Deep Reads: When Answers Get Cheap
Picture a research report landing. Thirty-odd slides, a confident summary up front, and someone in the meeting says thank you, this is really useful — and then the meeting carries on without it.
0:00 / 10:22
Deep Reads: When Answers Get Cheap
This edition reads across the last fortnight rather than the last week, because the pieces worth your attention landed on both sides of the calendar. All three end up in the same place: the gap between a finding arriving and a decision actually changing. Producing research gets cheaper and faster every quarter, so the question underneath all of them is what the work is worth now — and what has to be true about a finding before it is worth anything at all.
Who expects to buy research from a machine
Nelson Whipple runs the GRIT survey at Greenbook, and he published what it found on how research will be bought. The question put to the industry was blunt: will automated systems — AI agents — mediate most purchases of research and insights services within a few years? 1
Who agreed is the interesting part. Two thirds of brand-side people whose work leans on analytics said yes. Just over a third of those whose work leans on traditional research said the same. The gap is not noise: Whipple lines belief up against budget, and belief tracks the money. 1
About a quarter of the brand-side believers run research budgets above fifteen million dollars, against roughly one in fourteen of the doubters. Believers were also far more likely to have seen budgets rise, teams grow and outsourcing increase over the past year. On the supplier side the split runs the same way, with belief concentrated in the large houses. 1
Then the survey turns to what agents already do. Three quarters of the brand-side believers use them to analyse or model data, and roughly the same share use them to build reports and dashboards. Fewer than a third let an agent pick a supplier. 1
The criteria buyers name for choosing a supplier do not specify equally well either. Price and use of technology are easy to write down. Service quality, reputation and relationship are abstract and context-dependent. Whipple's practical reading: if a buyer hands an agent a shortlist, the job is to be on that shortlist.
The boundary to keep. This is the industry reporting its own expectations, collected by the person who runs the survey. It tells you who is positioned to act on the change. It cannot tell you what will happen.
An insight is a half product
Ed Keller runs the Market Research Institute International, and for two years he has co-hosted a podcast called Insights and Innovators. On October 2 he read his own interview archive as a single document — nearly ninety conversations with brand-side leaders, agency people, academics and founders. 2
His conclusion is that the value frontier for insights work is moving. He means where value sits, not how much of it there is. Oksana Sobol at Clorox calls an insight a half product: the real product is what comes next, whether that is an idea or a business decision. Pam Forbus, who runs global insights at Mondelez, draws the contrast as good teams answering questions while great teams help shape decisions. 2
Keller is careful about what this does not mean. Excellent research has not become unimportant; it has stopped being enough on its own. If the output is research, a model can increasingly produce it. If the outcome is a growth idea or a better decision, that part got more valuable. His worked example is foresight work at Frito-Lay that helped make the business case for new manufacturing capacity inside what became a multibillion-dollar business. 2
Two lines travel well. Stan Sthanunathan, who ran insights at Unilever, puts it as: it is not about how smart you are, it is about what decisions you can actually influence. Elaine Rodrigo at Reckitt pushes the same idea back to the start of an assignment — begin with the business question, not the research objective, and sometimes conclude that no new research is needed.
The boundary to keep. This is a synthesis of a podcast: conversations the author chose, with guests who agreed to be on it, read by someone who also takes this argument onto conference stages. It is a considered point of view rather than a study.
Three traps
Monika Rogers leads growth at CMB, a research firm, and wrote up what she took from a business summit at the University of Wisconsin. She names three traps organisations keep walking into. 3
The first is automating tasks instead of enabling decisions. Pointing AI at coding open-ends, drafting takeaways and summarising reports are all sensible places to start. The trap is letting that efficiency become the strategy: more analyses, more summaries, the same decisions underneath them. Rob Bradford of Kimberly-Clark reframed his team's unit of redesign away from jobs and departments to repeatable pieces of work, so that each piece could be sorted into what a model does well, what it should assist and what needs a person. 3
The second trap matters most. Load research into an AI system without its context and access scales misuse. A study's value is not only its findings; it is the audience definition, the field dates, the method, the decision it was commissioned for, the confidence level and the caveats. Strip those away and an answer can be faithful to the document while being wrong for the decision. Her example: a product team asks an internal agent which audience is likeliest to adopt an offer, the agent answers cleanly, and the study behind it sampled existing customers only, ran before a price rise and was labelled directional in an appendix. Her remedy is to make every reusable finding carry its own context, and to tell the agent which source wins, which populations count, how old evidence may be, and when to refuse. 3
The third is drift, and not only the model's. Meaning drifts too: a nuanced body of learning gets summarised over and over until the qualifications fall off it. Rogers points at a federal appeals court that sanctioned two attorneys in June over briefs containing fabricated cases, and was explicit that the failure was not the use of AI but the signing and filing of work nobody had verified. Her closing contrast is the same one the other two reads reach from other directions — efficiency is easy to demonstrate, enterprise impact is not, and it is the only thing that decides whether any of it was worth doing. 3
What to take into the week
Three reads, three angles on one shift. The buying side is starting to hand the specifiable work to machines and keeping the work that cannot be specified. The value frontier is moving from producing the research to changing the decision. And the way that goes wrong is almost mechanical: automate the output, strip the context, and let the meaning wear off one summary at a time.
Notice what all three put in the centre. Not the report, not the deck, not the answer — the decision, and whether anybody is willing to stand behind it. So before you commission, produce or forward a piece of research this week, name the decision it is meant to change, then check whether the context that makes the finding usable is travelling with it. If the decision has a name and the context is attached, you have something worth acting on. If either is missing, you have an answer that is cheap to produce and expensive to use.
References
- 1
- 2
- 3
This story was produced automatically by a channel. One sentence is all it takes for Neodrop to keep producing for you.