A conversation I had with the CEO of a mid-sized Indian financial services firm earlier this year. He had read the first two essays in this series. He agreed with most of it. The mindset principles, the allocation discipline, the J-curve, the last-mile problem — all of it landed. We are doing some of this. We could do more.
Then he asked the question that, in different forms, every CEO is asking right now. But how does any of this help me with what is really keeping me up at night, which is whether the entire structure of my industry is being rewritten while I am perfecting the basics?
He was right to ask. Because this third essay is not about the basics. It is about the question underneath the basics. The question of whether the rules of competition in your industry are about to change so fundamentally that the firm that wins the next decade will not be the firm that perfected the last one.
I have come to believe — and the evidence of the last three years has only sharpened my view — that this question is not a marketing slogan. It is the actual situation. Every classical principle of technology investment is being rewritten by AI. Some of the rewriting is incremental. Most of it is structural. And the senior leaders who treat this as an extension of the last technology cycle, rather than a discontinuity, will lose ground in ways they will only notice when it is too late to recover.
Eight principles, in this final essay, on what changes at the strategic level. Not the spending. Not the execution. The structure of what your firm is for, and how that gets answered when the underlying technology of cognition itself becomes a utility.
Technology strategy is business strategy
The first thing to understand at the strategic level is that the artificial separation between technology strategy and business strategy is over.
For most of the last forty years, this separation made some sense. Business strategy was about markets, products, competitors, customers. Technology strategy was about systems, infrastructure, vendors, integration. The CEO did the first. The CIO did the second. They met occasionally at the executive committee. The conversation was largely sequential — here is what the business needs to do, and here is how technology will support it.
That model is now broken. In every industry I observe, the most consequential business strategy decisions of the next decade are technology decisions. And the most consequential technology decisions are business strategy decisions. They cannot be separated, because the underlying causal chain runs in both directions.
The bank that decides to move to a real-time, AI-driven personalisation engine is not making a technology decision. It is making a decision about what kind of bank it will be — what its customer relationships look like, what its branch network is for, what its talent profile needs to become. The hospital chain that decides to deploy AI-augmented diagnostics is not making a procurement choice. It is making a decision about what its doctors do, what its hospitals charge for, and where competitive advantage will sit in five years.
The CEO who delegates these decisions to the CIO is delegating business strategy to a function whose remit was never designed to carry it. This is not a criticism of CIOs — most of the ones I work with are genuinely sophisticated. It is a criticism of the org-chart inheritance that treats their role as it was treated in 2010.
The discipline that works here is unglamorous but necessary. The CEO must own the technology strategy directly. Not the implementation. The strategy. Which platforms will be foundational. Which AI investments will be transformational. Which capabilities will be built, bought, or partnered for. These are not delegate-able decisions anymore. The CIO is the implementer. The CEO is the strategist. The boards I sit on that have made this shift explicit are the ones whose firms are widening their technology lead. The ones that have not are the ones whose CEOs are perpetually surprised by how quickly competitors are moving.
If there is one structural shift this entire essay is about, it is this. Technology strategy is no longer a sub-strategy. It is the strategy. And the firms that internalise this first will be the firms that compound first.
Platforms versus products
Closely related is the shift from product thinking to platform thinking at the strategic level.
In a product-thinking firm, the strategic asset is the offering. The bank's mortgage product. The insurer's policy. The retailer's catalogue. Investment goes into improving the product. Competitive advantage comes from the product being slightly better, slightly cheaper, slightly faster than the competitor's product.
In a platform-thinking firm, the strategic asset is the underlying capability that produces many products. The data platform that serves every customer interaction. The AI infrastructure that runs every decision. The integration layer that lets new products be assembled in weeks rather than years. Investment goes into the platform. The products are downstream consequences. Competitive advantage comes from the platform being meaningfully more capable than the competitor's — which lets the firm produce products faster, cheaper, and more relevantly than the competitor can.
The shift sounds abstract. Its consequences are not. A platform-thinking firm can launch ten new products in the time it takes a product-thinking firm to launch one. Not because the products are simpler. Because the underlying capability is reusable. Each new product, in a platform firm, requires assembly rather than construction.
The Indian BFSI industry is, by global standards, still mostly product-thinking. Each new product is built more or less from scratch. The data is fragmented. The integrations are bespoke. The systems were designed in eras where each line of business was a separate empire. The firms that have started the shift to platform thinking — and there are now a small number of them — are visibly accelerating. They launch products faster. They cross-sell more effectively. They respond to regulatory change with weeks of work rather than quarters. The product-thinking firms watch them and assume the difference is execution speed. It is not. It is structure.
The senior leader's question on this principle is — if I had to launch a new product line tomorrow, what fraction of my technology stack would I be reusing, and what fraction would I be rebuilding? In a platform firm, the answer is eighty percent reuse, twenty percent rebuild. In a product firm, the ratio is reversed. And the gap widens with every product launched.
