For all the discussion about how to interpret first call resolution (FCR), few dispute its importance as a CX indicator. But FCR has evolved with the addition of customer touchpoints and means of contact. Now, AI in contact centers is changing what counts as an interaction in the first place.
Refresher: What is FCR?
FCR measures whether a customer’s issue ends after one interaction with the contact center. In this context, “call” has come to include other touchpoints (email, chat, etc.). SQM says a good FCR rate is 70-79 percent—anything 80 percent or above is “world class.”
In a contact center running AI, a virtual agent and a live agent might both touch the same issue before it’s resolved. That shift is already changing what FCR counts, through records that split one issue into several, or through dashboards that conflate containment and resolution.
Getting a reliable FCR rate for an AI-assisted operation means updating what counts as a contact and a resolution, and scoring every AI and human interaction against those same rules.
FCR Makes Assumptions that No Longer Hold Up
It’s in the name: FCR originated as a measure for traditional call center interactions—an issue arrived as one inbound call, one agent owned it end to end, and that agent logged the outcome before moving on. Operations measure FCR in one of two ways:
Internal Measure:
No repeat contact on the same intent within a set window.
Customer-Reported Measure:
The customer confirms that their issue was handled.
Most run some blend of both, because each method covers the shortcomings of the other. For example, a customer who says on a survey that their issue was resolved may call back two days later. The system needs to accurately reflect the relationship between these two contacts.
Both methods assume the contact center can tell where one contact ends and the next one begins. With AI, that boundary gets harder to define.
AI Spreads One Issue Across Several Records
AI’s first effect on FCR shows up in the records. A single customer issue can now generate a virtual agent session in the CCaaS platform, a knowledge-base query, and a case in the CRM. It sometimes generates a voice call if the customer abandons chat.
Here’s where it gets complicated. Session timeouts create records that look like contacts. Bot intent taxonomies and agent disposition codes often come from different schemas, so the same issue carries two names. Identity stitching can fail when the customer authenticates in one channel and not the other.
These are data-model problems that AI exposes at volume.
Why it matters: Customers don’t see (or care about) a schema mismatch. They experience the resulting impact on their experience, i.e., having to repeat themselves, high-friction handoffs, and so on. For FCR, each of those moments raises the same question about whether the customer is still on their first contact.
Containment Rate Now Answers a Different Question
The second effect shows up on the dashboard. Once a virtual agent takes the first turn, containment rate becomes the headline number on the AI dashboard. It’s easy to read containment rate as resolution rate. But containment counts sessions that ended inside the automated channel. That can include abandonment, a timeout, or a customer who gave up and emailed instead. What’s more, that email’s arrival may itself be logged as a new contact.
Containment is a volume metric. It records where an interaction stopped. Resolution is an outcome metric that requires separate evidence.
It helps to consider containment and repeat contact rate together: if containment rises while repeat contacts hold steady, the same issues are still recurring, just through a different channel. A virtual agent can contain a high share of sessions and still frustrate the customers inside them. Forrester’s recent predictions forecast that a third of companies would harm customer experiences with frustrating AI self-service.
Deflection is harder still to score, because it can leave no record at all. A customer steered to a help article before any session opens never shows up as a contact. If that customer calls the next day, the call looks like a first contact.
Whether contained customers left with their issue settled is a separate measure, which we call containment quality. Scoring it takes follow-up data, and the most reliable signal is the customer who comes back about the same issue inside the window.
A contact center can run the same FCR formula for years and end up measuring something different once AI is added to the mix. Split records can count one issue as two contacts. Sessions that ended without an answer can get counted as resolved. Catching either problem starts with deciding what counts as a contact in the first place.
Define the Contact Before Scoring the Resolution
Redefining FCR starts with four decisions about what counts as a contact. Each one has bearing on the reported FCR rate.
DATAMARK recommends counting a conversation that AI handles from start to finish toward FCR. It can be reported on its own line and held to the same test that’s applied to human-handled contacts—with the same window and the same check for customers returning about the same issue. Otherwise, an AI resolution might look identical to a customer who gave up.
How to Score AI-Assisted Resolutions and Handoffs
With the contact defined, every resolution gets scored under it, whichever path it took. Those resolution paths fall into four categories:
| Resolution Path | What it Covers |
|---|---|
| AI-Only | A virtual agent or self-service flow resolves the issue, and the customer doesn’t return within the defined window. |
| AI-Assisted | A human agent resolves the issue with AI support such as knowledge retrieval or real-time guidance, including issues a virtual agent handed off. |
| Human-Only | An agent resolves the issue with no AI in the workflow. |
| Unresolved | The customer abandons the session or returns about the same issue inside the defined window. |
Agent-assist tools such as DATAMARK’s DataSmart, which pulls answers from a client’s knowledge base during the call, belong on the AI-assisted line.
A workable rule for handoffs: when a virtual agent passes a customer to an agent inside the resolution window, the exchange counts as one contact if a.) the agent receives the conversation history, and b.) the customer doesn’t have to verify their identity or explain the problem again. If the customer has to start over, count it as a second contact against FCR.
Handoffs are likely to only increase. Gartner predicts AI-related regulation will increase assisted service volume by thirty percent by 2028, in part because rules guaranteeing access to a human agent will lead customers to ask for one by default. Many of those requests will start as a handoff from AI to a person, and each needs a clear answer on whether the customer is still on their first contact.
Another consideration: AI assistance changes how agents perform. A field study published in the Quarterly Journal of Economics followed 5,172 support agents and found that access to an AI assistant raised issues resolved per hour by fourteen percent on average. Newer and lower-skilled agents gained thirty-four percent, while experienced agents saw little change. If AI-assisted contacts sit in the human-only line, any lift from the assistant gets credited to the agents. Nobody can see what the AI contributed.
Give the Definition an Owner
Every rule needs someone accountable for it. A single owner for the definitions of a contact and a resolution, across automated and human channels, keeps the virtual agent team and the agent team scoring the same thing.
That owner should settle the definitions before the AI goes live. Gartner found that ninety-one percent of customer service leaders face executive pressure to implement AI this year. When the definitions for a contact and a resolution get written after go-live, the pre-AI and post-AI FCR rates rest on different rules, and the comparison that many AI investments depend on can’t be trusted.
Finance leaders will want that comparison in cost terms. Gartner predicts that by 2030, generative AI cost per resolution will exceed three dollars, surpassing many B2C offshore human agent costs. A cost per-resolution figure built on a loose definition of resolution will understate what AI costs per resolved issue.
Whoever owns the definition needs a view of every handoff, from what the virtual agent captured to what happened once a person picked up. DATAMARK runs contact center operations for its clients, often inside the client’s own technology stack, which puts it on the receiving end of those handoffs. Every agent call is scored with AmplifAi, so the handoff rule can be checked on each transferred call, including whether the customer had to explain the problem again.
FCR is still the clearest read on whether customers had to come back. Keeping it that way in the age of AI requires one definition of a contact and a resolution, applied to every channel, and settled before launch.
Talk to DATAMARK about the best way to put that definition in place for your organization.
