AI ROI is moving from productivity to growth.

The first wave of AI ROI was about saving time. The second wave is something different — and the gap between companies still optimising for efficiency and companies pursuing growth is starting to show up in revenue numbers.

Google Cloud's most recent enterprise study found that 56% of executives say generative AI has led to business growth, and among those reporting revenue increases, more than half estimate gains of 6–10%. PwC's 2026 analysis is more direct: AI leaders are roughly twice as likely as the rest of the market to be using AI to identify and pursue growth opportunities, not just productivity gains. McKinsey's high performers, separately, set growth and innovation objectives alongside efficiency, not in place of it.

The companies whose AI investments look unimpressive on a board slide a year in are usually the ones that scoped exclusively for efficiency.

Why efficiency was the obvious first move

Efficiency is the easy AI story. The before-and-after is simple. The investment case is a labour-cost line item. The metrics fit on a single slide: hours saved, tickets per analyst, lines of code per engineer, drafts produced per writer.

That clarity is what made efficiency the default opening move. It is also what limits it as a strategy.

Why efficiency alone is a limited strategy

There are two ceilings on an efficiency-only AI programme.

The first is that you can only save the same hour once. After the obvious wins — drafting, summarising, triaging, looking things up — the marginal productivity gain on each new deployment gets smaller. By the second or third initiative, you are running diminishing returns on the same shape of metric.

The second ceiling is harder to spot but more decisive. Efficiency by itself does not change what the business is capable of. It speeds up things the business already does. A faster version of an existing process is still the same process. If the strategic question is "what new markets, products, or customer experiences are we capable of," the answer does not come from efficiency.

From cost reduction to business reinvention

The companies pulling ahead are using AI to do things they could not do before, not just the same things faster. A few patterns where this shows up:

• Better lead qualification. Sales pipelines used to be filtered by hand based on signals an SDR could collect. Now AI systems read across enriched data sources to score and prioritise — at a granularity that produces meaningfully better conversion, not just faster qualification of the same leads.
• New customer experiences. Personalised onboarding flows, conversational product discovery, dynamic service tiers — capabilities that were not commercially viable when each one required a human in the loop are now table-stakes for some category leaders.
• Faster product development. Engineering organisations that have redesigned around AI-assisted code generation and testing are shipping more often, not just typing faster. That shows up as faster product cycles, which shows up as competitive position.
• Personalised service at scale. What used to be reserved for high-value accounts — proactive outreach, custom analysis, recommendations — is now economically viable for the mid-market.
• Improved decision workflows. The cycle of "wait for the monthly business review, identify a problem, commission analysis, decide" is collapsing for the companies that built AI into how decisions get made. Better questions get asked, faster, in places they were not asked before.

These are revenue-side stories, not cost-side stories. They are also harder to scope, because the metric of success is "what did we build that we could not build before" rather than "how many hours did we save."

How to tell if an AI use case is strategic or merely convenient

A useful test: if the AI deployment succeeded perfectly and was running for two years, what would change about the business?

If the honest answer is "we would save N hours per week," that is an efficiency play. Worth doing, often, but not strategic.

If the honest answer is "we would be able to serve a customer segment we cannot serve today," or "we would ship a product capability we cannot ship today," or "we would make decisions on a cadence the competition cannot match," that is a growth play.

The first kind compounds linearly. The second kind compounds the business itself.

Use AI to change what the business can do, not just how fast people type.