In our own daily work, whether on the agency side or managing brands, we talk constantly about combining AI with human intelligence. It’s a great line for a pitch slide or a board deck. In this chapter, we set out to explain what that actually means in practice.

Turns out that’s not so easy. AI is so embedded in the research process and being without it now feels like cutting the WiFi. It’s indispensable when it works and we grow more fluent and more dependent with each passing day.

On faster and cheaper, AI wins comfortably. Sets you thinking, if machines can process data faster and cheaper than any human team then what is the distinctive contribution of researchers?

In adopting more AI in more parts of our work and more often, have we outsourced our knowledge and experience for the sake of getting the job off the list? It’s a question of accountability. Who actually owns understanding? Is it the system that processes the information or the person who decides what that information means?

Humans served by AI

We tend to paper over this question of accountability by wrapping it in the comfortable language of ‘AI plus human intelligence'. It’s become our standard industry positioning because it sounds balanced. It promises augmentation instead of replacement.

In practice, AI has already settled into parts of the research workflow, most of them operational and increasingly interpretive.

Seen this way, the distinction is more nuanced than it appears. Rather than a clean split between processing and interpretation, AI’s role is increasingly to generate and organise possible meanings. It can cluster themes and draft implications at warp-speed. The uniquely human task is deciding which of those meanings are actually valid, useful and safe to act on.

To our minds, the value shifts entirely from producing analysis to producing meaning. Anyone or anything can generate a pattern. The real human task is deciding whether that pattern actually matters and attributing meaning to it.

Consider a pack design test. The AI platform instantly aggregates eye tracking data and flags a massive success: a bold new logo is drawing huge visual attention.

The machine registers a winning signal. An experienced researcher applies experience. That high attention score isn't an index of desire. Consumers are staring because they are confused. The human exercised the judgement to take other information into account and urge caution.

Hybrid intelligence

We believe the future of consumer insight lies in hybrid practice rather than pure artificial intelligence.

Artificial intelligence replaces individual tasks, but hybrid intelligence entirely reshapes human roles. When you look at how this plays out on live projects, three patterns emerge:

1. The efficiency engine

The first role for AI is straightforward. Much of research contains structured, repeatable tasks that historically required substantial human labour. AI handles these quickly and reliably. This should reduce cost and shorten timelines.

Efficiency alone does not produce insight. Still, it accelerates the pipeline through which insight may emerge. Automation improves the mechanics of research but alone it does not resolve interpretation.

2. The collaborative colleague

A second model appears in more exploratory workflows, where AI becomes less a tool and more a partner in analysis. Used well, this dynamic accelerates thinking and enables rapid exploration and progression of ideas.

At the same time generative AI notoriously struggles to maintain a consistent analytical lens across complex datasets. If a project involves comparing multiple consumer sub-segments, the machine frequently drifts, applying different evaluative criteria to different groups.

This is why the ping-pong dynamic requires high experience. The researcher must constantly audit the machine’s consistency, acting as the stabiliser across the data.

3. The evidence amplifier

The most interesting role for AI emerges when it expands the evidence base of insight work. Many modern datasets are too large or complex for traditional analysis alone. AI systems can detect patterns that would otherwise remain invisible.

Use cases in this area include identifying emerging consumer trends from social media chatter, spotting anomalies across millions of transactions, uncovering hidden correlations in long-term customer engagement and mapping complex pathways in multi-channel journeys that humans will not easily track.

But expanded evidence does not automatically produce understanding. Data does not explain itself. Insight emerges when patterns are interpreted and placed within context. More evidence is only an advantage if you have the experience to cross-examine it.

The human edge

If machines dominate speed and scale, the human contribution becomes clearer:

1. Researchers act as narrative anchors, weaving together multiple sources of evidence and deciding which signals deserve attention.
2. They also function as cultural translators. Consumer behaviour sits within wider social and economic contexts. AI can detect statistical relationships. Humans explain their significance.
3. Finally, there is ethical responsibility. Research involves judgement about data sources, representation and interpretation. AI processes information. Responsibility for the conclusions drawn from that information still rests with people.

Why human intelligence cannot be abdicated

There is also a structural reason human interpretation remains necessary. AI systems analyse what is available to them but don’t recognise what is missing.

Large organisations are fragmented. Data sits across departments, platforms and programmes of work that rarely connect cleanly. Humans navigate these gaps.

Representation presents a further challenge. AI reflects the structure of its inputs. If certain groups are absent or poorly represented, the system may produce confident conclusions based on incomplete evidence. The machine cannot easily identify what it was never given the capacity to observe.

Worse, even with perfect inputs, the machine requires constant policing. True hybrid practice dispels the myth that we can trust the system to run flawlessly in the background. In reality, a significant portion of human time is spent checking that the machine has actually done its job properly.

Blind trust in the machine is not acceptable in the business of evidence; a human driver must step in to identify where it has misread nuance, smoothed over critical friction or simply cut corners.

In a fast moving environment, detailed prescriptions age quickly. What appear more durable are several guiding principles:

• Researchers must adopt a deliberate mindset about where AI adds value and where interpretation remains essential. Hybrid intelligence requires conscious design.
• Data quality becomes even more critical. AI amplifies whatever it is fed, which means source transparency and methodological discipline matter more rather than less.
• However, this pivot introduces a sharp craft problem. The entry level work of the past was also the training ground where junior researchers developed their category intuition. If we automate the mechanics entirely, we are hollowing out the next generation’s ability to interrogate the machine. True capability evolution means we must intentionally design new ways to build human expertise. You cannot cross-examine an AI output if you do not deeply understand the data it is summarising.

The work starts now. Hybrid models are still forming and the organisations that learn fastest will shape the future of insight.

Sarah De Caux is lead analytics and insights manager at Co-op and Adrian Sanger is co-founder at Fore Consulting

• The full Delphi report, ‘Who owns understanding? How AI is reshaping value, expertise and accountability in research’ is accessible via the MRS website.