Market Research: Methods, Evidence and Decisions

Market research is in the middle of the biggest shift in its history, and most of the available commentary is either breathless or dismissive. The breathless version says AI has solved everything: instant insight, synthetic respondents, automated analysis, the end of the survey as we know it. The dismissive version says it is all hype, that fundamentals have not changed, and that serious researchers should wait for the dust to settle. Neither version is much use to the person who has to make a decision next quarter.

This project is the alternative. It is sixty-nine articles that take the post-2022 developments in market research seriously — each one built on verified evidence, each one honest about what the evidence does and does not show. It is not a trends report. It does not predict the future or rank vendors. It asks, for each new capability that has arrived: what has it actually demonstrated? Under what conditions? What does it break? What does it not yet know? And — the question that turns out to matter most — is the confident version you heard first the same as the version the evidence supports?

Who this is for

This project is written for the working professional who commissions, conducts or acts on market research. You may run an insights team, manage a brand, design fieldwork, build data products, or sit in a strategy role that depends on understanding what people do and why. You are not looking for an academic literature review; you are looking for a reliable account of what changed, told in plain language, that helps you spend your time and budget better. The articles are substantive but not technical. They assume you are smart and busy, that you have no patience for jargon that replaces an explanation, and that you would rather hear "we don't know" than a confident claim that falls apart on inspection.

How it is organised

The sixty-nine articles are arranged in seven chapters, each covering one part of the research process.

Foundations comes first. Before you can evaluate any new tool, you need to know what counts as evidence now that the cost of generating evidence has dropped to near zero. This chapter is the trust layer: provenance, validity, disclosure, the new version of the old question "how sure should I be about this?"

Asking covers the reinvention of elicitation. AI interviewers, conversational surveys, voice and multimodal inputs, cognitive pretesting at scale — and the new problem each one introduces.

Simulating is the most radical chapter. It takes the claim that useful answers can come from simulated people and tests it rigorously in both directions — where silicon samples and digital twins deliver, and where they flatten, parrot and mislead.

Observing follows the shift from asking to watching. Behavioural traces, attention measurement, predicted gaze, brand perception read by machines, stores that observe their own shoppers. Powerful and structurally silent on "why."

Modelling covers the engine that turns data into answers. The return of marketing mix models, causal inference, embeddings, knowledge graphs, uncertainty as a deliverable — and the chronic temptation to hide the uncertainty in the dashboard.

Deciding follows the insight after it leaves the analyst. Institutional memory, agentic pipelines, rebuilt deliverables, hybrid forecasts, the democratisation of research tools, and the strange new problem of marketing to a customer that is itself an AI agent.

Governing closes the corpus. The rewritten code of conduct, disclosure standards, consent, ownership, privacy engineering, regulation, the panel economy, and the question the whole project refuses to set aside: whether effective and ethical will travel together or apart.

How to read it

You do not need to read in order. Each article stands alone. The chapter introductions orient you to what a group of articles shares, and cross-references within articles point you to the connections. If you want the fastest entry, start with whichever chapter name matches the decision in front of you. If you want the foundation that makes everything else land better, start with M0.

The project has a sibling. An earlier corpus of over a hundred articles covers the behavioural science that underpins research practice — the mechanisms, dispositions, contexts and frameworks that explain how people actually behave. The two are designed to be read together: that corpus tells you what we know about people; this one tells you how we study them now.

The honest note

Three commitments run through every article in this project, and they are the closest thing it has to a manifesto.

First, provisionality. Every article in this corpus passed an admission test: the decisive development had to be post-2022, meaning the method did not exist or the evidence that now settles or unsettles it did not. That makes the material genuinely new and genuinely uncertain. The articles are honest about that. They report what the evidence shows, flag what it does not yet show, and treat "we don't know" as a finding rather than a failure. An article that could not meet the admission test was parked, not softened.

Second, verified sources. Every empirical claim is traceable to a specific study, dataset or public benchmark. The articles carry full reference lists and make clear which claims rest on strong evidence, which on early or limited evidence, and which on a single study that has not yet been independently tested. No claim is allowed to float on authority alone.

Third, the refusal to separate effective from ethical. The tools in this corpus are powerful, and several of them can be used to exploit the people they study as easily as to understand them. The project does not treat ethics as a chapter at the end; the governing questions run through every chapter from the start. The final chapter makes them explicit, but they are present throughout, because the industry's credibility depends on getting this right, and getting it right means never pretending the question is settled.