Business decisions that will outlive today's AI

Artificial intelligence is changing rapidly, but the most important decisions facing CEOs are not about choosing the right AI tool. Theo Dix contends that business leaders should instead focus on the long-term strategic, investment, and organisational choices that will determine whether their companies thrive as AI reshapes business economics.

---

Most chief executives are being asked to make decisions about artificial intelligence without having had a fair chance to understand it. That is not a failure of leadership. It is a consequence of the speed and peculiar character of the technology.

The terminology changes almost as quickly as the products. A system that appears astonishing on Monday can fail at an elementary task on Tuesday. Vendors present polished demonstrations; employees report mixed experiences. Predictions range from modest productivity gains to the disappearance of entire professions. Much of the commentary assumes that business leaders either understand the technology or should urgently become experts in it.

Neither is realistic.

A chief executive cannot follow every new model, benchmark or product feature. Nor should they. Their responsibility is not to predict which AI company will lead in three years, or whether a particular system will master a particular task next March. It is to make decisions today whose consequences may last far longer than the technology's current limitations.

Strategy, investment and organisational choices often take years to implement—and longer to unwind. Each carries assumptions about how work will be performed, what it will cost, what customers will value and where competitive advantage will sit. Every business is built around assumptions about what requires human effort, how much it costs to produce and serve customers, and how quickly the organisation can scale. AI is beginning to change some of those assumptions.

The useful question is therefore not: What can AI do today?

It is: Which decisions are we making today that may look badly designed if activities that currently consume significant time, cost and capacity can be performed very differently?

That moves the discussion away from individual tools and towards the economics of the business.

The wrong lesson from an imperfect machine

It is tempting to draw comfort from AI's failures. Ask a model to complete a long sequence of actions, and it may lose its way. Asking it to build a voice system may cause it to misunderstand a customer. An automated agent may become trapped by an exception that a competent employee would resolve almost without thought.

These weaknesses are real.

They coexist with capabilities that would have appeared implausible only a few years ago. Current systems can analyse large collections of documents, write and test software, interpret images, undertake research, work with spreadsheets and perform sequences of actions across digital systems.

The latest independent evidence does not suggest that capability is reaching a plateau. Stanford's 2026 AI Index[1] reports substantial gains in coding, scientific reasoning, mathematics and computer use. On OSWorld, a benchmark based on real computer tasks, agent success rose from 12 per cent to roughly 66 per cent—still unreliable, but an extraordinary shift in a short period. The progress is uneven: impressive performance in one task can sit beside elementary failure in another. Researchers call this the "jagged frontier".

For a CEO, that frontier creates a trap. A visible failure can look like evidence of a durable limit when it may simply mark the present edge of a rapidly improving system. No sensible board should assume that every weakness will disappear. It would be equally unwise to make a long-lived investment that depends on a particular weakness remaining.

Capability is also only part of the equation. What the best models can do in a controlled setting is not the same as what a company can deploy reliably, safely and economically.

The frontier of model capability is moving rapidly. The frontier of deployable enterprise capability is moving more slowly. The company itself increasingly defines the gap between them: the quality of its data, the architecture of its systems, its controls and most importantly, its capacity to change.

That is precisely why the decisions matter now. The technology may move faster than the strategic and organisational choices companies make in response to it.

Companies routinely make investment and organisational decisions based on the resources required to support growth. AI adds another question: how might the cost, capacity and operating requirements of that work change by the time the investment is fully in place?

That is not the same as asking whether the work can be automated now.

Work rarely changes in one leap. Systems typically progress from assisting with tasks to handling routine cases under supervision, shifting the balance between technology and human judgement.

The strategic consequence arrives well before full automation. An activity can retain human accountability while becoming faster, less expensive and more scalable.

Productivity is not yet profit

Changing how work is performed is not the objective. The question is whether it improves the economics of the business.

AI may accelerate work, increase output or improve quality without creating economic value. Unless management converts those improvements into revenue, pricing, lower cost, avoided investment or a stronger customer proposition, the benefit may never reach EBITDA or cash flow.

The first wave of corporate AI has often been measured through activity: licences issued, employees trained, tools used and hours reportedly saved. These measures describe adoption. They do not establish a return.

Stanford reports that organisational AI adoption reached 88 per cent in 2025, while agent use remained in single digits across most business functions[2]. Adoption is widespread; economic transformation is less mature.

The financial questions are more demanding. Can the company process materially more volume without adding equivalent cost? Can it enter a market without recreating its entire support structure? Can it improve conversion or retention? Can it reduce inventory, avoid additional capacity or shorten the period between investment and revenue?

If so, it has not merely become more efficient. It has changed the economics of growth itself.

Cost creates strategic capacity

The immediate value in many companies will lie in cost. A structurally lower cost base can be a competitive weapon.

It can support sharper pricing, protect margins during a downturn, fund acquisitions, accelerate investment in technology and product development or create the capacity to enter new markets.