The transition from product to platform is one of the most expensive technology transformations a firm can undertake. It is also, in my view, the single highest-leverage strategic move available to most firms in the next five years. The CEOs who are willing to fund it are buying themselves a strategic option that the CEOs who are not, by definition, will not have.
Ecosystems and the leverage of partners
The third strategic shift is about ecosystems — the recognition that no firm, no matter how large or sophisticated, can build everything it needs alone, and that the ability to participate in ecosystems is itself a capability.
The classical view of competitive advantage is that the firm wins by being self-sufficient. The integrated firm. The firm that owns the value chain end-to-end. This view is being slowly overturned, and AI is accelerating the overturning.
The reason is straightforward. The capabilities required to compete in any modern industry are now too numerous, too specialised, and changing too quickly for any single firm to build them all. The firm that tries ends up with mediocre versions of everything. The firm that participates well in ecosystems ends up with access to best-in-class versions of most things — and concentrates its own building on the small number of capabilities that are genuinely core.
In Indian banking, the most visible expression of this shift is the UPI ecosystem. The banks that won market share in the digital payments era were not the ones that built the best in-house payments tech. They were the ones that integrated most fluently with the public infrastructure that NPCI built. Building the rails was someone else's job. Building on them was the bank's. The firms that tried to build their own walled gardens around payments lost most of the volume to the firms that did not.
The same pattern is now playing out in AI. The firms that try to build everything themselves — their own foundation models, their own training infrastructure, their own MLOps stacks — are mostly losing ground to the firms that build carefully on top of what the hyperscalers and foundation model providers offer. The strategic advantage is no longer in building the technology. It is in deploying it well, with proprietary data and process design that is genuinely differentiated.
The senior leader's discipline on this principle is to map the firm's capability stack and ask, for each layer, should we own this, partner on this, or consume it as a utility? The answers are different than they were five years ago. They will be different again in five years. The firm that is willing to revisit those answers — and to give up things it once owned in favour of things it can now consume — is the firm that stays competitive. The firm that defends its existing capability footprint out of pride or inertia is, in most cases, defending the wrong things.
AI as cost reduction, AI as reinvention
I want to spend a moment on what I think is the most important strategic distinction in the entire AI conversation. There are two ways to think about AI investment, and the difference between them determines whether the firm produces incremental returns or compounding ones.
The first is AI as cost reduction. Apply AI to existing processes to do them faster, cheaper, with fewer people. Underwriting AI that processes applications faster. Customer service AI that handles more queries per agent. Compliance AI that reduces manual review work. These are real returns. They are typically modest. They do not change what the firm is. They just make the existing firm more efficient.
The second is AI as reinvention. Use AI to do things the firm could not do at all before. New products that depend on AI capability. New customer experiences that are not available through any human-staffed equivalent. New business models that require AI-scale data processing to function. These are the returns that compound. They are also harder to defend in a year-one business case, because they involve creating value where none existed.
Most firms — and most CIOs — are currently optimising for the first kind of AI investment. The second is harder, riskier, and slower to show measurable returns. But it is also the only kind of AI investment that produces durable competitive advantage. The cost-reduction kind gets matched by every competitor within eighteen months. The reinvention kind, if executed well, becomes a moat.
The same firms that have been thoughtful about platform thinking and ecosystem participation tend to be the ones investing in reinvention AI. They are using the cost-reduction work to fund the reinvention work — taking the savings from the first and channelling them into capability investments that will not pay back for three to five years. This is exactly the right financial structure for the moment we are in. Cost-reduction AI funds reinvention AI. Reinvention AI builds the next decade's competitive position.
The CEO's question is which kind of investment dominates the AI portfolio. If it is mostly cost-reduction work, the firm is on the path to becoming a cheaper version of itself. If reinvention work is genuinely funded — if there are at least one or two AI initiatives whose business case is capability rather than savings — the firm is on the path to becoming something more than itself. Both are valid choices. They produce very different ten-year outcomes.
Why proprietary data is the moat
Of all the conversations about AI in the last three years, the one that has been most poorly conducted is the one about data.
The standard version goes — AI is the moat. The firm that has the best AI wins. This is wrong. AI capability is, increasingly, a commodity. Every serious foundation model is now within a few months of every other serious foundation model. The frontier moves fast, but the gap between the best and the second-best, at any moment, is narrow and shrinking.
The actual moat is not AI. It is data. Specifically, proprietary data that no other firm has, applied to a problem that data uniquely solves.
The reason is straightforward. AI models trained on public data are available to everyone. The firm that fine-tunes a public model on its own customer transactions, claims history, underwriting records, or operational data has access to a capability that no competitor can replicate. The model itself is generic. The data is not. And the value created by the combination is the firm's, not the platform provider's.
