Deep, empathetic consumer insights

The tools and techniques we use to find what real people really want

The principle

You can’t optimise a proposition you haven’t truly understood.

Every winning innovation we’ve worked on began with the same thing: an unsatisfied consumer tension, identified in the consumer’s own language, with enough specificity that the proposition almost wrote itself.

That’s harder than it sounds, and it’s the place most insight programmes stop short. Most deliver categories of dissatisfaction — “consumers want healthier options,” “consumers care about sustainability” — when what wins is something a hundred times more specific. The Ehrenberg-Bass Institute’s finding that 88% of growing brands grew by finding more buyers through meaningful differentiation rests on this distinction: meaningful differentiation comes from a real felt tension, not a category truism.

So the question we ask, on every project, is: what’s the actual lived problem here, expressed in a way the consumer would recognise?

The human work that doesn’t change

There’s no shortcut around three things, and we don’t try to find one.

Depth interviews and ethnography. Sixty to ninety minutes with a real consumer in their own kitchen, bathroom, fridge or pantry. Watching, not just asking. The specific things people do that contradict the specific things they say.

In-home and in-store immersions. Where the proposition will actually have to compete. We’ve never run a category-growth workshop that didn’t benefit from at least one of the team having stood in the aisle that morning.

Quant segmentation, where the budget and brief justify it. Particularly for brands with broad portfolios where segment prioritisation drives commercial decisions.

This is the work that builds the proposition. Skip it and you’re optimising in the dark.

How we optimise at this stage

Insight optimisation means refusing to settle for the first plausible tension. We pressure-test what we’re hearing back against the consumers themselves — in small co-creation sessions, with single-question stress tests, with the explicit question “have we got this right, or have we got it nearly right?” The difference between an insight that converts to a winning proposition and one that doesn’t is usually whether someone asked that question one more time.

Where AI is now genuinely useful

The honest position: AI is over-hyped as a substitute for human insight, and under-deployed as a way to scale it. We’re focused on the second.

There are five places where AI now meaningfully extends what an insight programme can deliver.

Scaled qualitative synthesis. Tools like Notably, Marvin and Dovetail tag, cluster and theme transcripts at speed. Used well, they let one analyst do the synthesis work of three — which doesn’t reduce the human contact, it lets you do more of it.

Pattern detection across digital signals. Pulsar, Brandwatch and Talkwalker now go beyond mention counting. The current state of the art is narrative clustering — identifying the stories audiences are forming around a category, not just counting volume. Useful for spotting an emerging tension months before it shows up in tracking data.

Multimodal analysis at scale. The current generation of large language models can analyse hundreds of in-home photographs or in-store videos in a way that was previously impossible without a research assistant for a fortnight. We use this to find patterns across visual material that human researchers would miss.

Behavioural segmentation from transactional data, in hours not months. This is one of the places AI has genuinely changed the economics. Given a customer purchase dataset of meaningful size, we can now identify behavioural segments, characterise them, name them, and surface their commercial implications in a single working session — for a fraction of historical cost. We’ve done this on real client data, including a recent piece of work for Clipper Teas that surfaced four behaviour-based archetypes and their respective margin profiles in less than a day. What used to require a quant supplier, a six-week timeline and a five-figure invoice now happens in the room.

Avatars to bring segments and respondents to life. This is different from synthetic respondents — and the distinction matters. An avatar is a richly characterised, evidence-based representation of a real segment or research participant: face, voice, language patterns, motivations, all grounded in actual data. It’s a stimulus tool, not a substitute. Used in workshops, an avatar lets a brand team interrogate a segment, react to a proposition, and stress-test claims against a consistent human-feeling proxy. Used well, it’s one of the most powerful ways to keep real consumers in the room when they can’t physically be there.

Where we don’t go: synthetic respondents

The hype around synthetic respondents is currently the loudest noise in the category, and we’re sceptical for specific reasons.

The validation studies that show 85–95% accuracy are testing structured questions at aggregate level — pricing trade-offs, feature rankings, message preference among options. That’s the work where synthetic respondents look best, and it’s also the work where you arguably don’t need them. The questions that matter — what do consumers genuinely care about, what tension are they actually trying to resolve, what would they say in their own words if you weren’t leading them — are the questions where the technology is weakest. A 2025 Emporia study on B2B synthetic users found a strong positive bias and herd behaviour that systematically distorted findings.

Our view: if the question is qualitative, talk to real people. If the question is “could a synthetic panel save us a step?” — talk to real people anyway. The economics aren’t worth the integrity risk.

What Brand Development is doing today

We are not selling you an AI tool stack. We have integrated AI into our own working practice — primarily through Claude, used systematically against tone-of-voice profiles, evidence thresholds and a structured research framework — and we are actively building the methodology with clients project by project.

What this means in practice: where AI accelerates the work, we deploy it and pass the time savings on. Where it dilutes the work, we don’t. The judgment about which is which sits with the people who’ve spent thirty years doing this — not with the tool vendor.

The lever it pulls

What used to be a twelve-week insight programme can now be a six- to eight-week one, with no loss of depth. The synthesis happens in days rather than weeks. The pattern recognition reaches further. The hypotheses are sharper before the fieldwork starts.

The work itself — the consumer in their kitchen, the time in the aisle, the conversation that ends up shaping the proposition — that part doesn’t change. It shouldn’t, and it won’t.

[Read about Pillar 2: Differentiated proposition and compelling claims →]

[Or talk to us about an insight project →]