iteration, support after launchRetainers with no defined deliverables or response times
Outcome-basedA fee tied to a measured result, such as cost avoidedShared, depending on measurementCases with a clear baseline and attributionDisputes over measurement; agree the baseline, data source, and auditor before signing

The US Federal Acquisition Regulation sets out the risk logic. A firm-fixed-price contract puts maximum risk and full responsibility for costs on the contractor and gives it maximum incentive to control costs (FAR 16.202-1). A time-and-materials contract gives the contractor no positive profit incentive for cost control or labor efficiency, which is why FAR requires oversight of performance, and it includes a ceiling price that the contractor exceeds at its own risk (FAR 16.601). Commercial buyers are not bound by FAR, but the same incentives apply. A common hybrid is a fixed-fee assessment or proof of concept with acceptance tests, followed by a rate card and ceiling for the build.

How much does an AI assessment, proof of concept, or production build cost?

Estimate cost as team size times weeks times 40 hours times a blended rate. With the planning rates below, an engineering-led proof of concept comes to roughly $48,000 to $252,000 depending on scope, and a single-workflow production build to roughly $216,000 to $324,000. These are estimates under stated assumptions, not quotes or benchmarks.

Planning assumptions (DCF Research estimates, with the basis for each):

  • Engineering-led team: $150 to $225 an hour blended. Estimate: chosen to fall inside the CALC+ medians by experience band for data scientist titles ($142 to $205) and AI and machine-learning titles ($160 to $237), so that a mixed-seniority team lands within it. Federal ceilings may sit above or below what a commercial buyer pays.
  • Strategy-house-led team: $350 to $475 an hour blended. Estimate: McKinsey's business analyst and expert categories and BCG's lead knowledge analyst and executive strategy associate categories list at $354 to $479 on their schedules.
  • 40 billable hours per full-time person per week.
EngagementTeam assumptionHoursRate bandEstimated fee
Assessment or roadmap, 4 to 6 weeks, engineering-led2 full-time consultants plus a senior lead at 20% (88 hours a week)88 × 4 = 352 to 88 × 6 = 528$150 to $225$52,800 to $118,800
Assessment or roadmap, 4 to 6 weeks, strategy-house-ledSame team352 to 528$350 to $475$123,200 to $250,800
Proof of concept, lean2 people for 4 weeks2 × 40 × 4 = 320$150 to $225$48,000 to $72,000
Proof of concept, standard2.5 people for 6 weeks2.5 × 40 × 6 = 600$150 to $225$90,000 to $135,000
Proof of concept, extended3.5 people for 8 weeks3.5 × 40 × 8 = 1,120$150 to $225$168,000 to $252,000
Production build, one workflow with usable data3 people for 12 weeks3 × 40 × 12 = 1,440$150 to $225$216,000 to $324,000
Production build, several systems plus governance5 people for 16 weeks5 × 40 × 16 = 3,200$150 to $225$480,000 to $720,000

Two AWS Marketplace listings for a GenAI proof of concept, from Colibri Digital and Flatiron Software, describe an 8-week engagement and list no price; both are sold by private offer. A quote far below your estimate means fewer hours, not a bargain. For example, a $20,000 proof of concept at $150 an hour is about 133 hours, or 3.3 person-weeks (Estimate: 20,000 ÷ 150 ÷ 40). Ask which roles and how many hours it covers. The GenAI proof of concept guide shows what the hours should buy.

What drives AI consulting costs up or down?

Seven factors move the number most: data readiness, the number of integrations, the quality bar and evaluation effort, security and compliance scope, team seniority and vendor type, run-time volume, and change management. Most can be pinned down in the scope before you sign.

  • Data readiness. In RAND's 2024 interviews, 30 of 50 industry practitioners described persistent data quality problems, and 14 of 50 said senior leaders underestimated the time needed to train an effective model. Fund a data audit before you fix a build price.
  • Integrations. Each system to read from or write to adds build and test hours. List them in the request for proposals.
  • Quality bar and evaluation. Higher thresholds need more test cases and expert review. Estimate: 300 test questions × 10 minutes of expert time to write and verify = 3,000 minutes = 50 hours; at $100 an hour of internal cost, about $5,000 of your staff time.
  • Security and compliance scope. Reviews, logging, data residency, and testing against known GenAI risks add hours. Agree the test scope early, for example against the OWASP Top 10 for LLM Applications.
  • Team seniority and vendor type. The 10-plus-year median for data scientist titles ($205) is about 1.4 times the 0-to-4-year median ($142), and larger vendors' medians run roughly 17% to 23% above small businesses' (see above).
  • Volume and latency at run time. Model usage scales with traffic; see the next section.
  • Change management. Training, workflow redesign, and internal enablement cost hours on both sides. Scope them or expect to pay for them in your own staff time.

