Start With Business Strategy

Between February and April 2025, IBM’s Institute for Business Value and Oxford Economics surveyed 2,000 chief executives in 33 countries and 24 industries. One question asked, in effect, whether fear was setting the technology budget. Before reading on, guess what share of those CEOs agreed that the risk of falling behind drives them to invest in some technologies before they have a clear understanding of the value they bring.

The answer was 64 percent. Half of the same CEOs said the pace of recent investments had left their organization with disconnected, piecemeal technology¹.

These are self-reported figures, and the instinct behind them is understandable. Nobody wants to explain to a board why a rival moved first. But the numbers describe a sequence that runs backwards. The technology arrives, the budget follows the anxiety, and only afterwards does someone ask what business result the spending was meant to change. The pieces do not fit together because nothing chose them to fit.

The core idea

AI should serve the business strategy. It should not become a strategy of its own. The starting point for an enterprise AI strategy is the business: its goals, its customers, its economics, the way it intends to win and the constraints that hold it back. As What Is an Enterprise AI Strategy? showed, every initiative should trace back to that strategy. This chapter is about the step before the trace: making sure the business question is asked first, so there is something to trace back to.

The two paths can involve exactly the same tools. What differs is the question that opens the discussion. “Which AI technologies should we adopt?” sounds responsible and produces a catalogue. “What are we trying to achieve, and where are we stuck?” produces a short list, and it gives every later decision, from data to talent to platforms, a reason to exist. Starting with the business does not mean ignoring technology. What the technology can and cannot do is an essential constraint on the choice. It informs the decision; it does not define the objective.

The strategy decides what AI is worth

A strong reason the business must come first is that the same AI capability is worth very different amounts to different strategies. Michael Porter described the classic routes to competitive advantage: be the lowest-cost producer in a market, or differentiate in ways customers will pay for, either broadly or within a focused niche². Michael Treacy and Fred Wiersema offered a variant that many executives find easier to apply. Market leaders, they argued, choose one of three value disciplines, operational excellence, customer intimacy or product leadership, excel at it, and only keep pace on the other two³.

The ratings in the table are illustrative judgments made for this chapter, not measurements, and that is the point. A grocery chain that competes on price should care most about forecasting and automation, because they attack the cost base it competes on. A private bank that competes on knowing its clients should care most about service that feels personal, and could reasonably leave its back office on yesterday’s tools. A drug maker that competes on new products should care most about anything that shortens the path from idea to tested candidate. None of these companies is wrong to ignore what the others prioritize. They are being consistent.

Strategy also sets the urgency. Is the market shifting? Are customer expectations rising? Is a competitor changing the cost structure of the industry? When the answers are yes, AI may matter strategically because the economics are moving, not because the technology is new. That is a business reason, and it can be argued, tested and funded like any other.

Find the constraint the strategy depends on

Knowing how the business intends to win narrows the field. The next step narrows it further, to the place where AI would change the result. In 1984 Eliyahu Goldratt set out what became the theory of constraints: the output of any system is limited by its tightest constraint, so an improvement anywhere else does not raise the output. His sharpest line was that “an hour saved at a non-bottleneck is a mirage”⁴. Richard Rumelt makes the same demand of strategy itself, which must begin with a diagnosis of the critical obstacle⁵.

Suppose, as an illustration, a distributor’s strategy is to win on next-day delivery, and orders miss the van because the warehouse cannot pick them fast enough. An AI tool that writes product descriptions in seconds is impressive, and it may even save hours in marketing. It does nothing for the promise the strategy depends on. AI that predicts tomorrow’s orders and re-slots the fastest-moving stock nearest the packing benches would go straight at the constraint, if its forecasts proved accurate enough in that warehouse; that is a capability to test, not a result to assume. Both projects would report savings. Only one moves the number that matters.

Most organizations have several constraints, one per strategic priority, and they move once one is relieved. The discipline is the same in each case: name the constraint before naming the tool.

Name the problem before the tool

The quickest test of whether an AI proposal started in the right place is the sentence that describes it. “We need an AI assistant for finance” names a tool looking for a home. An illustrative alternative, “Month-end close takes nine working days because reconciliations are done by hand across four systems, and that delays every decision that depends on the numbers” names a problem, an owner and a measure. Only once the problem is that clear should AI enter the discussion, and only if it is the best way to solve it.

If a process follows a stable rule, conventional automation is usually enough: when an invoice is below a set amount and the supplier is approved, route it for payment. AI becomes attractive when the problem involves language, ambiguity, images, prediction or inputs that vary too much for fixed rules. And when the real cause is a broken process, missing data or untrained people, the answer is to fix that first. AI laid over a broken process gives you a faster broken process. Proposals that skip this step tend to die later and more expensively. Gartner predicted in 2024 that at least 30 percent of generative AI projects would be abandoned after the proof of concept by the end of 2025, and named unclear business value among the reasons⁶.

A good physician works the same way. She does not arrive with a promising new drug and look for patients to give it to. She examines the patient, makes a diagnosis and then chooses the treatment, which may be the new drug, an old one, physiotherapy or a change of diet. The same drug that cures one patient harms another, which is why nobody trusts a doctor who prescribes before examining. A technology-first AI program is a doctor writing prescriptions in the waiting room.

Map capabilities by importance and AI potential

To see where the constraints sit across a whole business, it helps to map the business into capabilities rather than departments: winning customers, serving them, developing products, running operations, managing the supply chain, finance, risk and people. For each capability, four questions do most of the work. How well does it perform today? How important is it to the strategy? Where is it constrained? Could AI materially change its economics?

