AI Alone Is Not a Cost-Cutting Strategy

The pitch is always the same. Buy the tool, point it at your support queue or your back office, and watch a line item shrink. It's a clean story, and it fits neatly on a slide. The problem is that it keeps not working.

An MIT report released in 2025 found that about 95% of enterprise generative-AI pilots delivered little to no measurable impact on profit and loss — only around 5% drove real acceleration. In the same window, S&P Global Market Intelligence reported that 42% of companies abandoned most of their AI initiatives in 2025, up from just 17% the year before, and that nearly half of all proof-of-concepts were scrapped before they ever reached production. Cost was one of the top reasons cited.

That's not an argument against AI. It's an argument against treating AI, by itself, as a way to spend less. Because the companies losing money here aren't the ones who avoided the technology. They're the ones who bought it expecting it to be the strategy.

Why "AI equals savings" keeps breaking down

The savings assume the AI actually does the job. That's the assumption that fails most often, and the most public example is Klarna. The company famously said its AI assistant did the work of 700 customer service agents — a number it later put closer to 800. A year on, it started rehiring people. Its CEO put the lesson plainly: an overemphasis on cost-cutting had led to lower-quality service. The math looked incredible right up until customers had to live inside it. When the tool can't fully do the work, the "savings" were never real — they were a bill that arrived later, in churn and cleanup.

The costs don't disappear. They move. Cutting a team is visible and immediate. The costs that replace it are quieter: integrating the tool into real systems, monitoring it, correcting what it gets wrong, and staffing the hard cases it can't close. A support queue doesn't get simpler when you automate it — the easy tickets go to the bot, and the ones left for people are the complex, high-stakes, emotionally loaded ones that take longer and need your best staff. You don't get to remove the humans and keep the outcomes. You just rearrange where the money goes, and a failed rollout adds a sunk cost on top.

Cutting the team removes what the AI needs to be good. This is the part the slide never mentions. The MIT researchers found that pilots mostly failed not because the models were weak, but because of a "learning gap" — the tools weren't connected to real workflows, and the organizations hadn't adapted around them. The knowledge that makes support good — how this product actually breaks, what this customer segment actually needs, where the edge cases hide — lives in your people and your processes. Lay that off to fund the tool, and you've removed the very thing that would have made the tool worth having.

Where the savings actually come from

The companies getting real financial value from AI are doing something more boring and more difficult than buying software. They're redesigning how the work happens.

McKinsey's 2025 research found that while 88% of organizations now use AI in at least one function, only 39% report any EBIT impact at all — and most of those attribute less than 5% of profit to it. The organizations that do see meaningful returns share a specific trait: they're nearly three times as likely to have fundamentally redesigned their workflows rather than bolting AI onto the old ones. The value isn't in the tool. It's in the operational change the tool makes possible.

That reframes the whole exercise. AI is a multiplier, not a substitute — and a multiplier applied to a messy process just produces more mess, faster. Point it at a support operation with no clear escalation paths, thin documentation, and undertrained staff, and you get automated confusion at scale. Point the same tool at a well-run operation — good SOPs, capable people, clean handoffs — and it genuinely lifts throughput. The difference in outcome isn't the software. It's everything around it.

The honest framing

None of this means "don't use AI." It means AI is a tool inside a strategy, not a strategy on its own. Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027, largely over unclear value and rising costs — and most of those cancellations will trace back to the same root cause. Someone bought a capability and expected it to behave like a plan.

The real path to lower costs is less exciting and more durable. Build an operation that works: people who are trained and accountable, processes that are documented, and clear rules about what a machine should handle versus what a person must own. Then apply AI where it genuinely removes friction — the repetitive volume, the drafting, the routing — and let it make good people faster instead of trying to replace them. That's how the cost curve actually bends, and stays bent.

Cutting the team and hoping the software covers the gap isn't a cost-cutting strategy. It's a bet that the cheapest version of your customer experience is good enough — and the companies that made that bet are the ones quietly hiring their people back.

That's the part we care about. Efficiency isn't the absence of people. It's a well-run operation, with the right work in the right hands, and technology used where it earns its place.

Sources: Fortune / MIT NANDA — 95% of enterprise gen-AI pilots show no measurable P&L impact (Aug 2025) · CIO Dive / S&P Global — 42% of companies abandoned most AI initiatives in 2025 (Mar 2025) · McKinsey — The State of AI 2025 · Gartner — Over 40% of agentic AI projects canceled by 2027 · Forbes — Klarna reverses AI push, rehires human support (May 2025)