The second XLG Summit brought together CX leaders, product teams, and insights professionals from some of the world's most customer-obsessed B2C companies. Across six sessions, one theme kept surfacing: the gap between having customer data and actually acting on it is where growth is lost — and where it's found.
Here's what you missed. And if you were there, here's what's worth revisiting.
Session 1: The Silent Customer — Abdul Khaled, Director of Transformation (E.On Next)
Abdul Khaled opened with a deceptively simple question: why do customers leave without saying anything? As Director of Transformation at E.On Next, he'd spent four years embedding an experience-led growth model into one of the UK's largest energy suppliers — and the most important thing he'd learned wasn't about technology. It was about silence.
His data showed that 75% of churned customers had stopped engaging three months before they actually left. They hadn't complained. They'd just quietly withdrawn. Abdul drew on relationship psychologist John Gottman's concept of 'stonewalling' — emotional disengagement as a leading indicator of relationship breakdown — to reframe what churn signals actually look like in practice.
"Customers don't just fall off a cliff. It's a series of signals that would've happened months prior. They'd already made up their mind they were going to leave you." — Abdul Khaled, Director of Transformation, E.On Next
The practical example that landed hardest was office noise. Chattermill data showed just 0.09% of customers mentioning it — around 45 comments out of 50,000 a month. It would never appear in a top-10 issues report. But Abdul reframed the signal: those 45 people weren't the only ones affected. They were just the only ones who bothered to complain. Everyone else felt it and said nothing.
The fix was straightforward once they'd found it: the wrong headsets, left behind between shifts. But the principle was bigger — small-volume feedback, read in context, can represent a much wider silent population. The teams who act on it are the ones preventing churn before it's visible.
Abdul closed with two customer stories from the energy crisis — a single father keeping doors closed to conserve heat for his son during custody visits; a couple waiting until their children came home to use an appliance just once. A deliberate reminder that behind every data point is a person.
"We're too focused on ROI, revenue, profits and market share. We sometimes forget there's a human behind that number." — Abdul Khaled
Session 2: What's Next for Chattermill — Mikhail Dubov, CEO (Chattermill)
Mikhail's session was the product announcement of the summit — but it was framed less as a feature launch and more as a philosophy of what good AI actually looks like in a world where building software has become almost trivially easy.
The core tension he identified: when AI tools make execution so fast, the hard part is no longer building — it's deciding what to build. Teams can now ship faster than they can think clearly, which creates three new failure modes: prioritising the wrong things, converging too quickly on solutions before understanding problems, and accumulating 'Jenga tower' products — features stacked on fragile assumptions.
"Speed without direction just gets you to the wrong place sooner. If you work brilliantly on the tenth most important problem, that's still wasted effort." — Mikhail, Co-founder, Chattermill
The product announcement itself — Chattermill's agentic vision — was built directly from this insight. Rather than just helping teams analyse feedback faster, the new direction focuses on helping teams decide better. A proactive discovery mode that continuously surfaces emerging signals, a collaborative exploration layer that works like a human analyst alongside you, and an MCP server that brings Chattermill intelligence into the tools teams already use.
The framing that stuck: when Mikhail surveyed product managers about what they most want AI help with, the answer was no longer coding. It was user research — not to replace researchers, but to reduce blind spots and make decisions with more confidence.
"If you ask product managers what they want most from AI, it's no longer coding. It's user research. To understand customers better, reduce blind spots, and make decisions with confidence." — Mikhail, Co-founder, Chattermill
Session 3: Optimize for Surprise — Rory Sutherland, Vice Chairman, Ogilvy UK
If Abdul's session was the emotional anchor of the day, Rory Sutherland's keynote was the intellectual one — a fast-moving argument for why customer experience is the most underrated competitive advantage in business, and why most companies are too busy benchmarking each other to notice.
His central provocation: when everyone optimises for the same metrics, they become more alike, which destroys value for everyone — businesses, investors, and customers. The rational individual action (do what competitors do) becomes collectively stupid when everyone does it simultaneously.
