Adoption Is Not a Strategy.

The honest answer to the question in the title is: sometimes, but not in the way most boards are being sold it. AI is rarely the missing link in a growth strategy. It is far more often the accelerant applied to a growth strategy that already exists, and where that strategy is vague, AI spending reliably produces activity, dashboards, and no measurable change in the numbers that matter.

The adoption-to-impact gap

The most useful dataset on this is McKinsey's global survey of nearly 2,000 executives. Adoption is essentially universal: 88% report their organisation regularly uses AI in at least one business function, up from 78% a year earlier. Two-thirds use it in more than one function.

Then the curve flattens hard. Nearly two-thirds of organisations have not begun scaling AI across the enterprise at all. Only 39% attribute any enterprise-level EBIT impact to AI, and most of those put it below 5%. The population reporting both meaningful EBIT contribution and significant value, the study's definition of high performers, is about 6% of respondents.

On agents, the picture is earlier still: 23% are scaling an agentic system somewhere in the business and another 39% are experimenting, but in no individual business function do more than 10% report scaled agent deployment. Most of those scaling are doing so in one or two functions.

So the question for a chief executive in 2026 is not whether to use AI. That is settled, and using it confers no advantage because everyone else is too. The question is what the 6% are doing that the other 94% are not.

Why pilots stall short of the P&L

The survey's relative-weights analysis is blunt about the differentiator. The practice most strongly associated with real value is fundamental redesign of the workflow, not tool count, not budget, not model choice. High performers are close to three times as likely to have rebuilt processes rather than inserted AI into them.

The mechanism is easy to see once you look at a specific process. Suppose a quoting process takes eleven days: two days to gather requirements, four waiting on engineering input, one to build the quote, three in an approval chain, one to send. Deploy a tool that drafts the quote in minutes and you have removed most of one day out of eleven. The customer's experience is unchanged. Nothing reaches the income statement. The pilot is nonetheless recorded as a success, because the tool did exactly what it promised.

Redesign asks a different question: given that a compliant draft can be produced in minutes from source data, what should this process look like? Perhaps engineering input becomes a pre-computed rules library. Perhaps approvals below a threshold become automatic with sampled audit. Perhaps requirements are captured directly in the customer conversation. Now eleven days becomes two, and that is a competitive difference customers feel and the P&L records.

Redesign is harder because it touches roles, authority, controls and incentives, which is exactly why senior ownership matters. In the same research, high performers were three times more likely to strongly agree that senior leaders demonstrate ownership of AI initiatives. That is not a motivational point. Only the executive team can change who approves what.

Efficiency has a floor; growth compounds

Roughly 80% of organisations set efficiency as an objective of their AI work. The companies seeing the most value set growth or innovation objectives alongside it, and respondents pursuing growth or innovation were more likely to report enterprise-level benefits including profitability, revenue growth, competitive differentiation and market share movement.

The logic is straightforward. Efficiency is bounded: a cost can be removed once, the saving is one-off, and competitors are removing the same cost with the same tools on roughly the same timeline. Within eighteen months it is table stakes, not advantage. Growth applications behave differently. Better segmentation improves targeting, which improves conversion data, which improves segmentation. Faster product iteration produces more customer feedback, which produces faster iteration. These compound.

The pattern in the data is consistent with that: reported cost benefits cluster in software engineering, manufacturing and IT, while reported revenue benefits cluster in marketing and sales, strategy and corporate finance, and product and service development.

For most middle-market companies the answer is not to choose. It is to run efficiency work to fund the growth work, and to make clear internally that the growth work is the point, because efficiency programmes are easier to measure and will otherwise quietly absorb the entire agenda.

Building AI into the roadmap rather than beside it

Most AI plans we are asked to review are organised by technology: a list of tools, functions and pilot dates. Very few are organised by the constraint the business is actually trying to relieve. That ordering is the problem, and it is why so many of these plans cannot be evaluated: there is no stated business outcome to evaluate them against.

