Real Interviews First, AI Second

Why the order matters when you mix real customer conversations with AI research.

---

AI research tools promise to replace the customer interview. They can’t do the job properly, and it’s worth being specific about why.

The job isn’t collecting answers. It’s noticing what’s underneath them. The pause before someone answers a pricing question. The moment a buyer stops talking about the product and starts talking about their boss. AI can’t sit in that room. It can only work with what’s already been said somewhere, by someone, and written down.

That’s the real limit. AI research is built from existing data. It can’t tap into what a customer is still working out how to say. The insight that changes your strategy is usually the one nobody’s articulated yet, including the customer.

Where AI earns its place.

Run five pricing models past ten customer segments and AI will give you a view in an hour, not a quarter. It’s properly good at that. Comparing options. Covering more scenarios than a fixed research budget ever allows.

None of that replaces the conversation. It scales what the conversation taught you.

My own process starts with a cross functional workshop. We agree who to talk to and why, which isn’t always a given even inside a single leadership team. Then come the interviews themselves, an hour of focused conversation with the actual customer or prospect, sharing ideas as I hear them, listening to the pauses as much as the answers.

Only after that do I bring AI in, to organise what I heard, test the pattern against more scenarios, and challenge my own thinking. The AI never runs the interview. It works on what the interview already found.

Before partnering with Substribe, our understanding of our value proposition and audience priorities was largely anecdotal. Substribe not only validated many of our assumptions but also sharpened our understanding of value and revealed clear growth opportunities. Their highly structured, in depth interview process enabled Substribe to build key audience profiles, suggest evidence based content ideas, and map out price sensitivities. We’re also using their AI tools to test our new strategy pre-market launch.

That’s the order that works. Real conversation first, for the human read no dataset contains. AI second, to stretch that read across more ground than time or budget would otherwise allow.

Get the order backwards, and the research sounds confident and misses the point.

---

Seeing this pattern in your business?

If this is showing up in your subscription business, it’s worth a conversation. 30 minutes, no pitch — just diagnosis.

Related reading: You Got Claude, But Are You All Right?

Written by the Substribe Claude project, with ideas from Andy Burden.