Boffin: A clever individual or academic with deep knowledge in a specific field, often seen as socially awkward or consumed by their work.
Sometimes common sense isn’t common. And sometimes, particularly in business, most people agree on an idea and direction, but external validation still helps.
Throughout my career, I have held simultaneous, seemingly divergent views about professional management consultants. Like many other founders, CEOs, and operators, I have held the critical view of management consultants as people who ask you questions and then tell you what you already knew as though it was some profound invention they have created and laid at your feet. I have also generally viewed with disdain people who get paid to tell you what to do but own none of the accountability for actually doing it.
On the other hand, structured frameworks for thinking about strategy and management bring value and clarity. I’ve also long believed that as decision makers, we travel very few paths that haven’t already been well traveled by others. And there is much we can learn from those previous travelers, their successes, and mistakes.
So today, I want to share what some leading management gurus (“boffins”) have to say about finding return on investment in AI transformation.
I have sat as both a participant and observer in many boardroom and executive committee meetings over the years. History does not repeat, but it does rhyme. I think you will recognize this poem.
You are in the boardroom
A slide is shared
A “transformation office”
A “center of excellence”
An AI pilot portfolio
A dashboard of adoption
Progress from the last quarter
The top-line economics have regressed
Not good.
Finance wants the R.
You want the R.
The board wants the R.
Where is the R?
You cannot point at the enterprise P&L and show the return that was supposed to show up. Why not? Did we bring in the right tools? Are they integrated properly? Are people using them?
But the tools are already good enough for the work you asked them to do. People are using the new tools and are reporting gains in personal and team productivity. Local cycle time really did drop. And yet the enterprise still does not see it in revenue, margin, or strategic optionality.
The gap is structural.
Here is the likely error. You are optimizing efficiency where the constraint is not efficiency. This strategic fallacy was fully uncovered in 1984 when Eliyahu Goldratt published “The Goal”, his best-selling introduction to the Theory of Constraints.
Yet here we are in 2026, and most organizations have either never learned or have forgotten the lessons we all learned nearly 40 years ago.
Pouring time, money, and executive attention into non-constraints will not give you the return you seek. The Theory of Constraints describes this pattern. Improvement that does not elevate the constraint does not move the system.
Until you clearly understand and can observe the hidden structure (the constraints, the relationships, the measurement habits, the operating model), the R will keep failing to appear. AI did not fail. You optimized non-constraints in a stationary model while operating in a constrained and non-stationary world.
What the Management Boffins Are Telling Us
“I skate to where the puck is going to be, not where it has been.” - Wayne Gretzky
- Martin Reeves and colleagues, writing in Harvard Business Review in 2023 on radical optionality, describe a world of many hard-to-name futures rather than a few forecastable states.
- Michael Mankins and Mark Gottfredson made a similar point a year earlier: traditional strategy assumes you can lock a plan against a market you describe as a stable entity.
- McKinsey’s State of AI work has repeatedly found that redesigning operating models and workflows, not bolting assistants onto old tasks, is among the strongest correlates of capturing AI value.
- Masha Shunko and Serguei Netessine argued in HBR that firms miss returns when they automate individual tasks instead of evolving the operating models that actually produce business value.
- Thomas Davenport and Thomas Redman put it plainly: AI without an end-to-end process rethink fails to deliver.
Two failure modes show up over and over.
First, optimizing non-bottlenecks. The AI assistant ships and workflow velocity ticks up. But revenue and cycle time don’t move because the binding constraint was never addressed, or it moved elsewhere when something changed.
Second, mistaking a shallow local constraint for the system constraint. Maybe you had a large persistent backlog queue in some internal workflow. You gave it the AI treatment. The backlog disappeared. But the real gate (e.g., belief in the data, authority to act, trust from the customer) never got touched.
The rules, like those of chess, are both simple to understand and hard to master.
- Identify the constraint that actually gates revenue, cycle time, or trust.
- Recognize that systems change and previously “proven” hypotheses evolve to become false and misleading; there are fewer durable truths than you think.
- Redesign the operating model and workflows around those realities.
Human and relational constraints
When operating models evolve, business processes are reengineered, and technology improved, humans and human trust relationships become the binding constraint.
Julie Bedard and coauthors, writing in HBR in 2026 on “AI brain fry,” document what happens when oversight demand outruns cognitive capacity: higher mental effort, fatigue, information overload, decision fatigue, and more errors. The productivity story and the burnout story are the same story told from different seats.
Deloitte’s research on trust in AI found that organizations taking many trust-building actions were more likely to report a larger share of expected gains. Transactional rollouts (licenses first, relationships later) starve the networks that make adoption stick.
Skills are jagged too. The HBS–BCG “jagged technological frontier” experiments showed large gains on tasks inside the frontier and failure outside it, with distinct high-skill use patterns mattering. Hallman and colleagues in HBR note that leaders often measure usage rather than whether AI improves the quality or ambition of the work. Capacity to use AI well is heterogeneous and requires in-practice training.
How to Fix This Situation
You need to change the organization’s posture and mindset toward real evolution. People need to evolve. Optionality with sensing must become the north star.
- Build a constraint model of your business and the ecosystem it exists in.
- If possible, focus on small bets you can pivot as hypotheses evolve.
- Put feedback loops on the constraint you think you are elevating.
- Treat strategy as adaptive under turbulence.
- Evaluate your outcomes.
- Evolve the model.
- Repeat with the shortest cycle time permitted.
Along the way, do these things:
- Watch for oversight overload, and design work for cognitive limits.
- Treat trust as an operating asset.
