Partner playbook

Plan one client AI workflow from readiness to handover

A readiness checklist and an illustrative roadmap for consultants taking a client from an AI assessment to a service someone runs every day. Start with one workflow, prove it, then add the next.

What should an AI readiness assessment establish?

An assessment earns its fee when it ends in a decision the client can act on, not a maturity score. For a first workflow, it should establish five things: which recurring request the client wants handled, how many of those requests arrive and how long they take today, where the answers already live, who is allowed to approve a change, and who will run the service after you hand it over.

The last one is easy to skip. If nobody at the client, or at its MSP, is named as the operator before the pilot starts, plan on the pilot stalling after the workshop.

Which client workflow is suitable for a first pilot?

Choose a request that arrives often, has a documented answer, and ends in one well-understood action. Good first candidates: how to request approved software, how to get access to a shared application, or a device problem with a known fix. Avoid workflows that need several systems changed at once, or where the policy itself is still being argued about.

Write down the baseline before anything changes: last quarter's volume and the typical time to resolve. Without it, the pilot has nothing to be judged against.

What knowledge, access and device setup does it need?

Knowledge. NetZen answers from the sources you add for that client: documents, files and website pages, scoped to the client or shared deliberately across clients. Each source needs an owner who approves changes. One caution: an article marked Internal is hidden from the client portal but is still available to AI answers, so keep confidential material out of the sources entirely.

Access. Decide who uses the service: employees see their own requests, site managers their sites, client admins the whole client. Choose the channel: the client portal, Microsoft Teams through the client's Microsoft 365 connection, or Slack. If the client wants ChatGPT, Claude or Copilot Studio connected, each person signs in to NetZen and keeps their own role.

Devices. If the pilot action runs on a computer, the devices in scope need the NetZen agent installed and connected before the pilot starts.

Who approves changes and owns the fallback?

For actions started from a ticket, the operator chooses who approves: the requester, IT support, the site manager, or nobody for low-risk actions. The AI cannot grant an approval that a person is required to give. Actions run by hand or on a schedule do not ask for approval, so keep those for the operator's own team.

The fallback matters as much as the automation. When the AI cannot finish, it hands the ticket to a person with what it already checked, and stops. Agree who that person is, their hours and the response time before employees start using the service.

What must the pilot demonstrate before launch?

Four checks, each observed by the client: a known question answered with the right source; a question the knowledge cannot answer, handed to a person; the pilot action run once with the required approval; and a result compared with the baseline. A successful automation run on its own is not proof that the business problem is solved.

How does the assessment become an implementation roadmap?

Who maintains the service after handover?

The operator: the client's IT team, or the MSP that supports the client. Hand over with an upkeep schedule: who updates knowledge when a procedure changes, who renews credentials the automations use, who retests actions after a system update, and who reviews unanswered questions each month. If the client has an MSP, this is the managed AI service it can sell. See how MSPs package and price it.

When is NetZen the wrong platform?

NetZen fits IT support and the work around it: answers from client knowledge, approved actions in a browser, through an API or on a device, and handoffs to people. It is the wrong choice when the client needs a custom AI model, all-purpose Copilot adoption, compliance enforcement or data loss prevention, or wants to keep its current service desk and only add AI on top. Saying so early keeps your recommendation credible.