Mikhail Dubov has a small worry about the AI agent his company has put in front of customers: someone might ask it about the weather. Chattermill built the tool for difficult customer research, the sort of work that ordinarily occupies an analyst for days. Yet the first question a real user types may have very little to do with the months of work behind it. A launch is an invitation to be corrected.
He has described another complication. By the time a product ships, he is already thinking about its second or third version. Celebrating the present requires catching up with his own attention. It is a recognisable founder’s predicament, with a particular irony here: Dubov earns his living helping businesses pay attention to the people using what they build.
As co-founder and CEO of Chattermill, he has spent a decade working on what happens between a customer saying something and a company doing something about it. The gap can accommodate entire departments. It can also swallow a perfectly useful complaint.
A career with several former lives
Dubov’s professional introduction has included banking, consulting, data analysis and software engineering. He studied economics at Cambridge from 2005 to 2008, followed by postgraduate study at the London School of Economics and Political Science. He later worked as a software engineer at AlphaSights. Finance and code both appear in the route to Chattermill.
There is a temptation to tidy such a career into a straight line. The more interesting detail is the change in how he understood a subject he had already chosen. Dubov has credited Freakonomics with helping him fall in love with economics and look at the world differently. A discipline associated with numbers became a way of asking questions about behaviour.
That makes a useful entry into his present work. Customer intelligence asks a business to examine the distance between what it believes and what people experience. A score supplies a number. A person’s explanation can supply the reason. The trouble begins when those explanations arrive faster than anyone can read them.
The appetite that grew with the work
Dubov and Dmitry Isupov founded Chattermill in 2015 at Entrepreneur First in London. In his account of the early company, Dubov describes an ambition to automate the analysis agencies and advanced research teams were doing by hand. HelloFresh and MOO were early partners. The founders initially thought of the product as an analytics service for feedback.
Working on it enlarged their idea of the problem. Customer comments were scattered across products and departments. They came in different languages. Making sense of them was difficult; making the resulting insight useful was another job again. Dubov’s retrospective reaches for a saying from his native Russian: “appetite comes as you eat.”
It is a pleasantly domestic description of an expanding software ambition. One solved problem reveals the next plate. A company might learn that customers dislike part of its app, then need to know precisely which interaction causes the frustration, who should fix it and how urgently. The useful unit of insight becomes smaller while the organisational challenge becomes larger.
A conceptual guide to the work, rather than a measured product funnel.
By December 2017, Dubov was announcing a round that brought the company’s total funding to £935,000. The announcement carried a telling detail from Transferwise: more than a hundred people across the business were using Chattermill. Feedback analysis was already reaching beyond a solitary researcher’s desk. The point of listening was becoming a shared working practice.
Before the software, a conversation
For World Product Day in 2021, Dubov described how he begins thinking about a new product: an unbiased, critical conversation with a customer or prospective customer about their problems. Several conversations can reveal recurring patterns. From there, he looks for useful parallels, sometimes in entirely different fields.
This is a modest starting point for a company built around machine learning. A conversation has no impressive loading screen. It also permits the customer to spoil a promising idea before the engineers have spent months making it elegant. Dubov’s description keeps the customer’s difficulty at the beginning of the process.
“both Dmitry and I are product people at heart”
Mikhail Dubov · on Chattermill’s culture, 2021
The same emphasis appeared in his 2022 discussion of customer surveys. He favoured simple, open questions that let customers speak in their own terms. A questionnaire can become so carefully constructed that it leaves little room for an unexpected answer. A business then discovers, with admirable statistical confidence, what it had already decided to ask.
His case for machine learning depends on leaving that room open. Freeform language creates more work for the analyst, but it can reveal something a predefined response never offered. Chattermill’s job is to make that freedom manageable at volume.
The books that reached the office
Dubov’s reading habits offer a more personal view of how he learns. In a 2019 interview, he described listening to audiobooks while commuting, at lunch and sometimes while walking around a town. He read shorter pieces on an iPad and said he did not take notes. His favourites included Shoe Dog and Moonwalking with Einstein, chosen partly for the quality of their storytelling.
