From Research Findings to Business Decisions

For most of its history, market research ended the same way. Someone wrote a report, built a deck, presented it to a room, and hoped the findings would influence a decision. Sometimes they did. Often the deck sat in a shared drive and was never opened again. The industry's chronic complaint was the "last mile" problem: good research, poorly used.

That last mile is being rebuilt. An institutional memory system powered by a large language model can now read across every study a company has ever run and produce a single, cited answer to a new question — something enterprise search never managed. An agentic analyst can run a research pipeline end to end, from sample design to tabulation, supervised by a human who checks the output rather than doing the work. A deliverable can be an interactive instrument rather than a static file, queried and filtered by whoever needs it. A forecast can be assembled not by a single model or a single expert but by a hybrid of prediction markets, polls and machine-generated estimates, each correcting the others.

This chapter follows the insight after it is produced. It is about what the research becomes once it leaves the analyst, and how that is changing faster than most of the industry has noticed.

Why this matters to you

If you are a decision-maker who uses research, these changes affect you directly. The archive that answers back means you can ask a question and get a grounded response drawn from years of accumulated evidence rather than waiting for a new study. The agentic analyst means some categories of research run faster and cheaper, but only if someone competent is supervising the machine — unsupervised, it will produce confident nonsense at scale. The rebuilt deliverable means you can interact with findings instead of reading about them. The hybrid forecast means better predictions, if you understand what each component contributes. And the democratisation of research tools means more people in your organisation can run studies, which is a gain for speed and a risk for quality. This chapter helps you navigate all of it, and it closes on the two questions you did not expect to face: what happens to the insight function itself when much of the labour is automated, and what happens when the customer at the end of the chain is not a person but an AI agent making a purchasing decision on someone's behalf.

What you'll find inside

The pieces here follow the research output from storage through delivery to its strangest new destination. You will start with institutional memory — the archive that answers back — and what it takes to build one that is grounded rather than hallucinated. You will meet the agentic analyst and the supervision problem it creates, and the deliverable rebuilt as software rather than a document. You will see always-on research, the shift from discrete projects to a continuous nervous system, and forecasting hybrids that outperform any single method. You will watch agents war-game a competitive market before a launch. You will confront democratisation and its discontents — what happens when everyone in the building can run a survey. And the chapter closes with two pieces that point forward: marketing to machines, where the customer is an AI agent, and the insight function rebuilt, fewer hands, more judgement, and a different kind of career.

The honest note

The caution for this chapter is about amplification. Infrastructure amplifies whatever quality went in, errors included, and it amplifies them silently. An institutional memory built on poorly sourced studies will produce fluent, confident, cited answers that are wrong. An agentic analyst that is not supervised will make the same mistakes a junior analyst makes, except at pipeline speed and without the junior analyst's hesitation. A democratised research tool in the hands of someone who has never designed a question will produce data that looks real and is not. Every piece in this chapter is a capability gain, and every one of them makes the quality of the input more consequential, not less. The systems are faster and more powerful. They are not safer. The safety comes from the human who knows what to check, and this chapter is honest about the fact that that human is now the bottleneck.

The nine pieces in this chapter

1. "The archive that answers back" — institutional memory
2. "Research that runs itself, supervised" — the agentic analyst
3. "From deck to instrument" — the deliverable, reinvented
4. "From projects to a nervous system" — always-on research
5. "Polls, markets and machines" — forecasting hybrids
6. "Rehearse the market" — war-gaming with agents
7. "Everyone's a researcher now" — democratisation and its discontents
8. "When the customer is an agent" — marketing to machines
9. "Fewer hands, more judgement" — the insight function, rebuilt