New to Beyond Strategy? Subscribe here for free.
While 78% of organizations now report using AI (Stanford HAI), MIT NANDA studied more than 300 AI initiatives, 95% showed no measurable P&L impact. This gap can be referred to as “value leakage” - AI value created vs. AI value captured.
Where the value leaks
AI initiatives can produce a positive effect . Yet, what reaches the P&L as durable profit may be a lot less. The value difference generated by AI often leaks to competitors, customers, and vendors:
To competitors. Most AI capabilities run on the same foundation models. An automated customer service based on a standard model can therefore be easily replicated by competition.
To customers. Once a capability becomes industry-standard (commoditization), AI savings are usually lost to customers due to price competition.
To vendors. Through token cost, created value is flowing straight to suppliers. For instance, Uber spent its entire budget on AI coding tools in 4 months (Fortune). For any initiative, the figure that matters is net of this cost.
56% of companies have seen neither higher revenues nor lower costs from AI, and only 12% report both, because for most the business value has never materialized.
Why the same pilot may give opposite results
Built on proprietary data, embedded in a redesigned workflow, and tied to a position competitors cannot easily copy, an initiative keeps most of what it creates.
The leverage sits in the workflow and the data. The share of tasks AI can meaningfully improve rises from about 15% to roughly half once the workflow around it is truly rebuilt rather than layered on top. Externally built or partnered tools succeed about twice as often as internal builds (MIT NANDA), because they arrive with integration and iteration already solved.
The first move is not choosing a tool or a pilot. It is finding where value survives the next 3 years, after stakeholders have each taken their share, and what it would take to keep it. Isolate the few levers that are both large and defensible, fund the near-term ones that defend the core and generate cash, then use that cash for the longer bets that cannot pay back inside one planning cycle. Spending becomes investment only once that logic is in place.
The full framework, from the six-layer diagnostic to an aggregated value case can be found in our new paper, The AI Value Case.
3 questions before your next AI investment
What will still be yours in 3 years? For your largest AI initiative, estimate the value it creates, then subtract what competitors, customers, and vendors are each likely to take.
Are you measuring inputs or outputs? Efficiency and productivity are inputs. Revenue, margin, and retained value are outputs.
Which levers can a competitor not buy next quarter? Rank your initiatives by how fast a well-funded rival could replicate them with the same vendor.
Weekly Picks
Stanford HAI, AI Index 2025: the paradox at the level of the whole economy, record investment and adoption alongside a collapse in per-token cost. The single best place to calibrate the macro backdrop.
When Every Company Can Use the Same AI Models, Context Becomes a Competitive Advantage (HBR): the defensibility argument in one line. When the model is a commodity, the moat is the proprietary context you feed it.
AI's $600B Question, David Cahn / Sequoia: the same value-capture question at industry scale. Capex is racing ahead of the revenue needed to justify it, which is the leakage problem in macro form.
Stefan Benndorf
Founding Partner scaleon
scaleon
Build for growth.
The questions this newsletter raises are the same ones our clients bring to us. Which AI initiatives deserve a second round of investment? Where does P&L accountability for digital transformation actually sit? How does technology deployment translate into measurable business value?
scaleon works with CEOs and investors of digital businesses on exactly these questions, in growth strategy, operational management, and transaction preparation.
If anything in this edition is worth a conversation, we'd welcome a direct message.