AI Consulting Services: What Is Included and What Is Not

AI consulting services usually include finding use cases, a readiness check, a roadmap, a pilot, building, rules for staff, training, and support after launch. Ask for the deliverable behind each one in writing before you sign.
Below: what each service includes, what is usually left out, and which to buy first. To see who delivers this work, read what an AI consultant does. The AI consulting page covers the services I offer.
What do AI consulting services include?
Most packages are built from the same eight parts. The “what you should get” column is my own checklist, built from the sources at the bottom of this page. It is not a standard.
| Service | What you should get | Question to ask |
|---|---|---|
| Discovery | A ranked list of tasks worth automating, with the time they take today | How did you rank them? |
| Readiness check | A short report on your data, tools and people: what exists, where it lives, what blocks you | What would stop the first project? |
| Strategy and roadmap | A sequenced plan: first project, next two, an owner for each | What happens in the first 30 days? |
| Pilot | One working workflow on your own data, with a pass or fail test | How will we know it worked? |
| Build and integration | That workflow connected to the tools your team already uses | Who owns the accounts and files when we finish? |
| Governance | A written policy: what staff may do, what data may go where, who approves | Who reviews it when a tool changes? |
| Training | Live sessions plus written how-tos | Will a new hire be able to use it without you? |
| Support | Fixes and updates after launch | Is this in the scope or a separate agreement? |
Two firms’ public pages describe the same pieces in different words. Monterail’s glossary page lists stakeholder interviews, a ranked use-case list, a data readiness assessment, a prototype tested on the client’s data, a production roadmap and team training. Centric Consulting’s Joseph Ours, writing on October 19, 2026, adds that a consultant also helps you decide when to build something custom and when to adopt a proven tool. Both are vendors describing their own service, so treat them as a menu, not a standard.
I’ve noticed that the biggest firms in the search results sell data platforms and custom model builds. A company with 20 people rarely needs that on day one. The next sections show which parts matter at your size.
What does an AI consulting engagement look like from start to finish?
It runs in stages with a stop-or-go decision after each one. That keeps the first commitment small.
- Discovery. Someone interviews the people who do the work and maps what they do today.
- Readiness check. They look at your data and tools before anything gets built.
- Pilot. One use case is built on your data and tested.
- Production. If the pilot passes, the workflow is connected, monitored and handed over.
- Rules and review. Staff rules are written and someone owns the system after launch.
Monterail publishes its own timelines for this path: 2 to 4 weeks to diagnose, 4 to 6 weeks for a prototype on your data, and 8 to 16 weeks for a production build. That is one firm’s range for custom builds, with no date on the page, read on November 15, 2026. A small business automating one workflow and training a team can move much faster, because there is less to connect.
What a first call sounds like matters too. When I run discovery, I play back what I heard to make sure I have it right: you have a process, it is pretty manual, and you are copy-pasting into Claude and ChatGPT and then copy-pasting back. If a consultant cannot say your process back to you in plain words, they have not understood it yet.
What is usually not included?
Ask about this before you sign, because it is where surprise bills come from. Monterail’s page lists four common exclusions:
- Software and cloud license fees for the AI models
- Large data labeling or data entry jobs
- Long-term running of the system after handover, unless you agree a separate support contract
- General IT or business-process consulting
Add three of my own to the question list. Who pays for the tools your team will use every month? What happens when a tool changes its settings or its price? And who fixes the workflow if it breaks in month four?
My rule of thumb is that where AI is not ready yet, the report should say so, and nothing should go live without your approval. A consultant who only ever says yes to AI is selling, not advising.
Which AI consulting services should a small business buy first?
Start with discovery and one pilot. Then add a short rules document and a training session. Platform work and large data clean-ups come later, if a pilot earns them.
The US Census Bureau reported that fewer than 20% of firms with four or fewer employees used AI between December 14, 2025 and May 3, 2026, against 37% of firms with 250 or more (published May 26, 2026). Most small firms are still at the start. That is fine. It means the first service you need is usually not a roadmap. It is someone to find the one or two tasks where AI helps this month.
Here is how I begin with a new team. I ask what the biggest time-consuming tasks are and which problems most need fixing. Then I look for small, easy, quick wins. When a team sees how fast something gets built, they say wow, it was that easy, and they want to automate everything else.
