Research Doesn't Have A Data Problem. It Has A Memory Problem.
There’s a question I’ve hated getting throughout my career in research, and it usually comes just after the debrief. You present the findings, everyone discusses what they mean, and then someone asks, “How does this connect to that other project we did?”
At an agency, that other project might have been run by someone else, for a different team, two years earlier. In-house, it might be sitting in a deck on someone’s drive, alongside dozens of other studies that were incredibly useful at the time and have barely been opened since.
The insights industry spends a lot of energy on getting more data, faster. We talk about faster fieldwork, bigger samples and running more studies, and all of those things matter. But in my experience, some of the most expensive waste in research comes from the knowledge organisations have already paid for and can no longer easily access.
We saw the problem in how customers were pushing the boundaries of Ideally. Treasury Wine Estates was looking for consistent commercial signals across a large portfolio project. Another customer had gathered a wealth of discovery data, then found themselves using Excel and Miro to piece it all together.
Research is filed like documents. Questions arrive like conversations.
Every study tends to be saved in roughly the same way: one project, one report, one folder. That structure makes sense when the research is new, but far less sense six months later when someone asks a question that cuts across several different projects.
A marketing director might ask what you know about pricing, when the answer is spread across four tests from the past 18 months. A brand manager might want to know whether a particular claim worked for younger consumers, when that finding is buried on slide 34 of a concept test nobody has opened since it was presented.
Answering either question properly means someone has to remember which studies exist, open them and piece together the evidence. That person is usually a researcher, whose time is already one of the scarcest resources in an insights function. As a result, valuable questions can end up with partial answers, gut-feel answers or no answer at all, even though the organisation has already paid for the knowledge needed to answer them.
The knowledge exists. The problem is being able to use it when you need it.
Finding the insight is only half the job
There’s another part of research that is harder to see, because finding the relevant evidence is only half of what a good researcher does. The other half is translating that evidence for the person who needs to use it.
A head of marketing, a brand manager and a sales lead can ask essentially the same question and need three very different answers. Knowing which part of the evidence matters to each person, how much context they need and how to explain it in language that makes sense for the decision they are making takes years of experience.
That expertise is incredibly valuable, but it doesn’t scale particularly well. The people closest to a decision can end up with raw data they don’t have time to interpret, or waiting for a researcher who doesn’t have time to answer. In both cases, there is a gap between having the insight and actually being able to use it.
It’s one of the reasons I’ve been interested in how AI can change the way people interact with research. The opportunity goes beyond finding a result faster. Done well, it can help people interrogate what they already know in a way that feels much closer to having a conversation with an experienced researcher.
AI has made this easier, and riskier
Language models are very good at producing answers that sound right, which is exactly what makes their use in research complicated. If a tool fills a gap in your evidence with general knowledge or a confident guess, it can create false certainty at exactly the point where someone is trying to make an evidence-based decision.
The standard for AI in research therefore has to be higher than whether it can produce a convincing answer. It needs to show its working, point claims back to the evidence behind them and be clear when the available research can’t answer the question.
The goal is to give people an answer they can interrogate and understand, with the confidence to actually use it.
What we built
That thinking is behind Insights Assistant, which we launched this month to every Ideally customer, on every plan. It’s the newest part of Assistant, the AI experience that already helps our customers build studies and draft presentations, and we’ve designed it to work like an insights analyst across every study you’ve run.
You can ask a question of your research in the same way you might ask a colleague. Insights Assistant searches across the studies you have access to and brings the relevant findings together, then lets you keep the conversation going. You might ask, “Was that true for older consumers too?” or “How did these two ideas compare?” and explore audiences and comparisons wherever the research supports them.
Every answer links back to the studies behind it, so you can see where the evidence came from. If your research doesn’t cover the question, it tells you rather than filling the gap with general knowledge. When a chart explains the answer more clearly than words, it can build one from a single source with the citation attached.
For teams using Ideally Canvas, category data can also come into the same conversation. You might ask, “Which of our concepts fits the biggest demand zone?” with study evidence and Canvas evidence separately cited so it’s always clear what came from where.
We built Insights Assistant with insights experts, and I spent a lot of time shaping the research approach behind how it responds to questions. It incorporates the research methods I use every day, alongside the marketing science thinking I believe people working in insights should understand. The ambition was bigger than creating a search engine for old reports. We wanted something that could approach a question the way a good analyst would, while still letting people interrogate the evidence themselves.
The real shift: research that compounds
The part of this that interests me most isn’t actually the speed. It’s what happens to a study after the results come back.
Today, most research has a relatively short active life. A study answers a question, the findings inform a decision and, eventually, the work disappears into a folder. When every study remains searchable, connected and answerable, that starts to change.
Each new test adds to what you already know, old studies become useful in new contexts, and the value of your research library grows over time rather than declining as individual projects become harder to find.
I think that also changes the role of the research team. Researchers can spend less time digging through archives to answer questions like, “Have we already learnt this?” and more time on the work where their judgement adds the most value: framing the right questions, challenging assumptions, connecting evidence and identifying what the data still doesn’t tell us.
Where this goes next
The version of this I’m most excited about is the one we’re building towards: an assistant that can hold the full context of what an organisation knows about its consumers.
That means eventually bringing research from outside Ideally into the conversation and giving people the ability to ask questions across an organisation’s collective knowledge, rather than relying on someone to remember which team ran which study and where the findings were saved.
For me, that gets at a frustration I’ve had throughout my career. There have always been pieces of information you know exist somewhere, but can’t access at the moment you need them.
We’re not there yet, but for the first time in my career, I can see a path towards it. It starts with treating research as a body of knowledge that should keep growing and becoming more useful, rather than a stack of finished projects we slowly forget.
