AI finds the signal. Acting on it still requires someone who’s been in the room.

At Glossier, Cati Brunell-Brutman’s VOC team ran a five-year review analysis on the brand’s top 10 SKUs. The Ultralip lipstick had been flagging in CX emails for years, but ticket volume alone wasn’t enough to justify a full packaging redesign. The AI analysis changed that: 60 to 70% of reviewers said they loved the product but wouldn’t repurchase because of the cap. That number, tied to the SKU’s rank and customer acquisition cost, got the redesign approved. When new shades launched with the updated cap, Glossier posted that they’d heard the feedback. Customers on Reddit confirmed it.

The gap between surfacing the data and knowing what to do with it

Cati’s point isn’t that surfacing VOC has gotten harder. Tools like Hark aggregate channels and use AI to push trends before you have to go looking. She’s run that kind of analysis herself, loading thousands of customer comments into a spreadsheet and using AI to pull what customers love, what they don’t, and where patterns are forming. The surface-level problem is largely solved.

What isn’t solved is what sits between the data and the decision. Knowing what to ask for, and how to turn “customers keep saying this” into a recommendation that actually gets resourced. Cati names that gap directly: AI can surface the data, but it can’t apply the lived experience of a decade in customer support to figure out what to do with it.

She’s also clear about what that means for the role. CX leadership isn’t going to be automated out because the judgment isn’t in the model. It’s in the person who’s read every NPS comment on a Monday morning and knows how to bring the right story to the right room.

AI can surface your data, but it can’t figure out what to do with it.

What Cati’s team actually built at Glossier and Daily Harvest

Cati’s most recent structure at Glossier was what she called a team of teams: a core group of internal specialists who’d been with the company five to seven years and were deep knowledge and product experts, a remote BPO, and a rotation program where retail associates could spend six months working with the digital CX team. The retail rotation gave her team first-hand product context on launch days. She worked to make sure all three groups had the same access to training and shared knowledge, including virtual sessions where everyone could ask questions together.

The VOC infrastructure she built there started with a channel audit. She sat down with a retail director and listed every place a customer talked to Glossier, from Google reviews to owned social to email, then looked at month-over-month volume and channel coverage to identify where the largest share of customers were and where they were loudest. That audit became the foundation for deciding where to focus the analysis.

From there, she ran monthly VOC meetings open to any cross-functional partner who wanted in. Product, marketing, and ops could attend, ask for specific data pulls, and hear what customers were saying. The Ultralip cap story came directly out of one of those meetings: the product team asked for a deep dive on the top 10 SKUs, and the five-year review analysis made a small but persistent CX signal impossible to ignore. Between 60 and 70% of reviewers said they loved the formula but wouldn’t repurchase because of the cap. The product team had the data they needed to justify a redesign.

She built a similar rhythm with the social team. CX and social established a weekly meeting to preview upcoming posts in advance. The goal was straightforward: CX could flag product availability issues before they showed up in the comments. She flagged that a product featured in an upcoming post was out of stock in the UK, letting the social team swap it before the post ran. Both teams avoided a predictable headache.

At Daily Harvest, where she joined as an early employee, the trust between CX and product worked the other direction too. The chefs would bring new smoothie versions to CX specialists first for taste testing, asking what they thought customers would love and what would confuse them. When the brand was preparing to launch a vegan cheese bowl spelled C-H-E-E-Z-E, Cati’s team flagged that customers would think it was a typo. That feedback became part of how the product was positioned at launch. It’s still on the site.

For CX leaders who want to build a VOC program with no budget and no executive buy-in, she’s specific about where to start: identify a low-lift, high-impact problem that someone else already wants solved, and show what CX data can do for that problem. Do a listening tour. Ask cross-functional partners what they’d fix if time and money weren’t a constraint, and find the problem where CX data could move the needle. She also keeps the tooling simple. She has a personal Claude subscription she uses for analyzing large exports of customer data. Google Sheets and one AI tool are enough to build the proof of concept. Once you’ve shown that the data can solve something people already care about, the conversation about a dedicated VOC platform gets easier.

The question Cati leaves open is whether CX leaders building their experience now, with AI handling the data layer, will develop the judgment the role requires. The surfacing problem is solved. The critical thinking that turns a spreadsheet of customer comments into a redesigned cap, or a vegan cheese bowl that doesn’t look like a typo, still requires someone who’s been in the room.

If you’ve been working through how to keep that judgment layer intact as AI takes on more of the analytical work, that’s the conversation this episode is after.

• Cati Brunell-Brutman’s LinkedIn
• Cati Brunell-Brutman on Instagram
• Hark (VOC platform)

• Youtube
• Spotify