AI & Automation 3 min read

Eight questions to ask a Managed Intelligence Provider

The label has no licensing board behind it, so the burden of proof is on the provider. These questions put it there.

The question

We have two proposals in front of us and both providers call themselves Managed Intelligence Providers. One is our current MSP. How do we tell whether either of them actually does this?

A 45-person professional-services firm in a regulated industry, with Copilot already rolled out to half the staff and one department using an AI note-taker nobody vetted.

What we told them

Ask each of them these eight, in writing, and compare the answers side by side. The order matters: the first three are the job, the rest are how they charge and what they refuse.

  1. Show me the inventory you would produce in week one, and how you build it. The right answer is from sign-in logs, web filtering and SaaS discovery, not a questionnaire to your staff. People do not report the tools they are not supposed to be using.
  2. Who holds identity? Every AI tool gets gated the same way: MFA, SSO and Conditional Access in front of it. If the provider does not hold your identity, ask who does and how the two coordinate. A provider who proposes AI into an environment without enforced MFA has told you something.
  3. What is in the monthly report, and who is it written for? It should answer three questions, what is in use, is it safe, is it paying off, and be fit to hand to an insurer, an auditor or your board without a separate project. Ask for a redacted sample.
  4. How are you paid, and by whom? Ask whether they resell the AI vendors they recommend, and whether usage passes through at cost under a cap you set. A recommendation with a margin behind it is a sales call.
  5. What happens when an agent does something wrong? Ask for the runbook: spend caps, who gets paged, how it is stopped. In July 2025 a coding agent working for a small software company deleted its production database during a code freeze, after eleven explicit warnings. The provider's answer should not start with “that would not happen to us.”
  6. What is the price shape? Published US MSP prices in 2026 run $2,500 to $7,000 for an AI readiness assessment and $750 to $2,500 a month for a governance retainer; per-user layers run $10 to $25 a seat. Builds are quoted before they start. Anything priced “per intelligence,” or bundled into one number you cannot take apart, deserves a second look.
  7. What will you not do? A real one has a written list. Ours includes not training foundation models, not building a customer-facing chatbot as a first project, not deploying AI without MFA, not advocating a single vendor and not promising that an AI replaces a person. A provider with no such list has answered the question.
  8. What do I keep if we stop after the audit? The report, the policy and the pilot should be yours. A provider who holds the inventory back to force the next step is selling the next step, not the service.

Then compare. Your current MSP already holds your identity and your endpoints, which is a real advantage on questions two and five. It is not an answer to the other six.

The takeaway

The title costs nothing to claim. Make each provider show the inventory, the identity controls, the report and the written list of what they will not do. The one who has them is the one who does the work.

Got a question like this one?

Send it over — 20 minutes, no pitch, a straight answer either way.