How to Measure a Contact Center: Digital-First KPIs That Matter

Call centers were built to be measured in seconds, and the habit has carried into contact centers where it does harm. An agent who closes a chat in two minutes without solving anything looks efficient on handle time and produces a second contact tomorrow. This page proposes a balanced scorecard with four groups of measures, explains how to set targets per channel, and adds the measures you need once AI is handling part of the volume.

Why Speed Alone Fails

Customers judge an interaction by whether it resolved the problem, not by how quickly it ended. Zendesk’s 2026 research found 85% of CX leaders say one unresolved issue is enough to lose a customer ¹; Metrigy found 38% of consumers will leave after one bad experience ²; and Qualtrics found 53% say a bad experience will cause them to cut spending with a brand (November 2024) ³. A scorecard that rewards speed without resolution optimizes for the wrong outcome.

The Balanced Scorecard

Resolution

• First-contact resolution (FCR) by channel: the share of contacts resolved without a follow-up in the same or another channel. Define “resolved” explicitly and apply the same definition across channels.
• Repeat-contact rate within seven days: the share of customers who contact again about the same issue. This is the honest check on FCR.
• Transfer and channel-switch rate: how often a contact had to move to another agent or channel to be resolved.

Responsiveness

• First response time by channel, with targets matched to the channel: seconds for phone, under a minute for chat, hours for email and social.
• Service level for voice (share answered within a target) and abandonment rate, kept for voice and chat only.
• After-hours coverage: the share of after-hours contacts answered live or by a digital agent rather than left to voicemail.

Experience

• Digital CSAT collected in-channel, split by channel and by whether a person or a digital agent handled the contact.
• Quality score from a QA rubric that includes a “did not make the customer repeat information” item.
• Customer effort, measured by a single post-contact question where possible.

Cost

• Cost per resolved issue, not cost per contact. A cheap contact that creates a second contact is not cheap.
• Cost by channel and by handler (person or digital agent) so the economics of automation are visible rather than assumed. The 2019 Gartner benchmark often quoted, USD 8.01 per live contact versus about USD 0.10 for self-service, is seven years old and should be used only as a reminder of the gap, not as a target ⁴.

Adding AI to the Scorecard

Once a digital agent handles part of the volume, add five measures so the scorecard tells you what the AI is doing rather than hiding it inside averages:

The last two measures matter because of what customers say and what regulators are beginning to require. Consumers’ top annoyance with automated service is the inability to reach a human, named by 47% in Verizon Business research among 5,000 consumers ⁵, and California’s AB 1609 now requires the largest businesses to make a good-faith effort to connect a requesting customer with an agent within 15 minutes ⁶. For a front desk, the right target is seconds.

Setting Targets Per Channel

Set one resolution standard for the whole operation and different responsiveness targets per channel. Review the scorecard monthly by channel and by handler type, and resist the temptation to roll everything into one number. The purpose of the scorecard is to show you where the customer journey breaks, and averages hide breaks.

How Redial Fits

Redial builds the scorecard with the client before launch, agrees the definition of “resolved” per queue, and reports by channel and by handler. For programs that include the AI front desk we are previewing, IzzyOps (working name), the five AI measures above are the ones Redial intends to instrument from the first pilot call, so that the claims we make about answer rate, handoff rate, and context carried are measured rather than asserted.