As AI adoption accelerates, a leadership gap is opening. Leaders have invested in AI tools and platforms but haven’t aligned ownership, accountability and governance in ways that protect institutional expertise and create a competitive edge.
“This disconnect is how you end up with one foot on the accelerator and one on the brake,” says Derreck van Gelderen, global head of AI strategy at PA Consulting. He explains that the most common mistake leaders make is treating AI deployment as a technology decision: “Deploying AI is a series of leadership decisions: speed versus assurance; build or buy; and weighing up which enterprise knowledge gives you a competitive edge versus what you can safely open up to share with partners and AI platforms.”
Safeguarding and supercharging enterprise knowledge
Proprietary knowledge is critical for operational efficiency, informed decision-making and minimising risks.
At Sellafield, the UK’s first nuclear power station and now a decommissioning and reprocessing site, engineers relied on technical archives of critical documentation, from operating procedures to safety cases, stretching back more than 60 years. Finding or updating just one of these documents could require months of manual research. Sellafield also faced a major knowledge gap, with institutional knowledge lost as experienced staff retired.
To protect enterprise knowledge, Sellafield and the Nuclear Decommissioning Authority worked with PA Consulting to develop DANI2, the industry’s first AI agent, designed to transform how engineers access institutional knowledge in real time.
“For decades, the answer to knowledge loss in industries like nuclear was digitisation, but this meant engineers spent weeks searching through digital folders instead of paper ones,” says van Gelderen.
“We now have the technology to let the employees of the future have a conversation with those of today, to ask a question and get contextual intelligence at the point of need. For an industry where critical expertise is retiring faster than it can be replaced, it changes the entire equation.”
Reimagined decision-making
Carl Dalby, head of AI & digital at Nuclear Decommissioning Authority Group, says: “This project has been a resounding success, truly challenging the status quo in how we can embed AI in our approach to knowledge management and decision-making in the nuclear industry. The PA team has done a fantastic job in bringing this to life and helping train our team so that we can be self-sufficient.”
“With the use of AI, we’re turning dusty archives into living intelligence,” says van Gelderen.
From the outset, the programme prioritised governance and expert oversight. Engineers helped identify more than 80 potential AI use cases before narrowing the focus to those delivering the greatest operational value. Validation frameworks were built into the system so subject-matter experts could verify outputs, while guardrails ensured responses remained within defined operational boundaries.
Proprietary knowledge is critical for operational efficiency and informed decision-making
“There’s a temptation to hand everything over to the model and optimise for speed. But when you’re dealing with safety-critical knowledge, the governance architecture matters as much as the AI architecture,” van Gelderen says. “DANI’s design means subject-matter experts stay in the loop, outputs are verifiable and Sellafield retains full ownership of its institutional knowledge. The goal was to ensure judgement is informed by 60 years of expertise, not whatever someone can find in the time they have.”
Ownership and access
With AI models, organisations can inadvertently share proprietary expertise with external platforms without clear ownership and architectural control.
“Process templates, common standards and widely understood methodologies can sit on a shared platform. But domain expertise built over decades, or tacit operator knowledge, must be protected,” van Gelderen says. DANI2 gets this right. By integrating multiple language models, including open-source options, Sellafield avoids locking critical knowledge into a single vendor’s ecosystem. The infrastructure stays flexible and the intelligence stays theirs.
Turning enterprise insight into innovation
Accumulated expertise remains the most underused growth asset most organisations own. Used well, enterprise knowledge becomes the raw material for everything organisations haven’t built yet. This means the return on AI shouldn’t be measured in hours saved, but in the ideas that wouldn’t have existed without it.
With consumer expectations evolving rapidly across global markets, consumer goods company Unilever needed new ways to translate insights into faster product innovation.
Working with PA, Unilever developed DelphiAI, a platform that consolidates market research, consumer feedback and formulation expertise into a single intelligence layer. The system enables R&D and marketing teams to interrogate large datasets and generate recommendations for product features, ingredients and market-specific positioning.
“We wanted to use data to look beyond annual product and brand planning cycles to come up with better ideas, more quickly, that unlock the consumer delight that we strive for,” says Kumar Subramanyan, director of digital R&D at Unilever.
For PA, the programme highlights the importance of balancing central governance with driving innovation.
“By putting the right decision forums, leadership ownership and ways of working in place, we ensured the programme could move fast without fragmenting,” says Nyree Basdeo, digital, AI change and transformation expert at PA Consulting.
Everyone on board
Building quicker buy-in and AI adoption meant helping Unilever’s workforce understand the rationale and benefits of DelphiAI – particularly how the platform arrived at its decisions. By grounding adoption in trust and transparency, Unilever ensured DelphiAI strengthened enterprise knowledge – ensuring AI functioned as a strategic growth asset.
“With DelphiAI, we balanced innovation and IP by designing shared data layers for co-creation, while ring-fencing proprietary models behind controlled interfaces,” says Richard Chamier, business strategy and data science expert at PA Consulting.
The shift to next-level strategic asset
As AI capabilities rapidly commoditise, competitive advantage will come from how deliberately leaders protect and mobilise proprietary knowledge. The intelligent enterprises pulling ahead recognise that it’s this knowledge that determines where speed creates value, where caution is essential and how AI strengthens decision-making. Not a by-product of AI, but its foundational differentiation.
As AI adoption accelerates, a leadership gap is opening. Leaders have invested in AI tools and platforms but haven’t aligned ownership, accountability and governance in ways that protect institutional expertise and create a competitive edge.
“This disconnect is how you end up with one foot on the accelerator and one on the brake,” says Derreck van Gelderen, global head of AI strategy at PA Consulting. He explains that the most common mistake leaders make is treating AI deployment as a technology decision: “Deploying AI is a series of leadership decisions: speed versus assurance; build or buy; and weighing up which enterprise knowledge gives you a competitive edge versus what you can safely open up to share with partners and AI platforms.”