Here is what I believe: not every problem is an AI project. Sometimes it is just an inventory system, and even a great system is pointless if people will not use it. A consultant who says that on the first call is worth more than one who finds an AI use for everything.
Not sure where your business stands? The AI readiness assessment guide helps you check before you spend.
Is it advice only, advice plus building, or training?
Consulting services come in three shapes, and the price and risk differ for each. Know which one you are buying.
| Shape | You get | Fits when |
|---|---|---|
| Advice only | Analysis, a roadmap, a vendor shortlist | You have a team that can build |
| Advice plus build | The above, plus a working pilot or workflow | You want results, not a report |
| Coaching and training | Your team learns to build and run it | You want the skill to stay inside the company |
I do all three. Some clients want a coach, some want me to build and run it, and many want a mix. My AI workshops cover the coaching and training side, and my custom AI apps cover the building side. The right one depends on who in your company will own the work after the engagement ends. If nobody will, advice alone will sit in a folder. For how to pick between suppliers, see how to choose an AI consulting firm, and for local options, AI consulting near me.
How do you scope the work?
Scope it on money and time, not on technology. I scope AI work by asking how you do things today, how much time and manpower that takes, and how much it costs you. The answers give you a baseline. Without one, nobody can say later whether the project paid off.
Bring three numbers to the first call:
- Hours. How many hours a week does the task take, and who does it?
- Cost. What does that time cost you, including any tools involved?
- Error rate. How often does it go wrong, and what does a mistake cost?
Then ask each bidder to write down what success looks like in those same three numbers. For what consultants charge in the market, read AI consulting rates. For the measuring side, see AI ROI.
What about rules and governance?
Any service that touches staff or customer data should come with written rules. The US National Institute of Standards and Technology published version 1.0 of its AI Risk Management Framework on January 26, 2023, as a voluntary guide to managing AI risk. A small business does not need to adopt it in full. A one-page policy that says what data may go into which tool, and who approves new tools, covers most of the early risk.
Governance is also where services differ most. Some firms deliver a policy document. Others only mention it in a slide. Ask to see a sample, and read the AI governance guide for what a sensible first version holds.
How do you tell a real service from a buzzword package?
Run any proposal through four checks.
- Deliverable test. Every line item names something you can hold: a list, a workflow, a document, a recording. “Strategic alignment” is not a deliverable.
- Own-use test. The consultant can show AI running in their own business.
- Exit test. You own the accounts, prompts, files and documentation when it ends.
- Argue-back test. Before I adopt anything new, I make Claude argue against its own recommendation, weigh the pros and cons of each option as a neutral third party, then re-compare and tell me what is best. Ask a consultant what they would argue against in their own proposal.
Centric’s article, citing McKinsey’s 2025 State of AI survey, says nearly two-thirds of respondents were experimenting with AI but only 7 percent had fully deployed and integrated it across their organizations. I could not confirm that figure at McKinsey directly, so treat it as Centric’s reading. The point stands either way: many pilots stall, and a service that ends at the pilot leaves the hard part to you.
What should you do next?
- List your five biggest time sinks and the hours each takes.
- Pick one where the work repeats every week and the data is not sensitive.
- Ask two or three consultants for a written phase-one proposal with one deliverable per service line and the exclusions listed.
- Compare on deliverables, not hours or buzzwords.
If you want help with step one, the AI consulting page shows how to start a conversation.
Related guides
- What an AI consultant does, and when you need one: the job and a five-question test
- AI strategy: decide where AI goes in your business
- AI use cases: examples to start from
- AI automation: what you can hand off first
- AI training for employees: a plan for a company
- AI for small business: where to start
Sources
- Monterail, AI Consulting Services: What They Are & When They Fit, no date shown, read November 15, 2026.
- Joseph Ours, Centric Consulting, What Goes Into AI Consulting Services?, published October 19, 2026, read November 15, 2026.
- US Census Bureau, Large Firms With at Least 20 Employees Biggest AI Users, published May 26, 2026, read November 15, 2026.
- NIST, AI Risk Management Framework, version 1.0 released January 26, 2023, read November 15, 2026.