A growth and transformation strategy that takes AI seriously starts from the growth constraint and works backwards:

1. Name the two or three constraints genuinely limiting growth. Sales capacity. Quote turnaround. Churn in a specific segment. Time to launch in a new geography. If the leadership team cannot agree this list in a session, no AI programme will fix the underlying ambiguity.
2. Test each for an information or throughput bottleneck. Some constraints are informational: too slow to know something, too expensive to analyse it, too manual to do it at volume. Those are where AI has leverage. Others are capital, regulatory or relationship constraints, and AI will not move them.
3. Redesign the end-to-end workflow, not the task. Assume the informational bottleneck is gone, then rebuild handoffs, approvals and roles around that assumption. This is where the value is, and where the organisational work is.
4. Define the KPI and the baseline before deployment. Cycle time, conversion rate, cost to serve, revenue per rep. If you cannot state the metric and today's number, you will not be able to say afterwards whether it worked, and the programme will be defended on anecdote.
5. Set the verification rule up front. Roughly half of organisations using AI report at least one negative consequence, with inaccuracy the most common. Decide which outputs require human sign-off, who signs, and what evidence is retained. High performers are markedly more likely to have defined this.
6. Sequence for evidence, then fund. One redesigned workflow that demonstrably moves a real number earns the right to the next three. Twelve simultaneous pilots produce twelve inconclusive readouts and an exhausted organisation.

Market expansion, where the leverage is genuinely new

One area deserves separate treatment because the economics really have changed. Entering a new geography, vertical or customer segment has always carried a fixed research and preparation cost: market sizing, competitive mapping, regulatory review, channel identification, localisation, pricing calibration. That fixed cost is what has historically made smaller expansion opportunities uneconomic to pursue properly, not the opportunity itself.

A large share of that work is information processing, and it is now materially cheaper and faster. The practical consequence is that the minimum viable size of an addressable expansion opportunity has fallen. Segments a middle-market company previously had to ignore because the analysis alone would have consumed a quarter of the upside can now be assessed seriously.

Two cautions, both learned expensively. First, cheaper analysis is not better judgment: a synthesised market map still has to be tested against people who actually operate in that market, and the failure mode is a confident, well-formatted document with no primary validation behind it. Second, the constraint usually moves rather than disappears. Once analysis is no longer the bottleneck, the bottleneck becomes distribution, local relationships, or management attention, and a market expansion strategy built on analysis alone will stall at exactly that point.

What to do in the next ninety days

Concretely, for a US middle-market leadership team:

• Inventory what is already running. Most companies have more AI in use than the executive team knows about, arriving through individual tools and vendor features. You cannot build a strategy over an unknown base, and this inventory doubles as the answer to a diligence question buyers now ask routinely.
• Kill the pilots that cannot name a metric. Not because they are unpopular, but because they consume the scarcest resource in the company, which is management attention.
• Pick one workflow and redesign it properly. Choose one that touches revenue rather than an internal convenience. Rebuild it end to end, measure it against a stated baseline, and be honest about the result.
• Put a named executive on it. Not a committee, not the IT function alone. The redesign requires authority over process and roles.
• Set the governance now. An approved-tool list, a data-handling rule, a human-verification standard for customer-facing and financial outputs. It takes an afternoon and it is far cheaper than the first incident.

The companies that will be difficult to compete with in 2028 are not the ones with the largest AI budget in 2026. They are the ones that used this period to rebuild a handful of processes around what became possible, while everyone else was running pilots. The technology is broadly available to all of them. The strategic clarity about where to point it is not.

Questions we get asked

Why do most AI initiatives never show up in financial results?

Because tools are layered onto workflows designed around their absence. Time saved inside a process that still has the same handoffs and approvals gets reabsorbed by the process. The factor most strongly associated with real impact in the survey data is fundamental workflow redesign, which is organisational work rather than technical work.

Should AI be a cost play or a growth play?

Both, but growth is where the compounding is. Efficiency savings are one-off and competitors capture the same ones on the same timeline. Growth applications feed themselves. Use efficiency gains to fund the growth work, and say so internally, or the easier-to-measure efficiency agenda will take the whole programme.

Where should a middle-market company start?

With the growth constraint, not the technology. Name the two or three things genuinely limiting growth, identify which have an information or throughput bottleneck, redesign that workflow end to end, and define the metric and baseline before you deploy anything. One process that moves a real number beats a dozen pilots.