Ideas that connect.
Work that moves forward.

The announcement and its scope

Microsoft announced Contact Center Intelligence (CCI) on October 21, 2026, in global public preview excluding Austria. Microsoft announcement.

Microsoft Learn describes CCI as a layer providing self-service, quality, authentication and operational insights that complements existing telephony, queues and routing through supported integrations. It warns that preview features may have limitations and are not intended for production. We propose conducting the initial assessment in a test environment. Microsoft Learn overview.

Availability and capabilities: the state of the preview

The table summarizes the descriptions in the announcement, universal connector reference and quality guide. The final column contains checks proposed by Nexo. Actual availability in a specific environment remains unverified.

Integration: defining the work required

The universal connector's Onboard section requires initial provisioning with Microsoft. The token must be issued by the Microsoft Entra tenant that owns the Power Platform environment. We propose including these dependencies in the technical estimate. Provisioning and token.

The guide describes a 30-day trial. Its prerequisites include a work or school account, authority to accept the terms for the organization, and access to an eligible Dataverse environment or the ability to create a trial environment. Capabilities may vary during the preview. We have not checked any particular organization's eligibility. Trial and prerequisites.

For the technical team, we propose verifying the exact combination of vendor, version, channel and environment. The assessment should cover permissions, context transfer to a person, failure handling and the effort required to maintain the connection.

Quality and oversight: checking what is measured

The guide limits scoring to the self-service portion and excludes human handling. Plans do not evaluate retroactively or recalculate existing scores; coverage depends on conditions and sampling. Our interpretation: references to historical conversations do not establish an ability to score any pre-existing archive. Quality limits.

The dashboard's Deflection rate section defines the rate as the percentage of completed conversations that were not escalated to a person and clarifies that this does not necessarily confirm issue resolution. We propose measuring confirmed resolution alongside escalations and handling times, and separately reviewing what each real-time indicator means. Metric definitions.

Cost and data: conditions for an assessment

Microsoft describes consumption through Copilot Credits and possible additional vendor charges for API or voice-channel access. Conditions in the announcement.

The Data governance section of CCI's requirements calls for approval of residency, retention, consent, recording, transcription, responsible AI and data loss prevention requirements. This documentary review does not verify languages by capability, actual data residency, unit rates or the cost of a specific integration. Requirements and data governance.

As a project criterion, we propose separating AI consumption, vendor charges, integration and human review. The pilot budget should set a spending cap and record both total expenditure and cost per case whose resolution has been confirmed.

Editorial interpretation: gradual adoption with clear ownership

Our interpretation is that the assessment should begin with a bounded task, a reference answer, an owner and a route to human assistance. Responsibilities could be assigned as follows:

• Functional team: select a frequent customer request, define a valid answer and establish when a person must intervene.
• Technical team: define data, permissions and actions, and verify connectivity, errors and context transfer.
• Project management: agree acceptance criteria, budget, review owners and conditions for stopping or expanding the trial.

Proposed example: checking an order's status

A hypothetical Nexo scenario that has not been run: a read-only query using synthetic orders, with 20 ordinary cases in Spanish, 20 in English and 10 edge cases involving insufficient data or a request for human assistance. The following targets are illustrative and should be adjusted to the pilot's scope; they are not measured results or industry benchmarks.

What to resolve before expanding the scope

We propose making any expansion conditional on a demonstration in the selected environment, a language and channel matrix for each capability, a review of data handling and measured total cost. The documentary discrepancy over real-time evaluation requires checking its enablement and operation before including it in the acceptance criteria. Hypotheses about savings or service improvements will require comparison with the current process.