What this episode is about
Almost every QA form we see has one question: did the agent show empathy, or did they give an empathetic statement. That's easy to check, which is exactly why it ended up on the form. It's also false hustle. You check the box, the agent said "I'm sorry to hear that," and everybody thinks the call went great.
We thought about this for well over a year with our own QA product, because it came up almost weekly. Where we landed is two things: intent and outcome. Have the AI go through the whole transcript and find every pain point. On hold for 20 minutes, product doesn't work, it arrived broken. Then ask whether the agent's response actually correlated with that pain, not a blanket apology.
Then measure the outcome of that moment. Did it raise the tone, raise the sentiment, calm the customer down? That's very hard for humans to score at scale because a person would have to listen to each call all the way through. AI can do that part, and you can train agents on what it finds.
People ask me if AI QA will raise their NPS or CSAT. If you keep your form the way it is, no. If you use it to go deeper, like training on empathy this way, that's where the change comes from. On the second question, whether a bot can be empathetic: today, I don't think so. It doesn't feel anything, and I think customers in the hardest situations know it.
Key takeaways
- An "empathy statement" question measures a phrase, not empathy.
- Score intent: does the agent's response match the specific pain point the customer raised?
- Score outcome: did tone or sentiment improve after that response?
- AI makes this kind of whole-call scoring practical at scale.
- Changing the form, not just automating it, is what moves customer metrics.
In Tom's words
“did the agent show empathy? Did the agent give an empathetic statement?”Tom Laird · 04:28
“Right is two things is outcomes and intent, right?”Tom Laird · 06:24
“it is very hard for human beings to score empathy at scale, right?”Tom Laird · 08:34
“AI is helping humans become more empathetic”Tom Laird · 17:27
Transcript excerpt
These excerpts come from an automatic transcript and are lightly edited for readability (filler and obvious transcription errors removed). For exact wording, listen at the timestamp.
[06:51]We say: look through the entire call, the transcript, and find all the different pain points the customer said. I've been on hold for 20 minutes. Your product just doesn't work. This was broken when it came to me. And instead of just saying that blanket, I'm really sorry to hear that, have a statement that actually correlates. Have the intent.
[07:22]Then let's look to see if the outcome of that statement did something we can measure. Because we're trying to measure things with AI. Did it raise the tone? Did it raise the sentiment? Did it lower the tone? Depending on how it goes.
[08:13]It is very hard for human beings to score empathy at scale. You have to listen to the call, you have to keep listening, all the way through, to find out if there's true empathy. That's why we came up with, did they say the empathetic statement? Because it was easy to check. Now with AI we can really train and educate agents on intent and on the outcome of the issue the customer had.