The picture is of someone fitting ideas into ordinary intervals. A walk becomes reading time. A book provides another person’s experience to set beside your own. For a founder, that can be useful company: running a business does not come with a neatly sequenced syllabus.
In January 2020, Dubov made the connection explicit in an essay about books he had learned from. Patty McCord’s Powerful had encouraged him to make Chattermill more transparent about performance, objectives and decisions than initially felt comfortable. Patrick Lencioni’s The Five Dysfunctions of a Team brought lessons about vulnerability and disagreement.
He did not accept every author wholesale. His response to Ray Dalio’s Principles mixed useful takeaways with substantial reservations. Reading, in this account, includes arguing with the book. The personal detail matters because it puts a specific management change beside the abstract vocabulary of company culture. Openness had to be practised by the person setting the limits.
A billion pieces, one decision at a time
Chattermill raised an $8 million Series A in 2020. On December 6, 2022, it announced a $26 million Series B led by Beringea. At that point, the company had an 80-person team and customers including Amazon, Uber, H&M and Zappos. The financing was intended to support expansion in the US and Europe and further development of the platform.
Dubov and Isupov co-found Chattermill.
Series A announced in 2020.
Series B announced in December 2022.
Funding figures describe the company’s rounds.
Dubov set an ambition for the AI to analyse more than a billion pieces of customer feedback. The number measures the scale of the task, but a company still has to decide what to change on Monday. Large collections of comments are valuable only if the analysis retains enough detail to guide that choice.
In his later writing, Dubov identified three barriers that software alone does not neatly remove: unclear ownership, findings disconnected from revenue or retention, and information trapped between departments. Each is an organisational problem. An accurate finding can still arrive at a meeting where everyone agrees and nobody takes responsibility.
His argument asks an awkward question of a successful presentation: what happened afterward? A polished slide can make a complaint look as though it has been dealt with. The person who made it would probably prefer a repaired experience. Dubov’s focus on outcomes gives customer intelligence a test outside the analytics screen.
Taking the interviewer’s chair
In 2025, Dubov began hosting How To Love a Customer. Its guests have included customer and product leaders from E.ON Next, Footasylum, Miro, IKEA and Nike. The format gives his long-running interest in listening a literal expression: the CEO asks questions while someone else supplies the experience.
One conversation, with inDrive’s Olga Ivanova, sharpened a problem that a tidy dashboard can conceal. Automating common reasons for contacting support may lower ticket counts while leaving the underlying friction in place. Some customers can leave quietly instead of complaining again. Dubov discussed Ivanova’s description of this as a carpet under which problems get swept.
The image is useful because it makes an invisible failure easy to picture. A declining number looks encouraging until someone asks what it actually counts. His podcast conversations return customer intelligence to individual decisions and operating details. They also extend the circle of people contributing to his thinking beyond his own company.
Faster answers need better questions
On September 17, 2026, Dubov introduced Lyra Agent in a Chattermill product essay. The agent is designed to answer questions about feedback and surface findings proactively. It works on intelligence prepared by Chattermill’s analysis engine, with evidence available for users to examine. Early access opened with the announcement.
The argument behind the release continues a familiar concern. As AI makes execution faster, deciding what deserves to be built becomes more consequential. Speed can multiply the cost of following the wrong idea. Dubov’s proposed answer is to bring customer understanding into the decision before the work accelerates.
His essay also lays out future plans for richer business context, audit trails and connected memory. Those are ambitions for the product’s next stages. They return to the same practical difficulty that confronted the early company: a finding needs enough context to be useful to the person receiving it.
And then comes the user, with a question nobody rehearsed. Dubov’s weather worry is funny because the distance between a product’s intended purpose and its actual use can fit inside a single text box. After a decade spent studying what customers say, he still has to wait for them to say it. The launch opens the conversation again.