BusinessMCP
Academy

Module 5

The AI business analyst

It analyses freely, acts only when asked, and never spends money.

17 min2 lessons

After this module: You can demo the Assistant confidently and answer "what stops it doing something stupid?" without hedging.

What it will and will not do

Most AI products are sold on what the AI *can* do. Ours is sold as much on what it will not do, and once you see why, this becomes the easiest part of the demo.

Buyers are nervous about autonomous AI touching their customers. An AI that emails prospects unsupervised is a brand risk, and a sensible buyer knows it. So the constraints are not an apology — they are the reason it is safe to turn on.

The capability model
CategoryWhat happensExamples
AnalyseFreely, no approval neededAnalytics, SEO, ads, social, CRM, revenue, company data, live web research
ActOnly when asked, and gatedDrafting an email for approval, updating a contact, building an audience
NeverNot available at any settingLaunching or funding ad campaigns, auto-posting social content, unattended drip campaigns

Two specifics worth memorising. Sending email is approval-gated: the analyst drafts, you see the recipient and the full body, and nothing leaves until you approve. It never spends money on ads — it will build and sync an audience on request, but it cannot create or fund a campaign. That boundary is deliberate and it is a good answer to "what if it goes wrong at 3am?"

Check yourself · 1 of 2

Which of these will the AI never do, at any setting?

Check yourself · 2 of 2

A prospect says "I do not want an AI emailing my customers behind my back."

Assistant and Insights: pull and push

The same intelligence reaches the customer two ways, and knowing which to demo when matters.

  • Assistant is pull. The customer asks a question in plain English and gets an answer grounded in their own numbers, with a live timeline showing which reports it pulled to get there.
  • Insights is push. A daily analysis surfaces things worth attention — a conversion rate moving, leads going stale, a page leaking visitors, a sending domain at risk. Each card clicks straight through into a full analysis.

The tool timeline is the thing to point at in a demo. It shows the analyst pulling the specific reports and date ranges it used. That turns an answer from "the AI said so" into something a sceptical buyer can audit — and scepticism is the default reaction to an AI making claims about money.

It is also model-agnostic. A customer can connect their own frontier model or use the included free tier. For a buyer already paying for AI, "bring the model you already trust" removes a procurement conversation.

Finally, workspace memory is opt-in. With it on, the analyst distils durable facts about the business from each conversation into a living document it reads next time, so it stops re-asking who the customer sells to.

Check yourself

Why is the tool timeline worth pointing at during a demo?

Worth remembering

  • The capability model is the sales asset: analyse freely, act only on request and gated, and a short list of things it will never do.
  • Sending email is approval-gated — it drafts, a human approves, then it sends.
  • It never launches or funds ad campaigns, never auto-posts social content, and never runs drip campaigns on its own.
  • Insights is the same intelligence pushed the other way: a daily feed of things worth looking at, each one clickable into a full analysis.
  • It is model-agnostic. A customer can bring their own frontier model or use the included free tier.

Words that came up

Approval-gated
The AI prepares an action and stops. A human sees exactly what would happen and approves before anything leaves the system.
Tool timeline
The live list of what the analyst pulled while answering — which reports, which date ranges. It is what makes the answer auditable.
Workspace memory
An opt-in living document of durable facts about the business, so the analyst stops re-asking the same questions.

Do this before moving on

Reading about a demo is not the same as having given one.

  1. 1Ask the Assistant a question you do not know the answer to, and watch the tool timeline while it works.
  2. 2Find an approval card and read what it would send before approving or rejecting it.
  3. 3Open Insights and pick the card you would lead a customer call with.

Go deeper