What this episode is about
I'm a huge analytics guy. Speech analytics reshaped how we run our own contact center. But after speaking at an event on AI roadmaps, the questions I kept getting were about analytics, not scoring. Companies want real, granular data they can act on, and keyword bubbles and trending charts often don't get them there.
What changed my thinking is that our beta customers were using an AI QA platform as a compliance tool, or a retention tool, not just for scoring agents. You ask a question of the data and the AI answers it. Not just did the agent say these words, but was the intent there. A customer says they lost their job. You don't need the keyword "lost my job." The model understands that's why they're canceling.
With keywords, you're always going to miss things. You have to think through intent and outcome: what does the end result look like, and work backward. One example from a proof of concept was a rule about when a third party is allowed to take over a call. That's the kind of thing you can now prompt for.
Traditional analytics still has a place. It's better at telling you what you don't know, like a strange spike in calls after an email went out. And at the time of this recording, we were scoring transcripts from voice, email, chat and tickets, not what agents did on screen.
Key takeaways
- AI QA can answer direct business questions from calls: cancel reasons, buying reasons, compliance.
- Intent-based questions catch what keyword lists miss.
- Design the question from the outcome you want to know, then work backward.
- Traditional analytics remains better at surfacing unknown trends.
- Start small: upload a single transcript and start asking it questions.
In Tom's words
“we're asking a question of the data”Tom Laird · 07:34
“keywords you're going to lose out”Tom Laird · 13:28
“You have to look at intent. You have to look at outcomes.”Tom Laird · 13:28
“I kind of call it seeing the unseen.”Tom Laird · 12:41
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're now literally just asking that question and having the AI think and hear what's being said. So we're asking a question of the data, and the auto QA platform is answering questions on the data now. That's what we said analytics was going to do. But I think analytics is trending, it's macro. If you really want to dig deep into a topic, I think this blows away regular keyword-based analytics: looking for intent, looking at outcomes, asking questions.
[09:43]We're not looking for just did they say these specific words, but was the intent to do this. Did somebody say, hey, I've lost my job? We don't have to look for the keyword lost my job. The AI understands that this is the reason they have to cancel. So it's almost like talking to a human.
[11:36]We want to know why our customers are buying. What do they like, what do they not like, why are they canceling. All of these questions we can put in and literally score yes or no: did these happen? Then take all that data and make it an extremely useful tool. I kind of call it seeing the unseen.