There is a number on your operations dashboard that says something like 78%. It is your first-contact resolution rate, it has been broadly stable for four quarters, and nobody has looked at it properly in eighteen months because it looks fine. It looks fine because of how it is produced, not because of how your calls go.
Where does your first contact resolution number come from?
Almost every FCR number in existence comes from a post-call survey: the IVR asking the customer to press 1 if their issue was resolved, or the email that follows an hour later. Post-call surveys typically return response rates in the 15 to 25 percent band. So the number is built from a fifth of your calls, and it is not a random fifth.
Answering an optional survey costs the customer effort they have no reason to spend. The people who spend it anyway are the ones with a reason: they are delighted, or they are furious. Everyone in the vast, unremarkable middle — issue mostly handled, slight lingering doubt, will probably call back on Thursday — hangs up and does nothing. That group is precisely the group FCR is supposed to detect, and it is the group least likely to respond.
This is a sampling problem dressed as a measurement. You are not measuring resolution. You are measuring the resolution experienced by people with strong enough feelings to fill in a form.
The second-order problem: the customer is guessing
Even when they do respond, you have asked the wrong witness. Press 1 if your issue was resolved is a question the customer often cannot answer at the moment it is asked.
They were told a refund would appear in three to five working days. They believe it. They press 1. On day six there is no refund, and they call again. The survey recorded a resolution. The account records two calls about one problem. Only one of those is true.
The failure runs both ways. A customer who is annoyed about something unrelated — the wait, the transfer, the hold music — will report the issue as unresolved when it was resolved on the first call. FCR by survey blends resolution with satisfaction and reports the mixture under the name of the first one.
How do you measure first contact resolution without a survey?
Here is the awkward part. You do not need to ask. Resolution has an observable signature, and it is sitting in records you already keep.
A repeat contact is the same customer, about the same subject, within a defined window. Where those repeats come from is usually a routing problem rather than an agent problem. Three components, all of which you can compute:
- Same customer. Not the same phone number: the same person. Numbers are shared, spoofed, and changed. Match on the account or the case wherever you have one, and treat number-matching as the fallback it is.
- Same subject. This is the component everyone skips, and it is the one that decides whether the metric means anything. A customer calling about a billing error on Monday and a delivery date on Wednesday has not had a resolution failure; they have had two conversations. Without topic matching, your repeat-contact rate is really a call-frequency rate.
- Within a window. Not within twenty-four hours, which catches almost nothing. Seven days is the usual honest choice, and it should be set by how long your promises take to come due. If you tell people three to five working days, a seven-day window catches the broken promise and a two-day window guarantees you never will.
Get those three right and first contact resolution stops being an opinion poll and becomes a join across two tables.
| Subject | Survey said | Called back, same subject | Gap |
|---|---|---|---|
| Refund timing | Resolved | Yes | 6 days |
| Delivery date | Resolved | Yes | 3 days |
| Plan change | No response | No | -- |
| Billing error | Resolved | Yes | 5 days |
| Password reset | No response | No | -- |
Why topic matching is the whole job
Same customer and same window are database queries. Same subject is the part that historically stopped teams doing this, because it required either a human to read every call or a disposition code the agent selected under time pressure at the end of one.
Disposition codes are not a solution. The agent picking them is measured on handle time, the dropdown has forty entries, and the eleventh is Other. Any metric built on that field inherits its noise, which is why teams that try this with disposition data conclude the approach does not work.
Reading the transcript works, because the subject of a call is not hidden in it. Callix reads every transcript and groups calls by what was actually discussed. It is stated, usually in the first thirty seconds, by the customer, in their own words. A repeat-contact metric built on what was actually discussed is measuring the thing you meant to measure. One built on a dropdown is measuring how tired the agent was.
What changes when the number is real
The reason to do this is not a more accurate dashboard. It is that a survey-based FCR number cannot be acted upon, and a data-based one can.
A survey tells you 78% and stops. It cannot tell you which topics drive repeats, because the respondents are self-selected and too few to segment. The moment you compute resolution from call data, you have every call, so you can group by subject — and the distribution is never flat. Repeat contacts cluster. Two or three topics generally produce most of them, and they are usually not the topics your coaching programme is focused on, because coaching is aimed at where scores are low rather than where customers come back.
That is the practical difference. One number ranks your agents. The other ranks your problems. That reordering is what the operations teams we work with notice first.
Where this leaves the survey
Keep it, but stop asking it to do this job. A post-call survey is a reasonable instrument for satisfaction, which is genuinely subjective and genuinely requires asking. Resolution is not subjective. It is an event that either did or did not happen, and it leaves evidence. If you want it computed on your own calls, see it on a sample.
The one-line version: ask customers how they felt, and ask your call data what occurred.
Two honest limits. A repeat-contact measure cannot see the customer who gave up rather than calling back, and that silence looks identical to a resolution. And any first contact resolution number computed this way is only as good as your customer matching; operations without a reliable account identifier on inbound calls will undercount repeats and should say so on the dashboard rather than in a footnote.