How To Do Cool Stuff With AI: Analyze Your Call Recordings
There's gold buried in your customer calls, and AI is here to help you find it
Over the last few months, I've been talking to a hand-picked group of friends, investors, and operators about a topic I'm super curious about:
What are people out there actually doing with AI?
The AI hype is inescapable right now. Everyone is talking about it. And the data would tell you that most people out there are at least experimenting with AI tools like ChatGPT.
But in my world of B2B software, I've noticed an interesting kind of dissonance. Despite the hype, and despite the eye-popping surface-level adoption data, many people out there don't seem to have crossed the chasm between playing with AI and working with AI.
Many of us are still searching for how AI can reliably help us do what we get paid to do - faster, better, easier, and more rigorously.
That's what this series is about: The cool stuff you can do with AI that will actually make you and your business better.
This isn't theory. I'm building this series of pragmatic how-to guides by reverse-engineering the best of what other people like you are already using AI to accomplish. And over the next few months, I'll be unpacking the best "real world" AI use cases I've discovered, breaking them down, and giving you the tools you need to get started using them for yourself right away.
In case you can't tell, I want this to be as plug-and-play as possible. So along with the use cases, I'll also be sharing short prompt engineering guides to help you steal from what's already working for other AI newcomers like you.
How AI Can Analyze Your Call Recordings
If you work in sales, marketing, or product management, and you're not using AI (I'll use ChatGPT in this example) to analyze your call recordings, this is where you should start.
Here's how to engineer a ChatGPT prompt you can use to pull the best stuff out of your customer calls.
[1] First, you're going to transcribe your call recording. Most call recording software out there will spit out a raw txt file if you ask it to. No need to format it - you can paste it right into ChatGPT. Then it's all about asking the right questions.
[2] Tell the AI to look for, and what to ignore. In this case, I'm less concerned with "how well the salesperson is doing." I'm looking for the stuff that turns you into a customer mind-reader - the pain-points they're dealing with, the workarounds they've constructed, and generally all the ways they deal with that pain that YOU can help them solve. If you know that stuff, you can make your sales pitch, your marketing campaigns, your product decisions and every first impression with a potential prospect so much better.
[3] Ask ChatGPT to keep track of customer questions. These are priceless, and very hard to pull out of a long, unstructured customer conversation. So try this: Ask the AI to simply list ALL the questions the prospect asks during the call. Hint: If you do this for 5 calls, you'll have a couple quarters' worth of ideas for blog posts, whitepapers, social posts, etc. There's no better recipe for great content than (i) clear answers to (ii) frequently-asked-questions that customers are actually curious about.
[4] Tell the AI "what you're going for." Giving it context (e.g., "I'm going to use this output to do X") will almost always give you a better, more detailed answer that saves you some steps on the back end.
[5] Be specific. Tell the AI what NOT to leave out. In this case, I want the stuff that's going to help me speak the customers' local dialect - emotive words, concrete examples of competitive alternatives, and day-in-the-life vignettes. I want the stuff that's going to help me guess correctly at what these people do all day - and more deeply understand what a bad day at the office looks like.
[6] Separate your prompts and give the AI permission to clarify. I almost always include an additional clause that sounds something like "Please ask me clarifying questions if it will improve the analysis." Half the time, ChatGPT will ask me a few foundational questions that, if they had gone unanswered, might have led to a more generic response. The additional clarity on the back end is worth the extra step. (And it’s kind of fun to make this a dialogue, instead of just a set of instructions.)
[7] For more customer insights, just repeat. After you do this for one call recording, ask ChatGPT to put the summary into a word doc. Then download it, review it, and stash it somewhere. Then go do the same for 3-5 more call recordings. Then combine the documents and feed it back into ChatGPT and ask it to analyze the data in aggregate. Some examples of what you can ask it to do next:
💡 Describe what's not so great about those competitive alternatives
💡 Suggest the top 10 content marketing ideas to drive leads
💡 Create a list of the most common customer questions
💡Build you an email prospecting cadence for your ICP
💡 Write a landing page for a specific customer type
💡 Make a list of "what people use instead of you"
💡 Summarize the most common pain-points
Final Thoughts
If you're still getting a sense for what AI can do for you, and if you're wondering how to get more value out of your call recording software, don't stop at the "out-of-the-box" call recording summaries and "did I ask enough questions" talk-time splits.
Go deeper. Try using ChatGPT to unearth the stuff that can turn you into a customer mind-reader - what they're struggling with, what they've tried, and what's going on inside their heads while they wrestle with what to do next.
There's gold buried in your call recordings.
And with a little help from your new friend AI, it's surprisingly easy to find it.
This is fantastic guidance; thanks for sharing how you are mining customer calls for gold. We are in the early stages to identify emerging patterns and this is extraordinarily helpful.
Thanks Paul. I recently built a research assistant app to do just this. Unfortunately, results have been weak. My expectation is after analyzing 40 discovery calls, it would be able to unearth all of the hidden gems and help me design a hit product. What I learned, so far, is that AI is good at summarizing and giving middle of the road answers, not insight. So far, everything answer it has given me is stuff that I knew already without taking any notes. I guess one positive is that it validated my ideas, like having someone else in room who can say, "Yeah, that is what happened."