AI Alone Is Not a Cost-Cutting Strategy

The pitch is always the same. Buy the tool, point it at support or the back office, and wait for costs to fall. It sounds neat. It often fails.

An MIT report from 2025 found that about 95% of enterprise gen-AI pilots had little or no clear impact on profit and loss. Only about 5% drove real gains. In the same window, S&P Global Market Intelligence said 42% of firms dropped most of their AI work in 2025, up from 17% the year before. It also said nearly half of all proofs of concept were stopped before production. Cost was one of the top reasons.

That is not a case against AI. It is a case against treating AI as a way to spend less on its own. Many AI implementation failures are really AI change management problems. The firms losing money were not the ones that skipped the tech. They were the ones that bought it and hoped it would act like a plan.

Why the AI savings story keeps breaking down

• The savings assume the AI can do the work. That is where most AI implementation failures start. Klarna is the best-known case. The company said its AI helper did the work of 700 customer service agents — later, it said the number was closer to 800. A year later it began hiring again. Its CEO said the lesson was simple: too much focus on cost-cutting had led to worse service. The math looked great until customers had to live with it. When the tool cannot fully do the job, the savings were never real. They were a bill that showed up later as churn and cleanup.

• The costs do not vanish. They move. Cutting a team is easy to see and quick to do. The new costs are quieter: connecting the tool to real systems, watching it, fixing mistakes, and staffing the hard cases it cannot close. A support queue does not get simpler when you automate it. The easy tickets go to the bot, and the tickets left for people are the hard, high-stakes, emotional ones that take longer and need your best staff. You do not remove people and keep the same result. You just move the money around, and a failed rollout adds a sunk cost on top.

• Cutting the team removes what the AI needs. This is the part the slide leaves out. The MIT team found that pilots mostly failed not because the models were weak, but because of a learning gap. The tools were not tied to real workflows, and the companies had not changed around them. The know-how that makes support work — how this product breaks, what this customer group needs, where the edge cases hide — lives in people and process. Cut that to fund the tool, and you have removed the thing that made the tool worth having. That is the real AI change management problem.

Where the AI cost savings actually come from

The firms getting real AI cost savings are doing something less flashy and more hard than buying software. They are redesigning how the work is done.

McKinsey's 2025 research found that 88% of companies now use AI in at least one function, but only 39% report any EBIT impact at all — and most of those say less than 5% of profit comes from it. The firms that see better results are nearly three times as likely to have fully redesigned their workflows instead of bolting AI onto the old ones. The value is not in the tool. It is in the work change the tool makes possible.

That is change management for AI adoption in practice. AI is a multiplier, not a swap. Used on a messy process, it makes more mess faster. Put it on a support team with weak escalation paths, thin docs, and undertrained staff, and you get automated confusion at scale. Put the same tool on a well-run team — good SOPs, skilled people, clean handoffs — and it can raise output in a real way. The difference is not the software. It is everything around it.

The same rule applies in compliance work. Using AI for regulatory change management or finding the best AI change management tools for compliance teams still depends on clean process, clear owners, and human review. The tool can help. It cannot fix a broken flow by itself.

The honest framing

None of this means do not use AI. It means AI is one tool inside a plan, not the plan itself. Gartner expects more than 40% of agentic AI projects to be canceled by the end of 2027, mostly because the value is unclear and the costs keep rising. Most of those cuts will point back to the same root cause: someone bought a feature and hoped it would act like a strategy.

The real path to lower cost is less shiny and more durable. Build a team that is trained and accountable, write down the process, and be clear about what a machine should do and what a person should own. Then use AI where it cuts friction for real — repeat volume, drafting, routing — and let it make good people faster instead of trying to replace them. That is human-in-the-loop AI in practice. That is how the cost curve bends and stays bent.

Cutting the team and hoping the software covers the gap is not a cost plan. It is a bet that the cheapest version of your customer experience is good enough. The firms that made that bet are the ones quietly hiring back their people.

That is the part we care about. Efficiency is not the lack of people. It is a well-run operation, with the right work in the right hands, and tech 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)

Q&A

Question: Is the article saying companies should avoid using AI?

Short answer: No. The article says AI can help, but only when it is part of a broader plan. The problem is not AI itself. The problem is the idea that buying an AI tool will cut costs on its own, without workflow redesign, trained people, clear process, and oversight.

Question: Why do AI cost-saving efforts often fail to deliver real savings?

Short answer: They often fail because the forecast savings assume the AI can fully replace human work, while the real rollout adds new costs. Companies still need to connect the tool, watch its output, fix errors, handle edge cases, and support customers when automation falls short. In many cases, costs are not removed; they are moved into new areas.

Question: What does the Klarna example show?

Short answer: The Klarna example shows the risk of treating AI as a direct swap for human support. The company first promoted large staffing-like savings from its AI helper, but later began rehiring people after service quality fell. The lesson is that apparent savings can vanish if churn and cleanup costs are ignored.

Question: Where do meaningful AI-driven savings come from?

Short answer: Meaningful savings come from redesigning how work gets done. The article says firms with stronger returns are more likely to rebuild workflows around AI instead of adding AI to old steps. AI works best when it removes friction from repeat tasks, routing, drafting, and throughput, while humans keep ownership of complex or high-stakes work.

Question: What should companies do before using AI to cut costs?

Short answer: They should first build a well-run operation: train accountable people, document the process, create strong SOPs, define escalation paths, and decide which tasks machines should handle versus which need human judgment. That is change management for AI adoption. Once that base is in place, AI can make good teams faster instead of trying to cover gaps left by cuts.