This has profound implications for strategic decision-making. The firms that win the AI era will not be the ones that adopted AI fastest. They will be the ones that, over the last decade, accumulated the cleanest, deepest, most unified proprietary data — and now have something to fine-tune the AI on that nobody else does. The firms that were sloppy about data — that fragmented it across systems, let it decay in silos, treated it as a byproduct rather than as an asset — are entering the AI era without a moat.
I have watched this distinction become visible over the last eighteen months. Two firms in the same line of business, with roughly comparable AI investment budgets, are producing dramatically different results. The first has spent the last decade unifying its customer data into a single, clean, structured platform. The second has roughly the same data, scattered across thirty systems with inconsistent identifiers and uneven quality. The first firm's AI initiatives are accelerating quarter by quarter. The second firm's are stalling. The difference is not the AI. The difference is the substrate the AI is being trained on.
The senior leader's discipline here is to take a hard look at the firm's data estate and ask — is this an asset I can compound, or is it a liability I will have to reconcile before AI can produce real returns? For most firms, honestly assessed, the answer is the second. And the work of moving from liability to asset takes years, not quarters. Which means the firms that started this work three to five years ago are now positioned to extract AI returns that the firms that did not start cannot match — even with identical AI investment.
The data moat is real. It is also slow to build, painful to fund, and impossible to acquire quickly when you suddenly need it. Which is why, for most firms, the right time to start building it was a decade ago. The second-best time is now.
The real economics of human-in-the-loop
Closely related is what I have come to think of as the most under-discussed principle in AI economics. Almost every successful AI system in production today is a human-in-the-loop system. And the economics of human-in-the-loop are very different from the economics most firms have planned for.
The naive view of AI ROI assumes that the AI replaces the human. The agent. The underwriter. The analyst. The reality, in almost every meaningful production deployment I have seen, is that the AI augments the human — and the value comes from the augmented pair, not the replacement.
This changes the economics substantially. The cost is not "AI tooling minus salary saved." The cost is "AI tooling plus the same salary, mostly." The benefit is not "headcount reduction." The benefit is "throughput per head, plus quality of decision, plus capability to take on work that was previously infeasible."
The firms that have understood this are the ones that are extracting genuine value from AI. The firms that have insisted on the replacement narrative — cut headcount, deploy AI, save money — are the ones whose AI initiatives are quietly underperforming. You can deploy AI as a replacement and capture small, fragile savings. You can deploy AI as augmentation and capture large, durable productivity gains. The first looks better in the year-one business case. The second produces an actual return.
The reason augmentation works better is straightforward. AI handles the repetitive, structured, well-bounded part of the work. Humans handle the ambiguous, judgement-heavy, accountability-bearing part. Together they produce better outcomes than either alone. The challenge is that designing the augmented workflow is genuinely difficult — it requires rethinking what the human does, what the AI does, where the handoffs happen, how exceptions are managed. Most firms skip this work. They either deploy the AI poorly bolted onto existing processes (which produces little value) or replace humans entirely (which produces low-quality outcomes for the cases the AI cannot handle, which is more than they expected).
The CEO's discipline on this principle is to insist on augmentation design as a first-class workstream in every major AI deployment. Not as an afterthought. Not as change management. As the core of the project. The firms that do this end up with augmented workflows that are genuinely more productive than any pure-human or pure-AI alternative. The firms that skip it end up with AI deployments that produce theatrical demos and disappointing P&L impact.
How big tech actually thinks
The firms that have spent the most money, most thoughtfully, on technology over the last fifteen years are the hyperscalers — Alphabet, Amazon, Meta, Microsoft. They are not just heavy spenders. They are unusually disciplined heavy spenders. And the discipline they apply is genuinely different from the discipline most enterprise firms apply.
I want to extract three lessons from how they think, because I believe each one is portable to firms that are nothing like them.
First — they think in terms of irreversible bets. The hyperscalers have an explicit framework for distinguishing between "two-way door" decisions (reversible, can be unwound) and "one-way door" decisions (structural, hard to undo). They make two-way door decisions fast, with low ceremony, often by single leaders. They make one-way door decisions slowly, deliberately, with extensive review. Most enterprise firms have the inverse pattern. They torture two-way door decisions through committees and approve one-way door decisions in the rush of the next budget cycle. This is exactly the wrong allocation of decision discipline to decision type.
Second — they invest in capabilities, not in projects. The standard enterprise budget is structured around projects. Each project has a sponsor, a budget, a timeline, a benefit. The hyperscalers structure their investment around capabilities — things the firm wants to be able to do, persistently, that may take many years and many sub-projects to build. AWS is a capability. Search ranking is a capability. Recommendation systems are capabilities. The capability gets funded for as long as the firm wants to have it. The sub-projects come and go. This produces a fundamentally different kind of compounding than the project-by-project enterprise model.
