Resolution rate is the number every customer asks about first when they log into their AIVA dashboard, and it's also the number most likely to be misread. Across our customer base it sits between 82% and 96%, and the honest answer to "why isn't mine 96%" is almost never "the AI is worse for you." It's usually a decision, made somewhere in setup, that's working exactly as configured.
What counts as resolved
A conversation counts as resolved when it reaches a real, concrete outcome without needing a human to pick it back up — a question answered correctly, an appointment booked, rescheduled, or canceled, a status check completed. That's a stricter bar than the industry's standard first call resolution metric usually gets held to, since FCR commonly counts a call as resolved just for not needing a second contact, whether or not anything concrete actually got settled. It does not count as resolved just because the conversation ended. A customer who stops replying mid-conversation, or who's explicitly transferred to a human agent, isn't counted as a resolution, even though both of those end the conversation the same way a completed booking does.
That distinction matters because it's tempting to build a metric that rewards "the AI kept talking until the customer went away," and we deliberately didn't build that one. A resolution rate that only measured conversation length or absence of complaints would look great and mean nothing.
There's a verification piece underneath this too: a booking only counts as resolved once it's actually confirmed against the business's calendar, not just once AIVA has said something that sounds like a confirmation. If a calendar write fails silently for some reason — an integration hiccup, a slot that got taken in the seconds between checking and confirming — that conversation doesn't get credited as resolved just because the dialogue itself sounded complete. The number is meant to reflect what actually happened, not what was said.
Where the 82–96% range actually comes from
The businesses at the top of that range almost always share one thing: AIVA has real access to their systems, not just a document describing their business. A business that connected its live booking calendar gets real answers to "do you have anything Friday" — check, offer, confirm. A business that gave AIVA a static FAQ and nothing else gets a correct but shallower answer, and more of those conversations end in "let me have someone follow up with you," which is the right outcome for AIVA to reach, but it's a lower resolution number by design.
Escalation settings move the number too, deliberately. A business that sets a conservative escalation threshold — routing anything with a hint of complexity to a human — will show a lower resolution rate than one that lets AIVA work through more of the conversation itself. Neither setting is wrong. They're different choices about risk tolerance, and the dashboard reflects the choice rather than hiding it.
A lower resolution rate isn't always a problem to fix. Sometimes it's a business correctly deciding that certain conversations should always reach a human, and the number is just telling the truth about that choice.
What moves the number in practice
A few specific setup gaps show up repeatedly in the conversations that don't resolve. A service or price change that hasn't made it into AIVA's configuration yet is the most common one — a customer asks about something the business changed last week, AIVA gives the old answer or, correctly, admits it doesn't know and offers to connect them with someone who does. Neither is a resolution, and neither is really an AI failure; it's a configuration lag, and it's usually the fastest thing to fix once a business notices the pattern.
A second common gap is scope creep in what customers ask for versus what AIVA was actually set up to handle. A clinic that configured AIVA for booking and hours might start getting insurance-verification questions it was never given information about. Those conversations correctly escalate, which is the right behavior, but they'll sit outside the resolution number until the business decides to bring that category in-scope and gives AIVA what it needs to actually answer it.
Resolution rate isn't the same across every kind of question
The headline percentage is an average, and averages hide their own texture. A business sitting at a healthy 90% overall can still have one specific category — a confusing cancellation policy, a newly added service the FAQ hasn't caught up with — resolving far worse than everything else, quietly buried under a number that looks fine from a distance. This is exactly why resolution rate is meant to be read next to AIVA's auto-categorized intent breakdown, not on its own. The category view shows which specific type of conversation is dragging the average down, which the single top-line percentage never will.
What resolution rate deliberately doesn't measure
It's worth being clear about what this number isn't. It isn't a satisfaction score — a conversation can resolve cleanly and still leave a customer lukewarm about the experience, and resolution rate on its own won't show you that. It also isn't a repeat-contact check: if a customer's question gets a complete, accurate answer today and they call back tomorrow asking something adjacent, the first conversation still counts as resolved, correctly, even though it's worth noticing if that pattern happens often for the same topic. We don't fold satisfaction or repeat contact into the resolution number itself, because collapsing several different signals into one percentage makes all of them harder to read — the same reason ISO's guidelines for complaints handling treat resolution and satisfaction as separate things worth measuring on their own, not one score standing in for both. A business that wants the fuller picture looks at resolution rate alongside those other signals, not as a replacement for them.
That restraint is also why we've resisted the temptation to make the number look better than it is. It would be easy to loosen the definition — count a transfer to voicemail as resolved, count "the conversation ended" as good enough — and watch the average climb toward 96% across the board. We'd rather keep the bar where it is: a real outcome, verified, or it doesn't count. A percentage that's honestly 82 for one business and 96 for another, with a clear explanation of why, is more useful than a flattering number that means less the closer you look at it.
Does resolution rate differ by channel?
Businesses sometimes ask whether voice, SMS, or the web widget resolves differently, given how different the three channels are to actually use. In practice, the same conversation engine and the same definition of "resolved" apply across all of them — a booking is a booking whether it was confirmed on a call, over text, or in a widget chat, and it's verified against the calendar the same way regardless of channel. What actually moves a business's resolution rate is the setup factors already covered here — system access and escalation thresholds — far more than which channel a given conversation happened to arrive on.
What we want the number to be useful for
The dashboard shows resolution rate in real time, not as a monthly report, specifically so a business can catch a drop while it's still fresh enough to explain — a new service added last week that AIVA hasn't been told about yet, a policy change nobody updated in the FAQ. That's the actual use of the number: not a grade, but an early signal that something upstream of AIVA needs attention.
We'd rather publish a range that's honestly 82 to 96 and explain what moves a business within it than publish a single rounded number that flatters the product and tells a customer nothing about their own configuration. If you want the fuller picture of what "resolved" means and where the definition runs out, we've written a companion piece on the term itself. Both the resolution number and the intent data behind it live on AIVA's analytics dashboard, retained for 90 days by default and exportable anytime — and the same conversation engine driving the number runs identically whether the call came in over voice, text, or the web widget.