Numbers say what. People say why.

What I noticed at a sentiment analysis conference in Tokyo.

Last month I went to Tokyo for a few days of conferences. One of them was about sentiment analysis for market research. I wasn't sure what to expect: I work in performance marketing, not academic research. But the conversation stayed with me more than I thought it would.

A lot of what I do in marketing is still built on assumptions. There's so much data available: CTR, ROAS, CPA, conversion rates, session duration. I can see every step of the funnel. The conference made me think harder about what those numbers actually tell me.

They tell me what people do. They almost never tell me why.

The gap between data and understanding

When a campaign underperforms, my first move is to check the dashboard. Did the CTR drop? Is the CPA too high? Did the audience change? That works for a while, but at some point I hit a wall. The numbers go up and down for reasons I can't really explain from the data alone.

That's usually the moment I start guessing: A/B testing ten versions of an ad, changing the targeting, rewriting the landing page. The conference reminded me that there's another layer of data sitting right there - what customers actually say - and I don't always use it.

Where to find the "why"

In my experience, the useful "why" tends to live in a few specific places:

• The Google Ads search terms report: the exact words people typed before they clicked.
• Sales call recordings: objections, questions, hesitations.
• Support tickets and chat transcripts: what people complained about after they bought.
• Short customer interviews - 20 to 30 minutes with recent buyers (and recent non-buyers).
• Reviews, comments under ads, questions on Reddit or X about the product category.

None of this is as neat as a dashboard. It's the kind of input that explains what the dashboard shows.

In the accounts I've worked on, I've seen patterns like this: a page that converts poorly because it ignores the exact objection buyers raise on sales calls; an ad that underperforms because it answers a question nobody is really asking; a campaign that scales better once the headline uses the words customers used in their reviews.

What AI helps with, and what it doesn't

Sentiment analysis tools, and more recently language models, are good at grouping comments by topic and emotion. They can sort through thousands of reviews and surface patterns - for example, that a large share of negative reviews mention delivery times, even if no one used the same wording. (That's illustrative, not a specific client case.)

A year ago I would have read everything manually, so that's a real shortcut.

What AI doesn't do is decide what matters. A tool can tell me a theme shows up often; it can't tell me whether that's the most strategic thing to fix, or whether operations can actually solve it. That stays my job. The balance shifts: less time reading everything, more time deciding what to do with what I've read.

How this applies to performance marketing

Concretely, this changes how I approach a few things:

• When I write a new ad, I start from the customer's actual words rather than from a tagline I made up.
• When I redesign a landing page, I look at the objections buyers raised before they bought, and the reasons non-buyers gave for walking away.
• When a campaign isn't converting, I compare what the ad promises with what customers say after they buy.

Search terms, objections, post-purchase feedback: for me they give a clearer picture than ROAS alone.

Takeaway

Good marketing isn't only about efficiency, it's about understanding. Numbers tell me what people are doing. Customer language tells me why. If I only look at one, I keep optimising in the dark. Using both, I see the audience more clearly, and the ads stop being guesses.