Most companies deployed AI to reduce costs. A smaller group deployed it to understand their customers better and grow revenue. In 2026, the gap between those two decisions is becoming visible in the numbers - and increasingly hard to close.
The slide deck had fourteen pages. I counted. Every one of them showed some version of the same story: tickets deflected, headcount reduced, response times improved, cost per resolution down. The leadership team presenting it was visibly proud. The AI deployment had delivered exactly what it promised.
I asked one question when they finished. Where's the revenue slide?
The room went quiet for a moment. Someone offered that the cost savings were effectively revenue. Which is technically true and practically misleading. Because while this company had spent eighteen months optimising AI for operational efficiency, two of their closest competitors had been doing something different with the same tooling. They were using it to understand their customers better - to know which accounts were ready to expand before anyone asked, to identify which onboarding paths led to faster time-to-value, to surface the signals that told their CS teams where to show up and when. Their NRR was climbing. This company's was flat.
Same technology. Completely different decisions about what problem to solve with it.
The Reflex That's Costing Revenue
A Harvard Business Review article published this week - "Companies Are Using AI for Efficiency. They Should Use It to Grow" by Benartzi, Long and Puntoni - opens with a line that stopped me: "Ask a roomful of senior executives what AI can do for their business, and the answers will probably cluster around the same themes: lower costs, smaller headcount, faster processes. Efficiency, efficiency, efficiency - it's an almost universal reflex. It's also a badly misguided one."
That framing is precise. It's a reflex. Not a strategy. Not a considered deployment decision. A default - shaped by the fact that efficiency use cases are easier to measure, faster to demonstrate, and simpler to present to a board that wants to see margin improvement.
The problem with reflexes is that they feel like decisions while producing worse outcomes than actual decisions would. Companies that deployed AI for efficiency have the receipts - lower costs, faster processes, cleaner operations. What many of them don't have is growth. And in 2026, the window for correcting that direction is starting to close, because the competitors who chose differently are building compounding advantages that don't reverse easily.
What Growth-Oriented AI Actually Looks Like
The distinction isn't about which AI tools you buy. It's about the question you're trying to answer when you deploy them.
Efficiency-oriented AI answers: how do we do the same thing with less? It is an optimisation problem. It has a ceiling - you can only reduce cost so far before you hit structural limits.
Growth-oriented AI answers: what do we now know about our customers that we didn't know before, and what does that make possible? It is an expansion problem. It compounds - the more you know, the better you can act, and the better you act, the more your customers trust you, which gives you more data, which makes you smarter.
At Zendesk, I watched companies deploy the same support infrastructure in fundamentally different ways. Some used it to process more tickets faster. Others used it to identify patterns in customer questions that their product teams didn't know existed - signals of friction, confusion, unmet need. The second group didn't just improve their support metrics. They fed those signals back into product and CS, shortened their time-to-value for new customers, and reduced the category of problem that was generating tickets in the first place. They solved the upstream issue. The efficiency-first group kept optimising the downstream response.
The delta in customer retention between those two groups, over two to three years, was significant. Not because one had better AI. Because one had a better question.
The SaaS Dimension Nobody Is Pricing In
A second piece of research worth reading: "AI's Impact on SaaS Will Be Uneven. Here's What Leaders Need to Know" by Christopher Stanton of Harvard Business School, published in HBR on May 27. The central argument is that SaaS tools which survive the AI era will be the ones that deepen customer value - not those that simply digitised workflows.
Stanton's framing has a direct implication for CS and marketing leaders: the vendors whose tools you're currently paying for are facing the same efficiency-versus-growth question internally. The ones that oriented their AI toward making your team more capable - better signals, sharper customer intelligence, faster identification of expansion opportunity - will compound in value as your AI maturity grows. The ones that used AI primarily to deflect your support tickets and reduce their own cost to serve you are providing a commodity that AI can replicate more cheaply every quarter.
This means the vendor evaluation question has changed. It isn't just "does this tool do what we need?" It's "is this vendor's AI making us smarter about our customers, or just cheaper to operate?" Those are different products. They warrant different relationships and different renewal conversations.
At Adobe, working inside one of the platforms making this exact bet, I saw the internal version of this tension. The features that drove the deepest customer retention were never the ones that automated existing workflows. They were the ones that surfaced new insight - that showed a customer something about their own performance or audience that they couldn't have seen without the platform. That's the product that becomes embedded. That's the renewal that never becomes a negotiation.
Where the Gap Is Forming Right Now
The companies pulling ahead in 2026 are doing something specific in their CS and marketing operations. It isn't exotic. But it requires a deliberate reorientation away from the efficiency default.
What the efficiency-first companies are doing:
- Using AI to reduce CS headcount and automate tier-1 interactions
- Deploying AI-generated content to maintain publishing volume at lower cost
- Measuring AI success by cost savings, ticket deflection, and headcount ratios
- Treating AI as a cost-centre tool managed by operations
What the growth-oriented companies are doing:
- Using AI to expand what each CSM knows about their accounts - health signals, stakeholder changes, expansion readiness
- Deploying AI to identify which prospects are in-market before they raise their hand, and which customers are ready to grow before they ask
- Measuring AI success by NRR movement, expansion pipeline, and lead quality
- Treating AI as a revenue tool owned jointly by CS, marketing, and commercial leadership
The operational difference between these two groups is significant. The commercial difference, compounding over twelve to eighteen months, is becoming structural.
The Decision That's Still Reversible - For Now
There is a window here, and it is not permanently open.
The companies that have been using AI for efficiency for eighteen to twenty-four months have optimised their operations around that assumption. Their teams are structured for it. Their metrics are built around it. Their vendor relationships reflect it. Reversing that orientation isn't a technology decision - it's an organisational one, which is harder and slower.
The companies that haven't yet made a deliberate choice about which direction they're pointing their AI investment are actually in a better position than they might think. The tooling is largely the same. The question - efficiency or growth - is what determines the outcome. And that question is still answerable.
The slide deck I described at the start of this piece didn't need different technology. It needed a different brief. One that started not with "how do we reduce cost" but with "what would we be able to do for our customers if we knew twice as much about them as we do now?"
That's a different meeting. It produces a different AI deployment. And in 2026, it produces meaningfully different revenue.