The antidote is what he calls reverse benchmarking: instead of copying what competitors do well, find what they don't measure, don't care about, or culturally can't do — and make that your differentiator. His examples ranged from the restaurant that hired a beer sommelier when everyone else focused on wine, to Steve Jobs asking 'what does a computer feel like?' when everyone else was asking 'what can it do?'
"Don't copy the competition. Find something they haven't noticed, don't measure, can't do. Double down on that thing and make a lot of noise about it. That's reverse benchmarking."
— Rory Sutherland, Vice Chairman, Ogilvy UK
On customer experience specifically, he argued the standard is low enough that basic competence is surprising: AO has nearly a million Trustpilot reviews averaging 4.9, and attributes 90% of that to a single principle — they tell customers what they're going to do, and then they do it.
His closing argument: businesses consistently underinvest in retention not because it isn't valuable, but because its value is slow to appear. AI, used well, could rebalance this — giving CX the same quantitative credibility that sales has always had. But only if organisations choose to use it for value creation rather than cost-cutting.
"A personal experience at a timely moment trumps anything you can do in advertising. Customer experience can achieve something advertising rarely can — a complete brand quake." — Rory Sutherland
Session 4: From Insight to Action — Ipshita Chakraborty, Head of Data Science for Global CX (HelloFresh), Aji, Chief Scientist (Chattermill) & Sahil Rekhi, CRO (Graia)
The fourth session tackled the question that underpins everything else at XLG: what does it actually take to move from a customer insight to a business decision? The panel — Ipshita Chakraborty, Head of Data Science for Global CX at HelloFresh (dialling in from New York through a blizzard), Aji, Chief Data Scientist at Chattermill, and Sahil, CRO at Graia — brought real-world agentic deployments to a conversation that too often stays theoretical.
The moderator framed the stakes clearly: agents have landed. The technology is loaded and ready. But in Chattermill's customer advisory board sessions earlier that day, almost no one was ready to let an agent make a customer-facing decision without a human in the loop. The panel explored where that trust threshold actually sits — and why.
Ipshita's HelloFresh story was the most concrete moment of the session. Factor, HelloFresh's ready-to-eat brand, was growing so fast it couldn't keep up with demand. The response was to short-ship orders and issue full refunds — a decision that was bleeding money and quietly frustrating customers. When Ipshita's team ran the voice of customer data on care transcripts, the insight was almost embarrassingly simple.
"Customers didn't want the refund. They just wanted dinner. If you're out of broccoli, just give me green beans. Just send us the box." — Ipshita Chakraborty, Head of Data Science for Global CX, HelloFresh
The product team built a meal swaps hierarchy, launched it, and stopped the bleeding on full box refunds — saving millions, but more importantly, solving the actual problem customers had. Ipshita was clear this wasn't a one-off: the same VOC engine had since driven US packaging changes, real-time campaign refinements, and app experience improvements — all based on what customers were saying as they said it.
The discussion then turned to agentic AI specifically. Ipshita described what a genuinely useful agentic system would look like at HelloFresh: scanning thousands of delivery photos on Monday, flagging a crushed-box trend to the relevant warehouse by Tuesday, and drafting customer communications by Wednesday — all without a CX ops manager spending 20 to 30 hours pulling data into a spreadsheet.
"Instead of a CX ops manager spending 20 to 30 hours a week pulling data into a spreadsheet, the system acts like a chief of staff for the experience. It identifies the fire, grabs the hose, and only taps the human on the shoulder for the edge case." — Ipshita Chakraborty, HelloFresh
But she was equally clear about where HelloFresh is holding the line: customer-facing decisions. Anything involving PII, money movement, or direct customer trust is still routed through a human. Agentic wins, for now, are in internal, lower-stakes workflows — where productivity gain is the goal and the cost of a mistake is recoverable.
Aji rounded out the session with a sharp corrective on the hype: agents and LLMs perform well on problems with concrete, objective answers. Customer experience is by nature subjective, context-dependent, and human. The same analysis can be right one day and wrong the next, depending on what's changed in the business. Zero-human-in-the-loop systems aren't a near-term reality in this domain — and claiming otherwise is one of the bigger myths currently circulating.
