Nothing connected what customers did to what the company built
Insight came from third-party brand surveys, inconsistently, and rarely reached the people deciding. There was almost no direct customer contact during product development, so assumptions went untested. Departments were siloed with goals that did not line up. There were no standard metrics and no shared access to the ones that existed.
The symptom looked like missing research. The problem was one level down, in how decisions got made: there was no route by which something learned from a customer could reach the person choosing what to build.
The reporting line is the reason it was possible
A new Chief Growth Officer took over and I was his first key hire, reporting to him directly.
Building a research function is an organizational change before it is a hiring round. A function with no line to the executive table gets overruled by the next deadline, and then gets quietly rescoped into a service desk. That line is the single structural condition this work needs, and where it is missing I would say so before taking the job.
The infrastructure went in first
Dovetail as the central insights repository. Lookback for remote usability tests and interviews. QuestionPro for surveys. Calendly for recruitment. An interviewee pool the whole company could draw on.
More than 100 insights and over 50 interviews and usability tests went through it. Every study, survey and test landed in one searchable place, so a product manager in one brand could use what a researcher in another had learned, which in a group of more than 90 subsidiaries is the difference between research and 90 separate research efforts.
Then a team hired to teach
Eleven people: product researchers, product designers, and later the design system team. Ten of them I hired.
I hired for people who could teach the rest of the company how to work with customers. A function of eleven cannot serve a company of more than 11,000 employees by taking requests, so every hire had to raise someone outside the team. Craft was the entry ticket. The question that decided an offer was whether this person makes the people around them better.
A competency model before the first req. Ten hires at speed, across two disciplines and three countries, is where a hiring bar quietly becomes whoever the last interviewer liked. So the model came first: what each level of researcher and designer is expected to do, what separates one level from the next, and which of those are craft and which are the ability to raise others. It did double duty from the start, because the thing that decides an offer is the thing a person is later developed and promoted against, and the career ladder has to exist on the same page or the best practitioners leave for a title somewhere else.
Then a hiring process built on it. The old one was a sequence of conversations with no shared standard, which is fast to run and expensive to be wrong in. The new one scored against the model, used a work sample close to the actual job, and put the teaching question in front of more than one interviewer, so an offer was a decision the group could defend. Ten hires came through it and several of those people now lead teams themselves.
Decisions
Build the infrastructure before hiring the researchers. Planned: hire first. A new function with no people has nothing to show, and a headcount request is an easier thing to get signed than a tooling budget. Argued: a researcher who arrives into none of this spends two quarters building it anyway, except now it is being built in the gaps between studies by somebody who was hired to do something else. And the studies that run in the meantime are lost the day they are presented. Happened: the repository, the tooling, the recruitment pipeline and the pool went in before most of the team did. The first two quarters looked slower than hiring would have. Every study after was cheaper than the one before, and none of them went missing.
What I would do differently
Put a number on decision quality in the first quarter, for example the share of roadmap items backed by repository evidence.
The design system had a return on investment model and that is why it kept its funding through a difficult year. The research function had better stories and no equivalent figure, which made it the easier thing to question whenever budget got tight.