Resolution Rate vs. Deflection Rate: What Your Fin AI Numbers Are Actually Telling You

The Number That Gets Thrown Around the Most

When business owners ask me how their Fin AI setup is doing, they almost always lead with one number: the resolution rate. It's the headline figure Intercom puts front and center, and it's easy to treat it like a report card. The trouble is that resolution rate and deflection rate get used as if they mean the same thing, and they don't. If you're making decisions about staffing, pricing, or how much more to invest in tuning Fin, you need to know exactly which number you're looking at and what it actually counts.

What Fin AI's Resolution Rate Actually Counts

Inside Intercom's Fin AI Agent Report, resolution rate is calculated only across the conversations Fin was involved in, not your total conversation volume. That distinction matters more than it sounds. If Fin is only picking up a third of your incoming conversations and resolving most of those, your resolution rate can look great while your overall support workload barely moves. Intercom tracks this separately as an involvement rate, and reading resolution rate next to involvement rate gives you a far more honest picture than either number on its own.

Inside the resolution count itself, Intercom splits results into two categories: confirmed resolved and assumed resolved. A confirmed resolution means the customer gave positive feedback after Fin's last answer. An assumed resolution means the customer simply left the conversation without asking for a teammate and without leaving negative feedback. That second category is doing most of the work behind a typical resolution rate, and it's worth sitting with what it actually means: the customer didn't complain, but nobody confirmed the issue was actually fixed. The report also tracks escalated conversations, which were handed to a teammate, and abandoned ones, where the customer left before Fin could even answer.

Where Deflection Gets Confused with Resolution

Deflection, as a general concept across AI support tools, measures whether a conversation was handled without a human ever stepping in. On its own, it says nothing about whether the customer's actual problem got solved. A customer who gets frustrated and gives up mid-conversation can still count as deflected in a lot of reporting setups, simply because no teammate touched the thread. This is the gap that trips people up: deflection tells you how much work the AI kept off your team's plate, while resolution, even the assumed kind, is at least trying to measure an outcome rather than just handling.

Research into how AI support platforms report these numbers has found that leaders frequently struggle to tell the two apart when they're looking at a vendor dashboard, and that the two figures can diverge by a wide margin depending on how a platform defines them. The practical takeaway for your own Fin AI setup is simple: a high headline percentage doesn't automatically mean customers walked away satisfied. It means the conversation didn't reach a human, which is a related claim but not the same one.

The Breakdown That Tells You More Than the Headline Number

The most useful part of the Fin AI Agent Report isn't the top-line percentage at all. It's the breakdown underneath it. Intercom lets you filter results by escalation reason, pending reason, and resolution state, and it includes a content performance table that shows, item by item, how often each piece of help center content led Fin to actually resolve the conversation versus just answer it. If your overall resolution rate looks fine while one content category is quietly driving a disproportionate share of escalations, that table is where you'll actually see it. The headline number hides exactly the kind of problem that needs fixing.

Channel matters here too. Fin's involvement and resolution figures are broken out by how the conversation came in, whether that's chat, email, SMS, WhatsApp, or phone, and a setup that performs well in chat text can behave very differently on a phone call, where the back and forth works differently. A single blended resolution rate across every channel can quietly mask a channel that needs real attention.

Reading These Numbers Without Fooling Yourself

A few habits make a real difference once you know what's underneath the headline figure. Track resolution rate alongside involvement rate so a strong percentage doesn't disguise a low volume of actual automated handling. Watch the ratio of confirmed to assumed resolutions over time, and treat a rising share of assumed resolutions as something to investigate rather than a win to celebrate on its own. Review escalation reasons on a regular cadence, since a spike in one particular reason usually points to a specific content gap or a handoff rule that needs adjusting. And look at trends across weeks rather than judging by a single day's snapshot, since one unusual spike in topic volume can skew any individual number.

None of these metrics are useless on their own. They just answer narrower questions than the single headline percentage suggests. Resolution rate tells you how Fin performed on the conversations it touched. Involvement rate tells you how much of your overall volume it's actually touching. The confirmed versus assumed split tells you how much confidence to put in that first number. Put together, they give you an honest read on whether your setup is working, instead of a number that just looks good in a screenshot.

Getting This Right in Your Own Setup

Most of the Fin AI setups I look at have perfectly reasonable numbers sitting on the surface, with a specific and fixable problem underneath, usually in one content category or one escalation reason that's gone unreviewed for months. If you want a second pair of eyes on what your own Fin AI Agent Report is actually telling you, or help tuning the setup so those numbers hold up under a closer look, reach out and I'll walk through it with you.

Cover photo by Luke Chesser on Unsplash.