What costs sit outside the consulting fee?

Model usage, cloud services, your own staff time, evaluation data, and post-launch support sit outside most proposals. Model usage is the one that scales with traffic, so estimate it per query from the vendor's price list before you choose an architecture.

Vendor pricing pageModelInput per 1M tokensOutput per 1M tokens
AnthropicClaude Sonnet 5.5$2$10
AnthropicClaude Haiku 5.5 (prompts up to 100,000 tokens)$0.10$0.50
OpenAIgpt-6.1-sol$2.00$10.00
OpenAIgpt-6-luna$0.10$0.50
GoogleGemini 3.1 Pro Preview (prompts up to 200k tokens)$2.00$12.00
GoogleGemini 3.5 Flash-Lite$0.30$2.50

Standard-tier list prices as of October 2026, from Anthropic, OpenAI, and Google. Models, batch and caching discounts, long-context rates, and scheduled changes vary, so check each page. Google's table labels its output price as including thinking tokens, and Anthropic notes that its newer tokenizer produces about 30% more tokens for the same text than earlier models, so measure token counts on your own documents.

Estimate for a document question-answering assistant, assuming 4,000 input tokens and 500 output tokens per query and 300,000 queries a month:

  • Mid-tier models ($2 input, $10 to $12 output): 4,000 × $2 ÷ 1,000,000 = $0.008 for input, plus 500 × $10 ÷ 1,000,000 = $0.005 for output, so $0.013 per query, or $3,900 a month. At $12 output the figures are $0.014 and $4,200.
  • Small models ($0.10 to $0.30 input, $0.50 to $2.50 output): $0.00065 to $0.00245 per query, or $195 to $735 a month.
  • Proof of concept testing: 10,000 evaluation runs at $0.013 is about $130, small next to labor. Production traffic is where usage cost appears.

This excludes embeddings, vector database and hosting, caching or batch discounts, retries, and agent loops that make several model calls per task.

Other items to budget:

  • Your staff time. Estimate: a product owner at 0.25, a data owner at 0.2, and a security reviewer at 0.1 of a full-time person for 6 weeks is 0.55 × 40 × 6 = 132 hours; at $100 an hour of internal cost, $13,200.
  • Support after launch. Estimate: 20 hours a month at $150 to $225 an hour is $3,000 to $4,500 a month. Negotiate the rate card and response times before go-live.
  • Vendor funding. The Colibri Digital and Flatiron Software listings say completing the proof of concept may qualify for AWS funding, typically up to 10% of costs capped at $25,000. That is the sellers' statement, eligibility is not guaranteed, and funding ties the work to one platform.

How do I compare quotes and negotiate?

Normalize every quote to hours by role, a blended rate, assumptions, and exclusions before comparing totals. Then use public anchors to test the rates, and negotiate scope clarity and risk allocation before price.

  1. Ask for hours by role and level, not a lump sum. Illustration with assumed rates: one architect at $280 and two engineers at $120 blend to ($280 + 2 × $120) ÷ 3 = $173 an hour.
  2. Check the team mix against the work. A senior-heavy team costs more per hour and may need fewer hours; the quote should show both.
  3. List exclusions. Model usage, cloud, data labeling, support, and travel are the usual ones.
  4. Match the model to the risk. Require acceptance criteria for fixed fees, and a ceiling plus burn reporting for time and materials.
  5. Price the follow-on work now. Negotiate a rate card and team continuity for the build while competition still exists.
  6. Tie payment to evidence. Hold back part of the fee against passed acceptance tests and knowledge-transfer deliverables.
  7. Compare different firm types against the same scope. The buyer's guide covers the types, and the RFP template covers how to structure the request.

How was this guide researched?

DCF Research is an independent directory that tracks 50 data and AI consulting firms using public evidence, checked on 2026-10-23. Firms cannot pay to change their score or rank, and we have not audited client projects or tested any firm's work. Rate figures here come from the public sources below, and the estimates are labeled with their assumptions.

Sources

More in AI Consulting

← Back to Insights