- Level up judgment and reflection; capability is trainable, and seat licenses will not distribute it for you.
This is not a trivial undertaking, but it is necessary
Don’t look for R where R cannot be found
Some domains require AI to mitigate rising risks. Vanessa Lyon and colleagues at BCG describe cybersecurity as an arms race: offense scaling at machine speed while defense lags. Cyber defense and other relative-advantage contests are not the same as expandable markets. Confusing the two is how you either underfund survival or overclaim that every AI dollar must show a P&L hockey stick.
Recognize when you are metering the wrong thing
Brynjolfsson, Rock, and Syverson’s NBER work on the modern productivity paradox is still the clearest warning: major technology shifts can coexist with weak measured productivity when complementary capital, implementation lags, and mismeasurement hide the gain, and when some high-profit uses dissipate rents in races that mute average welfare. Gartner’s board-facing advice is simpler and still ignored by many. Outcome metrics tied to revenue, cost, cash, and retention beat vanity adoption dashboards.
Match operating-model innovation velocity to the environment. Fund relative constraints differently from expandable ones. Measure constraint throughput and outcome metrics.
Where the R actually is
Stop asking “What’s our AI ROI?” as if R were a property of the AI model.
Ask better questions. Where is the constraint? Is it stationary? Are we measuring its movement? Are we building the trust and human capacity to absorb what AI produces?
The R was missing from the spreadsheet because it was missing from the whole picture. Make the invisible structure visible, elevate the real constraint, and the return has somewhere to land.
References
1. Eliyahu M. Goldratt and Jeff Cox, The Goal: A Process of Ongoing Improvement, North River Press (current edition: Routledge), 1984. https://www.routledge.com/The-Goal-A-Process-of-Ongoing-Improvement/Goldratt-Cox/p/book/9780566086656
2. Martin Reeves, Mihnea Moldoveanu, and Adam Job, “Radical Optionality,” Harvard Business Review, 2023. https://hbr.org/2023/05/radical-optionality
3. Michael Mankins and Mark Gottfredson, “Strategy-Making in Turbulent Times,” Harvard Business Review, 2022. https://hbr.org/2022/09/strategy-making-in-turbulent-times
4. McKinsey & Company (Alex Singla, Alexander Sukharevsky, Lareina Yee, Michael Chui, and Bryce Hall), “The State of AI: How Organizations Are Rewiring to Capture Value,” McKinsey Global Survey, 2025. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-how-organizations-are-rewiring-to-capture-value
5. Masha Shunko and Serguei Netessine, “Stop Automating Old Processes. Design New Ones Instead.,” Harvard Business Review, 2026. https://hbr.org/2026/09/stop-automating-old-processes-design-new-ones-instead
6. Thomas H. Davenport and Thomas C. Redman, “How to Marry Process Management and AI,” Harvard Business Review, 2025. https://hbr.org/2025/01/how-to-marry-process-management-and-ai
7. Jorge Amar, Oana Cheta, Ivan Huang, and Stephen Xu, “From Promising to Productive: Real Results from Gen AI in Services,” McKinsey & Company, 2024. https://www.mckinsey.com/capabilities/operations/our-insights/from-promising-to-productive-real-results-from-gen-ai-in-services
8. Seb Murray (interviewing Paul McDonagh-Smith), “How to Accelerate AI Transformation,” MIT Sloan Ideas Made to Matter, 2026. https://mitsloan.mit.edu/ideas-made-to-matter/how-to-accelerate-ai-transformation
9. Julie Bedard, Matthew Kropp, Megan Hsu, Olivia T. Karaman, Jason Hawes, and Gabriella Rosen Kellerman, “When Using AI Leads to ‘Brain Fry’,” Harvard Business Review, 2026. https://hbr.org/2026/03/when-using-ai-leads-to-brain-fry
10. Diana Kearns-Manolatos, David Levin, Michael Bondar, and Beena Ammanath (Deloitte), “Trust in AI Boosts Benefits,” Deloitte Insights, 2024. https://www.deloitte.com/us/en/insights/topics/risk-management/trust-in-ai.html
11. Fabrizio Dell’Acqua, Edward McFowland III, Ethan Mollick, Hila Lifshitz, Katherine C. Kellogg, Saran Rajendran, Lisa Krayer, François Candelon, and Karim R. Lakhani, “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality,” Organization Science, 2026 (first circulated as HBS Working Paper 24-013, 2023). https://doi.org/10.1287/orsc.2025.21838 (free working-paper version: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4573321)
12. Nick Hallman, Zach Kowaleski, Anu Puvvada, and Jaime J. Schmidt, “What the Best AI Users Do Differently—and How to Level Up All of Your Employees,” Harvard Business Review, 2026. https://hbr.org/2026/03/what-the-best-ai-users-do-differently-and-how-to-level-up-all-of-your-employees
13. Vanessa Lyon, Shoaib Yousuf, and Mahmood Serry, “AI Is Raising the Stakes in Cybersecurity,” Boston Consulting Group, 2025. https://www.bcg.com/publications/2025/ai-raising-stakes-in-cybersecurity
14. Erik Brynjolfsson, Daniel Rock, and Chad Syverson, “Artificial Intelligence and the Modern Productivity Paradox: A Clash of Expectations and Statistics,” NBER Working Paper 24001, 2017. https://www.nber.org/papers/w24001
15. Nate Suda (Gartner), “Prove AI’s Worth to Your CEO and Board With These 5 Metrics,” Gartner, 2026. https://www.gartner.com/en/articles/ai-value-metrics