Third — they treat infrastructure as competitive advantage. Most enterprise firms treat infrastructure as a cost centre. Keep it running. Keep it cheap. The hyperscalers treat infrastructure as a strategic asset — sometimes their most strategic asset. The result is that the gap between hyperscaler-grade infrastructure and enterprise-grade infrastructure is now an order of magnitude. And that gap is one of the reasons hyperscaler-built AI products are pulling away from enterprise-built AI products in capability, even when the underlying models are similar.
The portable lesson for senior leaders is not to copy the hyperscalers. They have advantages — capital, talent, scale — that no enterprise firm can match. The lesson is to adopt the discipline. Distinguish between two-way and one-way doors, and apply the right ceremony to each. Fund capabilities, not just projects. Treat the parts of infrastructure that produce strategic advantage as strategic. Most enterprise firms can adopt all three of these, at no extra cost. Almost none of them do.
The CEO's new mandate
The final principle in this series is about the CEO role itself, because everything I have written across these three essays converges to a single conclusion. The CEO's job is changing.
For most of the last forty years, the CEO of a non-technology firm could treat technology as one function among many. Important, sometimes. Strategic, occasionally. But manageable through the standard delegation pattern — I hire a competent CIO, I review the budget, I approve the major projects, I stay informed. The CEO's actual job was elsewhere — markets, customers, capital allocation, talent.
That model no longer works. Not because the old job has gone away, but because a new responsibility has been added — and is, in many industries, becoming the dominant one. The CEO is now responsible for the firm's structural posture toward a wave of technology change that, every classical principle of investment notwithstanding, is rewriting the rules faster than any prior wave.
The new responsibilities, as I have come to understand them, are these.
Owning the technology strategy directly. Not the implementation. The strategic posture. Build, buy, partner. Capability priorities. Reinvention versus cost-reduction. These are CEO decisions now, not delegations.
Funding the things that compound. Foundations over features. Platforms over products. Data infrastructure over flashy applications. Reinvention AI over cost-reduction AI. The capital allocation has to favour the slower, harder, less-glamorous work that produces compounding returns. Defending those choices against the constant political pressure to fund the prettier demo is the CEO's job — because nobody else in the firm has the authority to do it.
Building the talent multiplier. Hiring genuinely good technical leaders. Pairing platforms with the people who can extract value from them. Protecting small teams of high-quality contributors from being absorbed into average ones. Most firms get this wrong because the cost of doing it right is friction with the existing org. The CEO is the only person who can absorb that friction.
Re-examining the strategy at AI cadence, not quarterly cadence. AI capability is doubling roughly every six months at the moment. The CEO who looks at strategic positioning quarterly is, by definition, falling behind. The firms that are pulling away are the ones whose CEOs are spending serious thinking time on this every two months, not every twelve.
I do not think most CEOs are prepared for this. Not because they are not capable. Because their training was for a different era. The CEOs who are doing this work well were, almost without exception, surprised by how much of their time it now takes. The ones who have not yet started are, almost without exception, the ones whose firms are quietly losing position to competitors they thought were further behind.
The mandate is real. It is also unevenly distributed. Some CEOs have absorbed it and reorganised their attention around it. Most have not. The ten-year separation between the firms run by the first kind of CEO and the firms run by the second is going to be one of the largest divergences in market performance of our generation.
The convergence
Three essays. Twenty-four principles. The convergence of all of them, if I had to put it in one paragraph, is this. The classical model of technology investment — distinct from business strategy, delegated to a competent CIO, evaluated through standard ROI lenses, executed through discrete projects — is breaking down. What is replacing it is a model in which technology strategy is business strategy, the CEO owns it directly, the discipline is upstream and structural rather than downstream and tactical, and the firms that execute this transition well are pulling decisively ahead of the firms that do not.
If you have read all three essays — and I am genuinely grateful if you have — the practical question is what you do with the principles. I would suggest one move per quarter. Pick the one principle, from the twenty-four, that you most clearly know your firm gets wrong. Address it. Properly. With the leadership stamina to see it through the J-curve and the last-mile problem. Then pick the next one.
Twenty-four principles, addressed at one per quarter, would take six years. Which is roughly the right time horizon for a serious technology repositioning of a non-trivial firm. I do not know of any shortcut. The firms that are pulling ahead are the ones doing this work, principle by principle, year by year. The firms that are falling behind are the ones still looking for the magic AI investment that will make all the structural work unnecessary.
There is no such investment. The work is the work.
That is what I have come to believe, after watching enough boards make enough decisions over enough years to develop the intuition for it. Technology is the work. The work is leadership. And leadership is harder than any of us were trained for.
Good luck.