From Research Projects to Learning Systems: Six Signals for the Future of Innovation

Innovation has never followed a perfectly straight line. But if there was one message I took away from the recent MRS Innovation Pipelines Conference, it was that our approach to insight needs to stop pretending that it does.

Across presentations from Mars, Zappi, Unilever, Conveo, Ipsos and others, a common thread emerged: the future of innovation research is less about individual projects and more about connected, continuous learning. That means connecting evidence, connecting teams, connecting learning across different stages of the innovation process, and increasingly, connecting human intelligence with AI.

Here are six themes that stood out to me.

Perhaps the strongest theme of the conference was connection.

Traditionally, consumer research has often happened at specific points within an innovation process. Develop an idea, test it, refine it, then validate it. But innovation itself rarely works that neatly. Mars and Zappi, as well as Conveo and Unilever, talked about moving beyond a linear “test and validate” model towards something much more iterative: explore, learn, improve and repeat. In this model, consumer understanding isn’t a checkpoint. It becomes part of the innovation system itself, providing direction at multiple decision points and helping teams continuously build on what they already know. It’s a shift that resonates strongly with the journey we’re already on at MMR with our clients: moving beyond the delivery of individual pieces of research towards more connected intelligence that accumulates over time and helps businesses make better decisions.

That connected approach also changes what we consider “evidence”.

Ipsos talked about connected evidence, bringing product feedback together with behavioral and observational data to build a richer picture of what is really happening. That matters because consumers can only tell us so much. What people say, what they do and what they experience aren’t always the same thing. Connecting different sources of evidence gives us a better chance of understanding the reasons behind behavior, rather than simply recording the behavior itself. And that brings us closer to continuous, agile learning loops rather than isolated studies.

One of the more provocative ideas came from Ipsos: the rise of the second consumer.

As AI moves from assistant to agent, it could increasingly make decisions on our behalf. Ipsos cited research suggesting that 48% of consumers would already allow AI to make purchases below $50 without approval. Take that trajectory far enough and we reach agent-to-agent commerce, where machines evaluate, recommend, buy and potentially sell products for us. That creates a fascinating challenge for innovation.

Products may increasingly need to work for two audiences:

• The human, where they need to be desirable, distinctive and relevant.
• The machine, where they need to be understandable, comparable and compatible with the rules used by AI agents.

Unsurprisingly, AI featured heavily throughout the conference. But what interested me was how the conversation appears to be moving on. The initial wave of AI adoption has understandably focused on efficiency: Can we do what we already do faster, cheaper or at greater scale?

The more interesting question now is: Can AI help us learn something we couldn’t have learned before?

There was some debate around ideation in particular. Should the model be human-to-AI, as discussed by Mars, or AI-to-human, as Pladis proposed? Whichever direction you start from, the important principle remains the same. Credible innovation still needs grounding in real human evidence. AI can help us interrogate, connect and extend that understanding, but the value comes from what we learn, not simply how quickly we produce an answer.

That principle is also reflected in MMR’s work around synthetic data. Our experimentation has shown why technical capability alone isn’t enough. For example, generating additional synthetic respondents from a small quantitative sample does not automatically create greater statistical confidence, because the output remains dependent on the quality and representativeness of the original human data.

There’s another side to AI that is easy to overlook. AI isn’t only changing how we conduct research, it is changing the world we’re researching.

If consumers increasingly use AI to discover products, compare options and make purchase decisions, established assumptions about choice, decision-making and the consumer journey start to shift. That poses some big methodological questions. 

Perhaps paradoxically, all the conversation around AI ultimately reinforced the importance of people. AI can generate, synthesise, connect and accelerate at extraordinary scale. But successful innovation still needs to understand what people value, what they experience and why it matters to them. That puts a premium on high-quality human evidence. It also means the future probably isn’t human or AI. It’s learning how to combine the strengths of both.

That principle is already visible in MMR’s approach to emerging technologies. Our work on synthetic data, for example, explicitly advocates client collaboration, experimentation and scientifically grounded guidance rather than applying new technology simply because it is available. 

Put these themes together and I think a bigger change starts to become visible. We're moving:

From projects to learning systems.
From validation to continuous improvement.
From individual data sources to connected evidence.
From AI as an efficiency tool to AI as a learning tool.

For MMR, that direction feels particularly important because it’s a journey we’re already taking with clients: finding better ways to connect learning, build on existing knowledge and turn consumer and product evidence into intelligence that can shape the next decision, not simply explain the last one.

The technology will continue to change rapidly, but the goal shouldn't.