The top-right quadrant earns attention first. The top-left still matters a great deal, but the answer is probably a process change, an investment or a hire rather than AI. The bottom-right is the dangerous one: impressive demonstrations in capabilities the strategy does not depend on. It is also where technology-first programs tend to start, because that is where the technology looks best. The map is a strategic filter, not a ranking. Feasibility, time to value and risk come in when individual opportunities are compared, which is the work of Finding Strategic AI Opportunities.

Competitor moves are information, not orders

Few things test a business-first discipline like a rival’s announcement. AI may lower an industry’s costs, raise what customers expect and reduce the barriers to entering a market. Those are possibilities more than settled results in most industries, but they are reason enough to take a competitor’s move seriously. But Porter’s warning applies with full force. Operational effectiveness, doing the same things better than rivals, is not strategy; when every firm copies the same best practices, they converge, and nobody gains a lasting edge⁷.

Matching is sometimes right, when a capability has become something customers simply expect. More often the better response is to differentiate where the rival is weak, to redesign the process so the copy is unnecessary, or to hold a different position on purpose because your customers chose you for something else. Whether AI can produce an advantage that lasts is the subject of AI and Competitive Advantage. The point here is narrower: the rival’s move is an input to your strategy, not a substitute for it.

Real priorities receive real resources

A strategy that starts with the business must also end with the business’s resources. Here many organizations are slow. Stephen Hall, Dan Lovallo and Reinier Musters studied more than 1,600 US companies from 1990 to 2005. For a third of the businesses, the capital received in a given year was almost exactly what they had received the year before. The companies that reallocated most, shifting an average of 56 percent of their capital across business units over the 15 years, earned on average 30 percent higher annual returns to shareholders than those that reallocated least⁸. The 0.99 is a correlation between one year’s capital and the next for each business; the 30 percent compares the top third of companies by reallocation with the bottom third. It is an association, not proof that reallocation alone caused the returns.

The study predates generative AI, but the lesson translates directly. An AI priority that receives no budget, no owner, no team, no data and no executive sponsor exists only on paper, however prominently it appears in the strategy document. The simplest test of a stated priority is to ask where its resources are. As What Is an Enterprise AI Strategy? noted, the AI leaders in BCG’s 2024 survey pursued about half as many opportunities as their peers⁹. Concentration is how a short list becomes a funded one.

Business strategy is not fixed, and the relationship runs both ways. Marco Iansiti and Karim Lakhani argue, from cases such as Ant Financial, that AI can reshape the operating model of a firm and, with it, the economics of whole industries¹⁰. When that happens, AI becomes an input to the next version of the business strategy, not just a tool for executing the current one. The order of the questions still holds. What AI might make possible is evidence for the strategy discussion; it is not a reason to skip it.

Story: a coach operator that answered the wrong question

What follows is an illustrative composite, not a single documented company. It is built from a pattern that recurs across industries and told here through an intercity coach operator, a long-distance bus company, so read it as a post-mortem of a decision rather than as evidence.

The operator’s strategy was clear and had worked for years: the lowest fares on a simple, no-frills service. Riders chose it for price and for departures that ran on time, and the board measured it on cost per seat-kilometer and punctuality. Then two larger rivals announced AI-driven personalized fares, with tailored offers and discounts predicted for each traveler. The board asked a fair question: where is ours?

Within weeks a program was launched to match them. It received the best data team and most of the year’s AI budget. A year later the offers engine was live, and technically it worked. But the operator’s riders were mostly price-focused people who already booked the cheapest seat, and many of the discounts went to trips that would have been booked anyway. The operator was paying for revenue it already had. Meanwhile the constraint its strategy actually depended on, coaches breaking down on the road and the hired replacement vehicles each breakdown forced, was still listed as a top priority. It had no budget, no owner and no data team.

The post-mortem asked what the rivals’ move had actually told the operator. It had treated the move as a mandate. The strategy pointed somewhere else: a low-cost coach operator wins by keeping its fleet on the road cheaply and reliably, which points AI toward predicting breakdowns before they happen and planning workshop time around the timetable. Four questions, asked in the first month, would have changed the decision. Which strategic priority does this serve? Which constraint does it relieve? Who owns it? Where are the resources? Nobody in the story was careless. The operator simply started with the rivals’ technology instead of its own strategy.

What this means for leaders

Open every AI discussion with the business question, not the technology one. Say how the business intends to win and which constraint holds each priority back, then ask whether AI changes that constraint. Expect some answers to be “not AI”, and treat them as success, not as a lack of ambition. Read competitor moves as new facts about the market and decide your response against your own strategy. And check that the priorities on paper have the budget, people and data to match, because the budget is where a strategy shows whether it is real.

Check yourself

1. If competitors use AI in an area, we must use it there too.MythFact
2. The same AI capability can be vital under one strategy and minor under another.MythFact
3. Starting with business strategy holds AI innovation back.MythFact
4. An AI project that saves time anywhere in a process improves the business result.MythFact
5. An AI use case with a high expected return is automatically strategic.MythFact
6. Most companies move their capital toward new priorities quickly.MythFact

Reflection: what the first question was

What comes next

Starting with the business tells you where AI belongs. It does not tell you how far to go. A low-cost coach operator might use AI to trim the cost of breakdowns, or to redesign how its whole route network is run. The next chapter, Defining AI Ambition, takes up that question: how ambitious the organization should be with AI, from targeted efficiency gains to a fundamental change in how the business works.