The assumed-resolution problem: how silence becomes revenue

Fin, formerly Intercom, publishes the simplest price in AI support: $0.99 per resolution, charged on outcomes instead of seats. It sounds like the fairest model in the category, which is exactly why the definition of a resolution deserves a close read. In Fin's accounting, a customer who goes silent for 24 hours counts as resolved. Then it bills.

How silence becomes a line item

The mechanic is called an assumed resolution, and to Fin's credit it is documented, not hidden. If a customer stops replying for 24 hours, the conversation is marked resolved and billed at $0.99. No confirmation from the customer, no check of the transcript. Silence is the whole signal.

How much of a bill this represents is hard to see from the outside. Estimates place 25 to 45 percent of billed resolutions in the assumed bucket: conversations no one confirmed were solved.

The reported bill-shock stories follow the same shape. Users report a monthly bill moving from roughly $4,000 to $9,000, and another from $119 to $854 after plan changes. We can't verify those figures, which is why we hedge them. The mechanic that makes them possible, though, is printed on the price tag.

Contract structure compounds it. Fin sells 12-month auto-renewing contracts with 30-day renewal-notice windows. Discover the assumed share of your bill in month three and you may be paying it until month twelve.

Silence is the most ambiguous signal in support

A customer goes silent for many reasons, and only one of them is satisfaction.

• They got their answer and left. This is the case the billing model assumes.
• They gave up and emailed instead.
• They gave up and bought from a competitor.
• They asked at 5 p.m. on a Friday and never came back.
• They needed a human, couldn't reach one, and stopped trying.

The gap between silence and satisfaction shows up in the numbers. Reported real-world resolution rates for text bots cluster between 42 and 53 percent, against vendor claims of roughly 76 percent. Those are estimates and we treat them as estimates. But if they are even directionally right, a meaningful share of what gets billed as resolved is actually unknown, and some of it is failure.

One more data point, reported and worth stating carefully: Fin's own customer support is rated 3.2 out of 5 on Trustpilot, and the most-cited complaint is not being able to reach a human. In an assumed-resolution model, a customer who stops trying looks identical to a customer who was helped.

The vendor profits from the ambiguity

We want to be fair here. Fin has real strengths: distribution, a deep catalog of integrations, and a brand support teams already trust. Per-resolution pricing was a genuine attempt at outcome-based pricing, and outcome pricing needs an automatable definition of an outcome. Twenty-four hours of silence is the cheapest definition available.

But follow the incentive. Every ambiguous conversation is worth $0.99 if it is called resolved and nothing if it is not. Disambiguating — asking the customer, auditing the transcript, tightening the definition — costs the vendor money on every conversation it reclassifies. Nobody has to act in bad faith; the structure settles every uncertain outcome in the seller's favor, automatically, at scale.

When ambiguity is billable, ambiguity stops being a measurement problem and becomes a revenue line.

The pressure is unlikely to ease. In June 2026, Salesforce agreed to acquire Fin for roughly $3.6 billion, and businesses bought at that price are expected to grow. In this model, revenue is a function of how resolutions are counted.

What honest accounting looks like

We think the fix is unglamorous: measure outcomes from evidence, and admit what you can't measure. Bubblio reads each conversation's transcript and classifies the outcome from what was actually said. When the transcript supports a call, the conversation is marked resolved or unresolved and the receipt is attached to it. When it doesn't, the conversation is counted as unmeasured — not resolved by default, not quietly dropped.

The dashboard reduces it to one line: X% resolved · measured on N. The denominator is always on screen, because a rate measured on twelve conversations should read differently from one measured on twelve hundred. Every conversation carries its own receipt, so any number we show can be audited down to the transcript that produced it.

The same rule runs through the rest of the product. The agent answers from your own site's content with citations, and when it doesn't know, it refuses to guess; every miss lands in a 'Customers are asking' feed so you can see exactly what was missing. An agent that won't invent answers and a dashboard that won't invent outcomes are one design decision applied twice.

Why we publish the smaller number

Honest accounting carries an obvious commercial cost. A resolution rate measured from transcripts, with unmeasured conversations counted as unmeasured rather than as wins, will usually read lower than a rate that books silence as success. Next to a reported vendor claim of roughly 76 percent, our dashboard can look like the weaker product.

We accept that for two reasons. First, you make real decisions on this number — staffing, escalation rules, what to write next for the knowledge base — and a flattering number corrupts every decision downstream of it. Second, we removed our own incentive to inflate it.

Bubblio is priced as prepaid minute packs: $49 for 100 minutes, $199 for 500, $599 for 2,000. There is no per-resolution fee, no overage, and no contract, and when minutes run out the bubble degrades to text chat rather than going dark. We do not earn a cent more when a conversation is labeled resolved, so the label has no job except to be true.

None of this means Fin is wrong for everyone. If your volume justifies its integration breadth, if you audit your billing exports monthly, and if an annual contract fits how you buy, the per-resolution model can work for you. The narrow point is this: whatever tool you use, know how it counts silence, because you are paying by the count.

If you want to see what assumed resolutions do to a real bill, our resolution-fee calculator models it against minute-based pricing, and What Fin really costs walks through the full structure. If you would rather look at receipts than claims, Bubblio is in private beta and you can request access.