After Microsoft Contact Center Intelligence, what counts as a completed service request?

Public developments and original analysis · Announcement date: 2026-10-21 · Sources

Microsoft announced the public preview of Contact Center Intelligence (CCI) on October 21. It aims to add AI service, quality evaluation and analytics to companies’ existing contact centers. For a business not ready to migrate its contact-center platform, that offers a route worth investigating: introduce AI within the current operation, then decide how far to extend the change.

An easier integration route does not make service outcomes easier to establish. Microsoft’s real-time dashboard documentation draws a precise distinction: deflection counts completed conversations that were not escalated to a human, and does not necessarily confirm resolution. That distinction matters when a company assesses the return from AI service.

Deflection is useful. Managers need to understand how many conversations reach a human queue, plan staffing and compare self-service performance. The mistake is extending “no human transfer” into “the customer’s request was completed.” One describes how the interaction was handled; the other depends on what the caller needed to accomplish.

An accurate answer may finish a query about opening hours. Changing an appointment requires a record to change. Compensation outside an established policy may require an authorized person’s decision. Combining these requests into one deflection target hides the differences between them. A better comparison uses similar requests under comparable conditions and checks whether the associated work progressed.

Verifying a rescheduling request against dispatch records

Suppose a customer asks to move a home repair from Wednesday to Friday. The AI understands the date, confirms the address and ends the call politely, but dispatch still shows Wednesday. The customer believes the change is complete, while the engineer may follow the original schedule. This is a hypothetical example of outcome assessment, not a reported Microsoft product fault.

Two results need separating: understanding what the customer wants, and the business accepting the new arrangement. The conversation can provide evidence of the first. The second requires finding the original appointment, checking availability and examining the saved record. If Friday is full, discussing another slot is more useful than a fluent assurance that the change is complete.

Natural conversation and accurate processing therefore need separate checks. An agent can correctly repeat “you would like Friday.” It can say “the request has been submitted” when the request reaches the responsible process. “The appointment has been changed” needs a confirmed arrangement. The wording should follow the actual state of the work; an unjustifiably confident answer can lead the customer to act on the wrong schedule.

A company evaluating AI service can retain both conversation and appointment records for a set of comparable changes and examine whether they agree. A date spoken correctly in a demonstration is not sufficient evidence. Nor should every unfinished request be attributed to poor model understanding. No available slot, a pending approval and a failed update need different owners. Knowing where the work stopped helps determine whether to change scheduling rules, the handoff or the integration itself.

Microsoft’s current quality-assurance documentation covers the self-service agent’s portion of the conversation; the human-handled portion after escalation is excluded from that score. Questions about acknowledging the request and appropriate language can be useful quality measures. They do not, by themselves, establish that dispatch changed. These are preview documents whose capabilities and scope can change, so an evaluation needs the version actually available.

For Mokyun, a practical scope is connecting that customer request back to the existing operation. Retrieve the current arrangement for the same appointment or work order, submit the proposed change through established dispatch or approval steps, and check the saved result. Where only submission is possible, make the next owner and confirmation route clear. An existing API, file exchange, data view or authorized interface operation would be chosen for the specific system. This is an integration approach, not a claim that Mokyun has a CCI connector or has validated every scheduling system.

Counting automated steps would miss part of the value. A dispatcher receiving the original booking, the customer’s acceptable times and the reason for the wait need not ask the customer to repeat everything. Even when the system cannot make the change, it can help route the problem to the right person. For a service with complex scheduling rules, improving that handoff may be a better first choice than pursuing an entirely unattended process.

Which requests need a human handoff

Microsoft’s announcement also describes deterministic paths, generative interactions or a combination, with human handoff when the organization’s criteria call for it. A company need not use the same approach for every request. A factual query, a booking change and an exception requiring discretion can have different automation boundaries and different evidence of completion.

For example, the system may verify identity and find the booking, then discover that none of the available slots works for the customer. Human coordination could be the right next step. The useful questions are whether the customer avoids repeating information, whether the coordinator receives the necessary context and whether there is a subsequent response. A transfer alone should not make that interaction a failure. Conversely, a conversation that avoids transfer but leaves the customer calling again should not earn approval solely from deflection.

This affects the financial comparison too. Microsoft says CCI uses Copilot Credits, excluding third-party contact-center API or voice-channel licensing charges. Integration, information maintenance and human exception handling also need consideration. Low-volume, simple queries already handled adequately by existing tools may not justify another solution. If repeat calls arise from a disconnect between appointments and customer service, improving the spoken answer alone may not address the main problem.

Our preference is to choose an initial AI service scope around a request whose completion can be explained. A query needs an accurate, usable answer. A rescheduling request needs the confirmed arrangement. A matter requiring discretion needs the right owner and continued handling. Deflection can still describe workload, while business records and subsequent service show the outcome. Using each for its proper purpose helps a company decide which work merits automation and which human effort it should retain.

• Mokyun system integration: connecting existing business records

Mokyun Editorial · neo@mokyun.com