Thought Leadership: The 81% Rule: Leading AI from Pilot to Practice
September 12, 2026
London, UK, September 12, 2026, AI81 Insights – Thought Leadership. Insights from Carl-Johan Nakamura, Co-Founder & CEO, AI81Works; Lalit Thakur, Chief Data Officer, Nissan Motors, and AI81Works Executive Advisory Board; and Steve Lok, CTO & Chief Scientist, AI81Works.
Most enterprises no longer have an AI technology problem. They have an AI leadership problem. Models are cheap to access, pilots are easy to launch, and nearly every large company now has a portfolio of proofs of concept. Far fewer can point to AI that has changed how thousands of people work, sell, build or decide.
At AI81Works we call the missing ingredient the 81% Rule: AI only creates enterprise value when roughly four in five of the people it touches adopt it at speed. Below that threshold, AI stays a showcase. Above it, AI becomes how the company operates. Crossing that line is not an engineering milestone. It is a leadership one.
AI-Equipped Is Not AI-Ready
An AI-equipped enterprise has bought the licences, stood up a platform and staffed a data science team. An AI-ready enterprise has done something harder. It has redesigned decisions, roles and accountability so that people outside the data team trust AI output and act on it.
The gap between the two shows up in familiar symptoms:
Dozens of pilots, few in production, and no one who owns the portfolio end to end
Business leaders who sponsor AI in steering committees but bypass it in daily decisions
Data that is good enough for a demo but not trusted enough for a regulator, auditor or customer
Value claimed in slide decks but never reconciled with the P&L
None of these is fixed by a better ML model. Each is fixed by putting the right leader in the right seat at the right moment.
The Four Leadership Moves, in Sequence
Enterprises that scale AI tend to add leadership capacity in a deliberate order. Each stage removes the bottleneck that would otherwise stall the next one. Skip a stage and the organization usually circles back to it, at higher cost.
Trust: can we rely on our data? You are stuck here when every AI result triggers a debate about whose numbers are right. The move is a Chief Data AI Officer with a real mandate over data ownership, quality and governance, not only over platforms and tooling.
Proof: does AI create value we can measure? The warning sign is pilots that succeed technically while no business owner will fund the rollout. The move is to embed AI product owners in the business and pair them with a finance partner who signs off on the value delivered.
Adoption: will people actually use it? Tools are live, but usage plateaus soon after launch. The move is a senior leader for AI literacy and change who is accountable for adoption rates, not for the number of deployments.
Stewardship: can we scale it safely? Legal, risk or works councils slow down every new use case. The move is a Responsible AI lead fluent in the EU AI Act, NIST, FDA SaMR, AIDA, Data Act and sector rules, backed by genuine AI fluency at board level.
Two patterns matter more than the exact titles. First, Trust comes before Proof: AI built on contested data rarely survives its first audit. Second, Adoption is a role, not a training budget. Someone senior must own the curve from first login to everyday habit.
Five Traits of Leaders Who Scale AI
Across the executive searches and transformation programmes we support, the leaders who move AI from pilot to practice share a recognisable profile:
Bilingual: they speak both P&L and data pipeline, and translate between the boardroom and the engineering floor without losing meaning in either direction.
Comfortable with probability: they make decisions on 85% confidence, design for error, and know when a human must stay in the loop.
Govern with guardrails, not gates: they set clear rules on data use and risk up front, so teams can move fast inside them instead of queuing for approvals.
Measure adoption, not deployment: their dashboards track who uses AI, how often and with what business result, not how many models went live.
Teach as they lead: they invest personally in AI literacy for managers and frontline teams, because people who understand AI are people who trust it.
Three Perspectives From the Field
Carl-Johan Nakamura: Trust is built in communities, not in policies. “As a Chief Data Officer, the biggest shift I saw was not a technology rollout per se. It came when hundreds of colleagues outside the enterprise data and AI team started treating data as something they owned and could not live without. Digital Councils, internal data cons and AI literacy programmes did more for AI adoption than any platform decision. Governance documents set the rules. Communities make people want to follow them.”
Lalit Thakur: Scale exposes every weak link. “In a global industrial business, an AI use case that works in one plant or one market has to survive dozens of data sources, languages and regulatory regimes. The leaders who succeed design for that from day one: common data definitions, clear ownership across regions, and a single view of value. A pilot proves an idea. Scale proves an operating model.”
Steve Lok: Treat AI like a product with a cost curve. “Every AI capability has its own unit economics: what it costs to run, what it saves or earns, and how both move as usage grows. Too many programmes measure accuracy and ignore economics. The best technology leaders put a product owner and a cost model behind every use case, and are willing to retire what does not pay its way.”
The Board’s Role, and the Traps to Avoid
Boards do not need to understand transformer architectures. They do need to ask the questions that keep AI honest:
Where is AI changing a decision today, and who is accountable for the outcome?
What share of intended users are active, and what is blocking the rest?
Which AI systems fall under high-risk rules, and who signs off on them?
What value has been realised and reconciled by finance, not just forecast?
The same few traps recur across industries:
Hiring a star before fixing the seat: a world-class Chief AI Officer without authority over data or budget becomes an expensive adviser.
Centralising for too long: a central team builds capability, but adoption only happens when ownership moves into business units.
Treating compliance as a final checkpoint: under the EU AI Act, risk classification belongs at the design stage, not at the launch gate.
Confusing activity with progress: pilot counts and model counts flatter. Adoption and realised value tell the truth.
Closing Thoughts
The next wave of AI advantage will not go to the companies with the most models. It will go to those whose leaders build trust in data, prove value in the P&L, drive adoption past the 81% line and steward AI responsibly as it scales. That is a leadership agenda, and it starts with putting the right people in the right seats.
AI81Works helps enterprises do exactly that, through executive search for data and AI leaders, strategy advisory, AI literacy programmes, and fractional and board-level expertise.
About the Authors
Carl-Johan Nakamura, Co-Founder & CEO, AI81Works. Former CDO of the ZEISS Group, first CDO of Oracle Cerner, and CDO/CAO of Siemens Health Services; earlier a Data & AI executive with IBM.
Lalit Thakur, former Chief Data Officer, Nissan Motors, and member of the AI81Works Executive Advisory Board. Former SVP & CDO of ATD, Novant Health and Solera Holdings.
Steve Lok, CTO & Chief Scientist, AI81Works. Former Expert Associate Partner at Bain & Company, Global Head of Martech at The Economist, and Director at MediaMonks.
Media Contact
For more information or to schedule an interview, please contact:
Liliana Trinkle
Email: ltrinkle@ai81works.com
Phone: +1-781-650-6800