Change across a large organisation is rarely quick, nor does it always need to be. Companies continuously hire, replace and redeploy people. As AI changes productivity, the question is whether those investments rebuild yesterday's operating model or help create tomorrow's.

If companies continue adding resources to activities likely to become materially more productive, they risk locking yesterday's economics into tomorrow's business. Using natural organisational change and normal investment cycles to redirect resources can shift those economics over time.

The transition may be gradual. The ambition should not be.

Boldness is not spending more on AI. It is making capital and operating decisions that assume the business itself will change.

CEOs that act early can improve the cost trajectory before pressure forces the issue, creating investment capacity while competitors continue to carry both old and new operating models.

The top line will move too

Cost is the more visible opportunity because it begins with work the company already performs. Growth is harder—and potentially more consequential.

AI is increasing what companies can produce and deliver with a given level of resources. But greater supply does not create equivalent demand.

If competitors can produce more too, markets may become crowded with competent, inexpensive output. Customers will have more choice, stronger negotiating power and less reason to pay a premium for products or services that have become easier to reproduce.

This creates a strategic problem that is easy to miss. AI does not simply reduce the cost of producing an answer. It can also reduce the price customers are willing to pay for one.

The exposure will vary by industry, but the underlying test is straightforward: how much of our revenue depends on something remaining scarce?

That scarcity may lie in specialist capability, access to information, production capacity, distribution, responsiveness or the customer's inability to obtain the same result elsewhere. If AI weakens it, protecting today's margin will not protect tomorrow's revenue.

Growth therefore cannot be reduced to adding AI features or making the existing proposition more personalised. In many markets, those improvements will rapidly become expected. Producing more will not, by itself, create growth.

The company must use its new capabilities to improve something the customer genuinely values: the quality or speed of a decision, the level of risk, the convenience of the service or accountability for the result. It also needs something that gives the proposition substance: customer access, trust, proprietary information, specialist judgement, distribution or the ability to execute.

Some companies will use it to produce the same work more cheaply in markets where prices are also falling. Others will use AI to solve a more valuable customer problem at economics that were not previously possible.

The first may become more efficient without becoming more valuable. The second has a chance to grow.

That choice will not emerge from a list of AI use cases. It requires management to decide which parts of the existing business will remain attractive, where value is likely to move and where the company can still earn superior returns.

This is why AI cannot sit apart from corporate strategy. It should influence where the company competes, what it continues to own, where it invests, how it prices and which acquisitions remain attractive.

An acquisition whose value depends on scarce capabilities, high production costs or structural information advantages may look different if AI weakens those barriers. Conversely, proprietary data, trusted customer access, effective distribution or deep integration into customer workflows may become more valuable because AI expands what can be done with them.

The relevant question is whether AI changes the return on the next unit of capital committed to the business—not whether it can complete a task.

Data, governance, skills and adoption matter, but they are not the strategy

A model is only as useful as the information, access and authority a company gives it. Many organisations still face fragmented data, disconnected systems and unclear ownership of information. These challenges should be addressed from the outset, but not allowed to delay deployment.

Governance must be equally practical. Management needs to decide what AI may recommend, what it may execute and where human judgement remains necessary. The requirements are different for a system drafting an email than one approving a payment or changing a customer record.

Skills matter too. Employees need to know how to divide work between people and systems, challenge outputs and recognise where judgement remains important. Managers must redesign work rather than simply add AI to existing processes.

Data, governance and skills are not prerequisites for experimentation. They are prerequisites for transformation. Companies that postpone them may find their ability to scale AI constrained just as the technology becomes more capable.

Adoption remains the bridge between capability and value. A tool used occasionally by individuals may improve personal productivity. Economic impact appears only when better ways of working become embedded and translate into revenue, lower cost, stronger cash generation or avoided investment.

The decision before certainty

The CEO does not need a better demonstration. Use what is already good enough, concentrating deployment where value can be measured through revenue, cost, cash, capacity or customer outcomes.

Start with a small number of economically important parts of the business. Ask how their cost, capacity and customer value might change as AI capability improves—and what would prevent the company from capturing that value.

Then apply an AI sensitivity test to major investments:

If AI capability improves materially during the life of this decision, would we still make it in the same form?

Where the answer is uncertain, preserve options. Avoid unnecessary rigidity. Separate the elements of an investment that are durable from those that may be overtaken.

The CEO does not need to know which model will win. The more important task is to identify where the economics of the business still depend on work remaining slow, expensive or difficult to scale.

Some of those assumptions will survive. Others will not. The chief executive cannot know exactly where the boundary will move. But investments being made today will still be in place when it does.

That is why waiting for certainty is not the cautious choice. It is a decision to build around constraints that may already be disappearing.

Source

[1] Stanford Institute for Human-Centered Artificial Intelligence (HAI) (2026) AI Index Report 2026. Stanford University.

[2] Stanford Institute for Human-Centered Artificial Intelligence (HAI) (2026) AI Index Report 2026. Stanford University.

---