"The hype is around 'you don't need any humans in the loop.' In a domain as subjective as customer experience, where the same answer can be right one day and wrong the next depending on context — that's one of the biggest issues I see." — Aji, Chief Data Scientist, Chattermill
Session 5: All Systems, No Soul — Panel moderated by Alison Blair, Managing Director, Customer Led Transformation (PwC), Obi Santos Senior CX & Insights Leader (Vivobarefoot), Agne Schveyher, CX Manager (MPB)
The PwC-moderated panel tackled the sharpest tension of the day: as automation accelerates, emotional intelligence is quietly disappearing from customer interactions — and most organisations don't notice until their CSAT scores are already slipping.
PwC opened with research from over 3,000 UK consumers showing that the problem isn't automation itself. It's that automated systems handle every situation the same way, while customers arrive in wildly different emotional states — from routine to crisis. The mismatch is felt more than it's measured.
"In this pursuit of efficiency, emotional intelligence is quietly disappearing. Customers don't necessarily have a bad experience when empathy is missing — they're just left cold. There's no depth. They don't feel heard." — Alison Blair PwC
The panel — featuring representatives from Vivobarefoot and MPB — offered practical frameworks rather than abstract principles. Vivo Barefoot's approach: before automating anything, ask 'should we?' not just 'can we?'. Product inquiries are left entirely to humans because customers want to speak to someone who's actually worn the shoes. MPB identified the high-value, emotionally loaded moments and built a dedicated seller experience team to jump in precisely when hesitation appears in the data.
The key insight: Chattermill sentiment data plays a critical role not in replacing human judgment, but in identifying where to deploy it. When NPS is high but sentiment is low on a specific theme, that's the signal to send a human — not another automated touchpoint.
"Rather than thinking 'what can we automate' — we could probably automate everything — I like to think: what should we automate. That's how you balance emotional intelligence with technology." — Obi Santos, Vivobarefoot
Session 6: Clarity at Scale — Kevin Griffin, Head of UX and Design (Zepz) & Vincent Dollet, VP of Product (Chattermill)
The final session was the most practical of the day — a close look at how Zepz, the global payments platform, operationalises customer insight at scale across a fast-moving, competitive industry where pricing sensitivity is high and silent churn is a constant risk.
Kevin walked through Zepz's feedback architecture: weekly triage reviews using Chattermill as an early warning system, quarterly deep-dives to surface structural issues, and customer sentiment sitting alongside conversion and activation metrics in leadership reporting. Net sentiment isn't a 'CX metric' at Zepz — it lives in the same dashboard as revenue and retention.
"We always couple data with a verbatim that represents the issue. We want stakeholders to hear it in the customer's own words. We don't want it to just be numbers." — Kevin Griffin, Zepz
The session also explored how Zepz handles the tension between planned roadmap work and unplanned signals — like a spike in login failures traced back to a single buggy app release, caught and rolled back within a week because the Chattermill dashboard showed all complaints clustering on the same version.
Vincent then previewed how Chattermill's upcoming 'Views' feature would extend this kind of workflow — and the conversation ended on the MCP integration: the prospect of a UX auditor that can pull in live customer sentiment directly, so a screenshot of a new design triggers a check against what customers are actually struggling with today.
"We'd throw in a screenshot and it would say: 'this could solve a problem you're already seeing in your customer insights.' That's customer data we don't have to worry about feeding in — it's already there." — Kevin Griffin, Zepz
The Takeaway
Six sessions, and the same thread running through all of them: the organisations doing this well aren't the ones with the most data. They're the ones who've built the discipline to look at what others ignore, act on signals that don't appear in top-10 reports, and keep the human in the loop at exactly the right moments.
The tools to do this are better than they've ever been. The question — as Rory put it — is whether you use them to cut costs or to create things worth talking about.
"Experience-led growth is actually the best kind. And it's often surprisingly cheap — if you're prepared to do it differently." — Rory Sutherland
