From pilots to production: how to embed AI in the everyday

Launching an AI pilot is easy. Scaling it so it changes how your business runs? That’s the hard part.

Across industries, the “AI Pilot Trap” is real: promising trials that never integrate, never scale, and never deliver lasting impact. McKinsey reports that while 70% of organisations pilot AI, fewer than 20% achieve meaningful business value. Why? Because scaling AI isn’t just a technology challenge, it’s about culture, operating model, and leadership intent.

In 2026, we expect to see a greater focus on AI payback, so we will share why we see pilots stall and what to do about it.

At dunnhumby, we see this pattern clearly both through our work with global retailers and brands, and through our own journey as a customer science business. Moving AI from promising experimentation to sustained business impact is not just a technology challenge. It is a question of culture, operating model, and leadership intent.

Why pilots stall and how to break free

From our work with global retailers and brands, and our own experience at dunnhumby, stalled pilots are rarely a technology problem. They’re almost always an operating model problem. Four common blockers stand out:

1. Innovation in isolation – Pilots built in silos, disconnected from everyday systems.
2. No clear ownership – Without senior sponsorship and a business case, projects drift into “interesting but not urgent”.
3. Underestimating the real work – Data integration, compliance, workflow changes, and training are essential but often invisible.
4. Cultural resistance – If people don’t trust or understand the tool, they won’t use it.

Scaling innovation isn’t just about smarter algorithms. At dunnhumby, we’ve learned that AI scales only when it is treated as an operating capability explainable, embedded in daily workflows, and tightly connected to how retailers actually operate, from promotions and assortment to media and measurement.

What works instead

Scaling AI means treating it as an operating capability: explainable, embedded in workflows, and aligned to real business problems. Here’s what we’ve learned:

• Start with the business problem – Tie every pilot to a clear outcome and define success upfront.
• Build for scale from day one – Think beyond the demo: robust data pipelines, governance, and compliance.
• Focus on people, not just tech – Adoption is human. Upskilling and visible leadership matter.
• Start small, prove value, then expand – Deliver one high-impact use case, then scale patterns.
• Embed governance early – Guardrails for fairness, transparency, and accountability are non-negotiable.

What scaling AI looks like in a customer science business

At dunnhumby, scaling AI means moving beyond experimentation and embedding intelligence into everyday decision-making.

• AI as business-as-usual: From targeting and measurement to promotions and media effectiveness, AI underpins how value is delivered day to day.
• Trust by design: Responsible AI principles ensure transparency, explainability, and confidence for retailers and brands.
• Closing the last mile: Cross-functional AI enablement brings science, product, engineering, and commercial teams together to turn insight into action.
• Built for reuse: Shared components, patterns, and best practices reduce cost, increase speed, and enable consistent scaling.
• A learning flywheel: Real data, real feedback, and continuous iteration drive ongoing improvement.

At dunnhumby, we’re focused on embedding AI into how we think, build, and deliver customer-first value. The next phase of AI transformation won’t be about running more pilots, it will be about scaling what works, responsibly and repeatedly.

And we’re just getting started.

References

• https://www.gsb.stanford.edu/insights/andrew-ng-why-ai-new-electricity
• https://nexla.com/blog/from-poc-to-production-breaking-the-ai-stalemate
• https://medium.com/@williamflaiz/why-your-ai-pilot-is-stuck-and-how-to-scale-it-to-production-1d0de31d1391