What Every Interaction Leaves Behind: Inside Cisne’s WFM Data Integration with Avaya Infinity

Almost everything written about AI in the contact center focuses on the conversation itself: latency, transcription, and whether the model knows when to stop talking. Workforce management sits outside the conversation.

Nothing we do happens while the customer is on the line. We rely on what every interaction leaves behind, across every channel and type of work. Calls, emails, chats, and back-office tasks handled by human agents and, increasingly, AI agents. All of it has to reach us, and it has to be complete.

That may sound like the easier assignment. It is not.

We integrated Cisne with Avaya Infinity earlier this year. It was the first platform in a long time to hand over all of it without an argument.

In my first blog, I explained what the integration means for contact center operations. Here, I want to focus on the data and technical decisions behind it. 

What Data Does a WFM Integration Need?

A WFM integration needs historical contact statistics, timestamped agent-state events, administrative data about agents and queues, and information about non-voice and back-office work. These data sources must reconcile so forecasting and adherence reports start from the same operational record. 

Start with interval-level contact data by queue that ties out to what the automatic call distributor (ACD) reported, across every channel rather than voice alone. 

Agent-state events need actual timestamps, not summaries: logged in, available, on contact, in after-call work, and away, with the applicable reason code. Underneath that sits the operating taxonomy: queues, skills, attributes, and the mapping between platform language and business language. 

Back-office and non-voice work also have to be included, which is where many platforms quietly stop. 

Now there is another category, one that did not exist when most WFM systems were designed. Some volume is handled by AI agents rather than people. Without visibility into which contacts AI handled, at what volume, and with what handle time, you cannot forecast the human capacity still required. You also cannot quantify what the AI is saving.

AI-handled volume is now a staffing input. 

Finally, all of it has to reconcile. If interval-level handle time does not align with the agent-state timeline, the forecast and the adherence report tell two different stories.

Three Interfaces for Historical, Real-time, and Administrative Data 

Avaya Infinity separates the three things WFM needs, which sounds obvious and almost never is.

Authentication uses a client ID and shared secret with role-based access, issued separately for each tenant. Our connector asks for the base URL, account ID, and credentials. 

Historical contact center statistics 

Historical statistics come through a dataset API. Queue data is available in fifteen-minute, thirty-minute, forty-five-minute, and daily intervals. It includes contacts offered and answered, abandon rate, average handle time, and volume by channel. 

The metrics arrive already calculated. We are not reconstructing queue performance from raw interaction records and then explaining why our AHT is four seconds off from the number on the customer's dashboard.

Real-time agent-state events 

Real-time data comes through a notification service using webhook subscriptions. It includes agent-state changes across channels and media types, login and logout events, busy states, and reason-code changes. Those events drive live adherence.

Two event classes are worth understanding going in. User events cover roughly ninety percent of what adherence requires. Interaction events carry hold time and the idle-versus-busy picture.

Agent and queue administration 

The third interface is an administrative API for agent profiles. Over REST, it handles queue assignments, proficiency by queue, and tags, which work the way skills and routing attributes do on other platforms.

Maintaining users separately in a contact center platform and a WFM platform eventually creates discrepancies. Always. Synchronizing through an API instead of relying on manual updates prevents an entire category of support tickets before they happen.

How Mapping and Edge Cases Affect WFM Data Accuracy 

Anyone can produce a list of APIs. What made this integration work is in the details underneath them.

Reason codes are configurable by the customer and map cleanly to WFM activities. If a customer defines Comfort Break as an away state and Team Call as a busy state, those distinctions remain intact through adherence. They do not collapse into a generic not-ready category that tells a planner nothing.

Wrap-up can be fixed or agent-controlled, and Avaya Infinity makes clear which one applies. That sounds like a footnote. It changes how handle time is calculated, and handle time feeds everything downstream.

The edge case I would tell anyone to examine, on any platform, is redirect on no answer. A contact is offered to an agent, the agent does not answer, and the platform pulls it back and places it in the queue again. Without the right logic, that same contact can be counted more than once, sometimes across intervals or queues.

Handle it correctly, and offered volume reflects real demand. Miss it, and the forecast begins learning from demand that never existed. The result stays wrong in a way that is difficult to trace back to its source.

None of this required a workaround. It required reading carefully and asking the right questions, which is a much better position to be in.

How Cisne Tested WFM Data Reconciliation 

Testing a WFM integration is mostly about disagreement. Historical statistics and real-time events describe the same day from different directions, and the job is to determine whether they tell the same story.

Do the historical queue statistics reconcile with the event stream, or do the forecast and the adherence report start from numbers that do not match?

An interaction that starts at 14:58 and ends at 15:04 has to land in the same interval every time. Interval boundaries look minor until the same interaction is attributed differently depending on which interface reported it.

Do concurrent chats break occupancy calculations? An agent handling two interactions at once cannot be measured as though the work happened sequentially.

When a contact goes unanswered and returns to the queue, it must be counted once. Otherwise, one customer creates demand twice.

Those are small cases with large downstream effects. A discrepancy that looks negligible in one interval becomes part of the history used to forecast the next week, month, and season.

None of the tests exposed a gap.

What the Cisne Integration Supports Today 

Voice, email, and chat all flow through the same integration today, not on a roadmap. If an interaction happens, Cisne can forecast it, schedule for it, and show how actual performance compares with the forecast, both in real time and historically.

Avaya Infinity Elements are JavaScript and HTML widgets embedded in the agent desktop, while custom tabs allow a complete web interface to run inside the platform. For WFM, that means agents can view schedules, monitor adherence, and submit shift-swap requests without leaving the place where they already work. Supervisors can remain there as well.

What Comes Next 

The next milestone is service-level attainment. When it lands, answer thresholds will be configurable by queue and channel, allowing Cisne to forecast and schedule directly against the service target the business has committed to.

That level of granularity is not common, and it matters. Forecasting to a service-level target is the job, which is why I know this is one people are watching for.

What the Integration Reveals about Avaya Infinity 

An integration is a fair way to judge a platform because it is difficult to fake. Either the data arrives in a usable form, or you spend months building scaffolding around everything that does not.

Interval history, agent state, user and queue taxonomy, and non-voice channels all came from the same place. The interfaces were documented and designed to be integrated against, rather than opened up after the fact.

That is the difference between an integration and a project.

Cisne is now live on Avaya Infinity through Avaya's Innovation Hub. Once the credentials are in place, the connector takes minutes to configure.

Explore Avaya Infinity