Almost everyone pointed these models at content production, which is where they're weakest — they can't know anything your customers said, so they default to the consensus. Point them instead at the material your company already generates and never reads, and they become genuinely transformative. This is the highest-return use of AI in marketing that I've found, and it's the one nobody demos.
The archive you're already sitting on
Sales call recordings. Support tickets. Churn interviews. Onboarding sessions. Sales emails. Between them, most B2B SaaS companies generate an enormous corpus describing exactly what buyers worry about, in buyers' own words. Almost none of it is read systematically, because reading two hundred call transcripts is a job no one has time for. That constraint has now materially changed.
The questions worth asking of it
Not 'summarise these calls' — that produces bland output. Ask specific, comparative questions. What objection appears in deals we lost that doesn't appear in deals we won? What words do customers use for the problem, as opposed to the words we use? Which competitor gets mentioned unprompted and in what context? What do people say immediately before they go quiet? These produce answers you can act on within a week.
- Objections that appear in lost deals and not in won ones
- The customer's vocabulary for the problem — usually not your category name
- Unprompted competitor mentions and the context around them
- The specific moment in onboarding where confusion recurs
Verify before you believe it
The output is a hypothesis, not a finding. These models will confidently assert a pattern that appears in three transcripts out of two hundred, and it reads identically to a pattern that appears in a hundred and eighty. Always ask for the specific quotes supporting a claim, then read those transcripts yourself. That verification step is what separates genuine insight from a plausible-sounding summary — and it's still an order of magnitude faster than reading everything.
What changes downstream
In the engagements where I've run this properly, the output has redirected messaging, reordered the objection-handling on the website, and occasionally changed the ICP. It's the closest thing to a shortcut to voice-of-customer research I've found — not a replacement for talking to customers directly, but a way of knowing which ten customers are worth an hour of your time and what to ask them.
Everyone's using these tools to produce more words. The advantage is available to whoever uses them to finally read the ones their customers already wrote.
Want to apply this to your business?
Book a free 30-minute strategy call and we'll diagnose your specific growth blockers together.
Book a Free Call