AI Just Beat Experts at Understanding Customer Needs

A fine-tuned LLM just identified 100% of customer needs from product reviews and interviews. Professional analysts, using the same data, caught 87.5%.

If you know anything about Voice-of-Customer studies, you know this shouldn't be possible.

Understanding customer needs isn't about parsing text—it's about recognizing unstated desires. When someone complains their wood stain "goes on pink," they're not asking for a different color. They're revealing a need to see which surfaces they've already covered. Analysts train for years to make these leaps.

And base AI models, indeed, failed spectacularly—identifying real customer needs only 40% of the time, worse than humans. The breakthrough came from supervised fine-tuning with about 1,000 professionally-extracted customer needs from past Voice of Customer studies.

Turns out, with the right training data, AI does it better.

MIT researchers partnered with Applied Marketing Science, a firm with 30 years of VOC experience. They trained the model on real studies spanning ten product categories. Once trained, it processed entire datasets without fatigue—all 14,341 customer reviews in the wood stain study, not just the curated subset analysts typically review due to time constraints.

And it didn't hallucinate. When evaluators checked whether extracted needs followed from source material, the fine-tuned LLM scored 92% versus 84% for human analysts.

The practical test: a professional product development association needed to understand member experience. Standard consulting project, $50K minimum, months of work. They uploaded interview transcripts to the fine-tuned model instead. Got results in minutes. The insight: new members found insider jargon off-putting, so they redesigned onboarding to be more accessible.

It is not just about quality. Customer insight has always been episodic:you commission a study, get a report, make decisions, then operate blind for months. By the time you realize a need has shifted, you're behind.

One caveat: this only works if you do it right. You need quality training data and proper fine-tuning. But for anyone sitting on past VOC studies, interview transcripts, or review data, you already have what you need. The methodology generalizes across categories.

If you've been treating customer needs analysis as something only specialists can do, that assumption just got invalidated.

— Torsten & Peter

Our sponsor today:

Startups who read beehiiv can receive a 90% discount on Intercom's AI-first customer service platform, plus Fin—the #1 AI agent for customer service—free for a full year.

That's like having a full-time human support agent at no cost.

What’s included?

Who’s eligible?

Intercom’s program is for high-growth, high-potential companies that are:

Apply now

We wouldn’t be marketers if there weren’t a Black Friday deal for you.

From today until EOD November 28 (Black Friday), we have a once-in-a-year offer.

The 50% off deal is available only to current subscribers and is live from today.

Go to this checkout page. The discount is already applied to the pricing.

Scroll down to the bottom of our emails (any).

If you are a free subscriber, you’ll see a paywall.

If you are a paid subscriber, you’ll see your access links.

Our newsletter partner today:

AI Dash delivers weekly AI insights for solopreneurs and creators who need results, not hype. Get practical tool reviews, strategic frameworks, and actionable workflows that actually move the needle. Skip the AI hype. Get the intel that counts.

» Subscribe to AI Dash