Example / Customer follow-up

Customer follow-up in ChatGPT starts with one approved change.

A useful first AI workflow does not need to replace the CRM. It needs to remove the small gap between a clear customer request and the approved update that already belongs in the product.

Direct answer

For a customer follow-up, let the conversation find one account, show only the relevant context, propose the next step, and require a clear confirmation before anything is changed.

Reference workflow. It describes an implementation pattern, not a customer result or a deployed client integration.

01

01 / The user job

Name the action that ends the loop.

“Follow up with this customer” is too broad to expose as a single black-box command. The product team should decide what the finish line means: a saved CRM note, a scheduled meeting, a drafted message, or a task for an account owner.

That sentence gives the workflow a boundary. It also makes it possible to tell a person exactly what changed when the conversation is over.

  • Find the account from a stable identifier.
  • Read a limited set of recent activity and ownership fields.
  • Prepare one proposed next step before a write action.
02

02 / The tool surface

Keep the model-facing actions small and named.

The model does not need every CRM endpoint. A first version may need only a read action, a draft action, and one confirmed write action. Each tool should say what it needs, what it returns, and which product rule can reject it.

That is easier to test than an assistant that is asked to “handle follow-ups” with an open-ended API connection.

Model Context Protocol · tools specification
03

03 / Confirmation

Show the proposed write before it happens.

The conversation can summarize the account, date, owner, and message before it asks the user to continue. The product then applies the same authorization and validation rules it would apply outside chat.

A confirmation point is not a decorative extra. It gives the person a chance to notice a wrong customer, a missing detail, or an action that should remain with an account owner.

04

04 / Result

Return a record a person can inspect.

A good response says what was updated, what was scheduled, and what still needs attention. It links back to the source record when that is useful, rather than leaving the user with an untraceable chat response.

The CRM remains the record of work. ChatGPT is the request surface, not a second system of truth.

Continue reading

Order support in chat

Apply the same pattern when policy and order state must be read before a proposed support action.

Product interface

See how a product keeps authorization and business rules while AI becomes the request surface.

Start a scoped estimate

Bring one workflow, the affected systems, and the host you want to reach.

AI / Second opinion

Ask your AI first.

Use this prepared question to assess where an AI Product Interface could remove friction from your product.

“I run a software product. Help me identify one high-value workflow customers could complete through an AI Product Interface. Ask me about the product, the user, the action, required data, permissions, and the safest small first release.”

The prompt is copied as a backup. Some AI hosts may ask you to paste it after sign-in.

FAQ04

Questions, answered.

Is this only an MCP server?

No. The MCP layer is one part. We also scope the product actions, permissions, host-specific behavior, UI, testing, and release path.

Do we need to rebuild our product?

Usually no. The interface sits in front of approved capabilities in your existing product. We start with a narrow workflow and expand from evidence.

Will it work in ChatGPT, Claude, and Gemini?

The shared architecture can support all three. Each host still has its own UI, authentication, approval, and publishing rules, so we verify them separately.

How do you keep actions safe?

We define explicit tools, validate inputs, keep user approval visible, and preserve the product's existing authorization rules.