Leveraging customer feedback and data to iterate on your product

How to iterate on product has always been one of the top concerns at startups. But while they previously had to rely solely on customer feedback and instincts, they now also have a trove of data to balance. This dynamic creates new opportunities but also requires a new kind of arbitrage.

We discussed this new context during TechCrunch Disrupt 2021 in a panel conversation with Jean-Denis Grèze of Plaid, Stephanie Mencarelli of InVision and Pete Thompson of eBay. We explored different perspectives on being data driven, segmenting your user base, iteration speed and more.

The role of data

Our conversation started with a provocative question: Can a startup or a tech company ever be too data driven?

According to Thompson, the answer is “absolutely not” — but Mencarelli wasn’t so affirmative.


It’s a matter of how you use the data and how you balance it with other forms of feedback that you get. But I would say that it’s just getting more and more important to uncover things within the organization with data that you can’t do with any other means. It identifies things that human curation or manual processing wouldn’t uncover.


I will maybe slightly disagree and say that there is a point where you can get too data driven, where you’re not seeing what I call the edges of the innovation bell curve. So it’s really important to see what smaller cohorts are doing, because they could be early adopters to a new behavior.

However, Thompson himself had added some nuance earlier, noting that “You need to be able to also think about other things that you’re not doing or features that you could build that the data won’t actually tell you.” He gave an example of this at eBay:

We’ve recently been launching something that we call searching through images, so not just text queries. And this is a great example where our data would never have told us that on our site, for instance, for fashion, the younger audiences just want to be able to browse and see other things that look like the same image, not based on a text-based search. And these are things that you have to glean through other forms of feedback loops.

The importance of segmenting well