The challenge

Taylors of Harrogate has been crafting high-quality coffee since 1886, building a strong leadership position in the Roast & Ground category. But like many brands, they faced a familiar tension: what people say doesn’t always match what they do.

With an upcoming innovation in the pipeline, the team knew the say–do gap would play a critical role. Traditional research could capture intent, but not always behaviour. The question became clear: how can we get closer to real-world decision-making and anticipate where intention might break down?
At the same time, Taylors were curious to explore the role of emerging methodologies. Could synthetic participants help go further than primary research alone and provide a more predictive view of consumer behaviour?

The solution

We designed a two-step approach, combining deep human understanding with AI-powered simulation.

First, we built a rich picture of reality. Through a week-long online insight community, consumers shared their coffee rituals, motivations and trade-offs via diaries, in-store missions and discussions. This allowed us to map the tensions shaping behaviour and identify where intention starts to unravel in the path to purchase. The learning was clear: while Taylor’s innovation resonated strongly, success would depend on reinforcing the right messages at the right moments.

This were useful learnings, but insight alone isn’t always enough. The real challenge lies in translating and applying those insights over the next few iterations of the innovation.

So, in a second step, we created a twin community. These are AI-powered digital twins grounded in the lived experiences of real participants. These twins allowed us to simulate scenarios, test unasked questions and explore “what if” decisions at speed. Through a grounding score we measured how much each simulation is rooted in real human data versus synthetic generation.

• When simulations closely matched the original research, grounding scores reached up to 89%, giving strong confidence in the outputs.
• As scenarios moved further away from known data, scores dropped. This made it clear when AI was “filling in the blanks.”

This created a powerful balance: the speed and flexibility of AI, anchored in the depth and truth of human insight.

To ensure this balance held up in practice, we rigorously validated the twin community. We tested reliability, looking at consistency over time and saw strong test–retest alignment (up to 85%). We also assessed validity. That’s whether the twins could accurately predict real-world outcomes. In blind simulations, they correctly anticipated consumer preferences in the majority of cases, giving the team confidence in the direction of decisions.

At the same time, we remained conscious of limitations. Twins are grounded in existing data, making them less suited for true white space exploration. And like all AI, they can lean towards overly rational or “idealised” behaviours. Which is why human interpretation remains critical at every step.

The impact

By combining human-first research with AI-powered simulation, Taylors gained a more dynamic and actionable understanding of the say-do gap.

The twin community became a strategic companion, extending the life of the research, stress-testing decisions and accelerating iteration. It allowed teams to explore more scenarios, faster and move from insight to action with greater confidence.

Crucially, it helped bridge the gap between knowing and doing. Insights didn’t just inform decisions, they were actively applied, refined and pressure-tested along the way.

The result? Smarter choices, made with clarity and conviction, grounded in real human behaviour and ready for what comes next.