Your CSAT Only Scores 8% of Support Conversations

The short answer: your CSAT number is built on a sample so small it barely qualifies as a measurement. Across all Intercom customers who use CSAT, just 8% of conversations get scored, by Intercom's own count. AI can score all of them. The only real question left is whether 100% visibility is table stakes or a $99 add-on.

We think the answer to that one is obvious. But let's start with the 8%.

Jump to a section:

• How many conversations does CSAT actually measure?
• What is the other 92% costing you?
• AI can score every conversation, so why is it an add-on?
• What does included scoring look like?
• Measure everything, pay for outcomes

---

How many conversations does CSAT actually measure?

About 8 in 100. That number is not ours. It comes straight from Intercom's AI Agent Blueprint: "Across all Intercom customers who use CSAT, just 8% of conversations get scored." Intercom publishes the benchmark AI agent and the benchmark admission in the same document.

And the 8% you do get is not a random sample. The customers who fill in surveys are disproportionately the delighted and the furious. Everyone in the middle, which is most of your customers, closes the chat and moves on. The blueprint's own economics chapter quotes customer service expert Shep Hyken on why: "People have survey fatigue, and they're not answering because they don't think their feedback is going to be appreciated."

So the metric on your quarterly dashboard is a tiny sample with a built-in skew toward extremes. It measures who responds. It only loosely measures how your support performed.

What is the other 92% costing you?

Run the numbers on a team handling 5,000 conversations a month. An 8% response rate means 400 conversations produce a score. The other 4,600 produce silence. Over a year, that is more than 55,000 customer interactions with no quality signal attached to any of them.

Now think about what gets decided on top of that sample. Staffing levels. Knowledge gaps. Which macros to rewrite. Whether the AI agent you just deployed is resolving conversations well or just closing them. A "resolved" conversation can still leave a customer annoyed enough to churn, and CSAT at 8% coverage will never tell you. That blind spot matters more than ever now that support is a strategic function, not a cost center.

Here's the uncomfortable version of the question: would you accept any other business metric measured this way? Revenue reported from 8% of invoices? Uptime sampled from 8% of servers? Support quality is the only number teams still accept on these terms, mostly because surveys used to be the only instrument available.

They aren't anymore.

AI can score every conversation, so why is it an add-on?

The same blueprint makes the case for the fix: AI can analyze 100% of customer conversations in real time. Intercom built exactly that. Fin evaluates each conversation for resolution status, customer sentiment, and service quality, then produces a CX rating from 1 to 5 with reasons attached. As a measurement instrument, it's genuinely the right idea.

Then comes the invoice. CX Score ships inside Intercom's Pro add-on at $99 per month, covering analysis of 1,000 conversations. Our team at 5,000 monthly conversations is back to sampling, this time at 20%, unless they pay for more capacity. The pricing page doesn't say what more capacity costs.

Read those two facts together. Intercom's published position is that scoring 8% of conversations is inadequate and AI should evaluate all of them. Intercom's published pricing puts that evaluation behind a paywall with a cap. The diagnosis is free. The cure is $99 a month, sold by the milliliter.

What does included scoring look like?

We built SupportWire on the other reading of that diagnosis: if an AI handles customer conversations autonomously, knowing whether it did a good job is part of the product, not a product.

Every SupportWire plan scores every AI conversation. When Kal resolves a ticket, the resolution gets a quality score, the reasoning behind the score, and a flag if something looks off. Trend detection and automated alerts are included. There is no analysis cap, no sampling, and no add-on to unlock any of it, on the free plan or the paid ones.

That isn't generosity. It's incentives. We charge $0.49 per resolution, and a per-resolution business only works long term if the resolutions are actually good. Hiding the quality data from you would be hiding it from ourselves.

Measure everything, pay for outcomes

CSAT was a reasonable instrument for an era when measurement was expensive. That era ended. The vendor that published the 8% number agrees, which is why the disagreement left is narrow and easy to evaluate: should 100% quality visibility cost extra, or come standard?

If you're pricing out the whole stack, the real math on AI support cost covers the per-resolution side, and the Intercom vs SupportWire page covers the rest. The short version: half the per-resolution price, and the measurement layer Intercom sells as an add-on comes included.

Your support quality is already being graded by your customers, 100% of the time. The only choice is how much of that grade you get to see.