Building a research function that teams actually use

When I joined Selfridges there was no research function. There were opinions, there was trading data, and there were designers who spoke to customers occasionally when a project allowed for it. None of that's unusual, and none of it's a criticism of the people who were there.

The first thing I'd to work out wasn't what method to use. It was who would trust the output. I could have run a beautiful set of studies that nobody read, and in the early days I probably did run one or two of those.

What changed things was starting small and being useful quickly. I found the decisions that were already being argued about, offered to test them, and made sure the answer arrived before the decision had to be made. Being on time matters more than being comprehensive.

We went from roughly three studies a year to running something every week. That sounds like a budget story and it partly is, but the budget was about £60k a year, which isn't a large number for a business of that size.

The real shift was making research a habit rather than an event. When a study is a big deal, everyone wants to load it with every question they have ever had, and it takes six weeks to organise. When it happens on Thursday every week, people bring one question, get an answer, and bring another one the following week.

That rhythm changed the conversation. Instead of arguing about what customers might think, teams started saying we can put that in front of someone this week. It also meant the cost of being wrong went down, which made people more willing to try things.

One of the things I've had to get more honest about is that research doesn't make decisions for you. It reduces the range of things you might reasonably believe, and then someone still has to choose.

I only really put words to this during the L7 programme. I'd spent a long time treating more evidence as always better, and there's a point where gathering more is just a way of avoiding the commitment.

So I try to be clear with teams about which part we're in. Are we still genuinely uncertain, or are we uncertain because nobody wants to own the call? Those need different responses and confusing them wastes quite a lot of time.

Getting research to travel across an organisation is harder than running it. A finding that's obvious to the people in the room can be completely invisible to a team two floors away who are making a decision that depends on it.

I've tried repositories, showreels, newsletters and standing sessions. All of them help a bit. None of them solve it, and I think the honest answer is that repetition and relationships do more work than any tool.

The other ongoing difficulty is research on things people can't easily articulate. Asking someone why they abandoned a basket usually produces a tidy story that may or may not be what happened. In those cases I lean much more on behavioural evidence and treat the interview as a source of hypotheses rather than answers.