What AI Agency Work Costs and What Changes the Price

A price range is useful only when you know what is inside it.

Planning ranges are not quotes

The figures below are starting points for a budget, not market guarantees. Two projects with the same label can have different data, integrations, security requirements, and definitions of done. Ask an agency to replace the range with a scope, assumptions, exclusions, and payment schedule.

Type of work	Planning range	Planning timeline
AI strategy or roadmap	$5K – $25K	2–4 weeks
Chatbot or virtual assistant	$5K – $150K	2–12 weeks
Workflow automation	$10K – $75K	4–12 weeks
Data analytics or BI dashboard	$15K – $100K	4–12 weeks
Custom ML model	$25K – $250K	8–24 weeks
Computer vision system	$30K – $300K	8–24 weeks
Enterprise AI integration	$50K – $500K+	12–52 weeks
Ongoing AI retainer	$5K – $50K/mo	Monthly

What changes the price

The state of the data: Clean, structured data reduces preparation work. If the agency must build pipelines, reconcile sources, or clean a messy dataset, a draft budget might carry another 30–50% as a planning allowance. The proposal should replace that allowance with named tasks.

Configuration or custom work: A GPT-4 chatbot using your documents might be planned at $5–15K. A custom-trained model on proprietary data might start at $50K+. Those are different systems even when both are sold as chatbots.

Security and compliance: Work in healthcare, finance, or government may need additional security, audit, and compliance tasks. Keep 20–40% as a planning allowance only until the agency identifies those tasks and prices them.

Team and location: For planning, allow a scenario in which a Silicon Valley proposal is up to 2x a Midwest or remote-first proposal. That is a comparison case, not a market rule or proof that the work is equivalent. Compare the named people, their time, and the scope they will deliver.

What a proposal should separate

A useful proposal shows where the money goes. It separates discovery from the build, data preparation from model work, integration from interface work, and launch from ongoing support. It also says which roles are included and how much of their time the estimate assumes.

Use these hourly figures as role-based planning ranges, not fixed market rates:

AI or ML engineer: plan with $150 – $350/hr until the proposal names the person and rate.
Data scientist: plan with $125 – $300/hr until the role and deliverables are clear.
AI strategy consultant: plan with $200 – $500/hr and define the decisions their work must support.
Full-stack AI developer: plan with $100 – $250/hr and identify the application work included.
Project manager: plan with $75 – $175/hr and confirm the reporting and coordination work.

Hold the scope constant. Give each agency the same requirements, systems, constraints, and acceptance criteria.
Price discovery on its own. A $5–15K discovery phase can be compared with a $100K build decision, but it does not guarantee that the larger project will work.
Use milestones. Tie each payment to a named output and a decision about what comes next.
Read the exclusions. A low total may leave data cleanup, testing, deployment, or internal training outside the quote.
Compare more than the total. Submit one project request only after the brief is specific enough for responses to cover the same job.

Budget for the work after launch

The build price is not the whole cost. Put hosting, model or API use, monitoring, incident work, data updates, retraining, and support on separate lines. A cheaper build that depends on $10K/mo in cloud costs may cost more over time; the proposal should show the assumptions behind that example.

If the plan uses an ongoing AI retainer of $5K – $50K/mo, define what the agency will monitor, fix, and improve each month. “Support” is not a scope.

Browse AI agency profiles to identify firms to question, or send your budget and scope.