[object Object]

A VP of customer service reviews quarterly numbers and notices the deflection rate has been flat at 38 percent for two years. Every IVR upgrade and chatbot rollout moved it by a point, then it drifted back. The Wave 1 agentic self-service capability is the first release in years that has a credible chance of moving that needle past the ceiling, and the pattern looks different from previous attempts.

What’s New

Agentic self-service in Contact Center handles deflectable intents autonomously — account inquiries, order status, password resets, simple billing questions. Escalates only on complexity or explicit request. The agent maintains state across channels, so a customer who starts on chat and switches to voice does not have to re-authenticate or re-explain.

The architectural shift is that the agent owns the conversation, not a topic tree. A traditional bot fails when the user phrases something off-script; an agent reasons from the customer’s intent and the available tools.

Deflection Beyond 40 Percent

Traditional self-service topped out around 30-40 percent deflection. Agentic capability pushes past 50-60 percent for supported intents. Real deflection (problem solved without agent) rather than deflection-as-avoidance. The honest measurement excludes hangups, abandoned chats, and users who eventually called back.

Old metric: any session that did not transfer to an agent
New metric: any session where the customer task was completed

Track both for the first quarter. The new metric will look worse on day one, then improve as you tune the agent.

Assisted Service

When agents engage, AI surfaces context — customer history, similar cases, suggested next steps. Handle time drops while consistency improves. Assisted-service acceleration across every case. The supervisor now sees a confidence score on each suggestion, which lets coaching focus on the cases where the agent overrode AI guidance and the call still went poorly.

Operational Intelligence

Supervisor dashboards surface real-time queue health, intent trends, sentiment shifts. Staffing and coaching become data-driven. Gut-feel management declines. The intent trend chart is the most useful new tile; it shows which intents spiked in the last hour and lets the supervisor pre-staff accordingly.

<fetch aggregate="true">
  <entity name="msdyn_ocsession">
    <attribute name="msdyn_intent" groupby="true" alias="intent" />
    <attribute name="msdyn_sessionid" aggregate="count" alias="cnt" />
    <filter>
      <condition attribute="createdon" operator="last-x-hours" value="1" />
    </filter>
  </entity>
</fetch>

Build this query into a refresh-every-five-minutes dashboard tile and the floor supervisor stops chasing yesterday’s data.

Tuning the Confidence Threshold

The agent decides whether to handle or escalate based on a confidence threshold. The default is conservative, around 0.85. Track misroute rate weekly and tune by 0.01 increments. Going below 0.75 starts to leak edge cases to customers; going above 0.92 erodes the deflection gain.

threshold 0.85: 52% deflection, 1.8% misroute
threshold 0.80: 58% deflection, 2.6% misroute
threshold 0.75: 63% deflection, 4.1% misroute

The right threshold depends on the cost of a misroute in your business. For account password resets the cost is low. For billing disputes it is high.

Channel Coverage

Wave 1 covers chat, voice, and SMS. Email is on the roadmap but not in the box. If 30 percent of your contact volume is email, the deflection lift in this wave only applies to two-thirds of your demand.

What to do this week

Stand up the new agentic self-service in a sandbox with three intents, run the dual measurement on deflection for two weeks, and tune the confidence threshold with weekly reviews. Build the intent trend dashboard tile and put it on the supervisor wall.

[object Object]
Share