2025, the year AI has to show results
- 1Time saved is not money saved: it only becomes a financial result when the freed capacity is redeployed or removed.
- 2Every AI use case needs an owner, a measured baseline and a written 90-day rule to scale, correct or stop.
- 3Value, full cost and risk (FADP, EU AI Act) are the three tests to pass before a budget is renewed.
- 4AI is not always the right answer: some gains come first from better organization and processes.
The year of proof
2023 was the year of discovery and 2024 the year of pilots. Nearly every company we work with has opened licences, trained a few volunteers and produced some convincing demonstrations. In 2025 the question changes. It no longer comes from IT or the innovation team but from the CFO and the board: what has this changed in the income statement?
Our view is simple. This demand is healthy, and it plays to the strengths of mid-sized companies. They cannot fund scattered experiments indefinitely, but they have something large groups rarely do: short decision cycles and little distance between those who decide and those who execute. The task is to use that advantage.
What the pilots produced, and what they did not
Pilots produced learning, which is valuable. They rarely produced proof. In most cases we see, three things are missing: an owner accountable for a quantified result, a baseline measured before deployment, and a written rule that says when to stop and when to scale.
Without these, a pilot can neither fail nor succeed. It simply continues. In our article on moving generative AI from experimentation to value creation we explained why this grey zone is expensive: it consumes attention, budget and credibility without ever producing a decision.
A worked example: time saved is not money saved
Take a company of 300 people, with deliberately simple assumptions. It rolls out an office assistant to 150 employees at roughly CHF 30 per user per month. The licence comes to CHF 54,000 a year. Assume, for the sake of the example, an equal amount for training, integration and management time: the total is around CHF 108,000 a year.
With a fully loaded hourly cost in the order of CHF 100, break-even sits at about 1,080 hours saved per year, or roughly ten minutes per user per week. On paper the bar is low, which is why the business cases look so persuasive.
The trap lies elsewhere. Ten minutes a week per person produces no saving at all as long as nothing changes in the organization. The time dissolves into the working day. It becomes a result in only four situations: overtime or temporary staff that disappear, a hire that is avoided, a shorter cycle that speeds up cash collection or sales conversion, or capacity moved to an activity that earns money. In each case someone has to decide, and the decision is managerial, not technological.
That is why we say that in 2025 the issue is no longer to prove that AI saves time, which is often true, but to show what the company does with that time.
Three tests before renewing a budget
The value test
Does the use case act on an identifiable line of the income statement, or on an operational indicator tied to one: cost of processing an order, time to answer a tender, rework rate, overdue receivables? If the answer is "general productivity", the use case is not ready to be weighed against others.
The full-cost test
Licences are the visible part. Full cost includes data preparation, integration with existing tools, change management, human review of outputs and maintenance. In many cases these items exceed the licence. A use case that only holds up on visible cost will not survive scrutiny.
The risk test
In Switzerland, the revised Federal Act on Data Protection (FADP) has been in force since 1 September 2023: transparency on processing, impact assessments in certain cases, care over transfers outside Switzerland. In Europe, the AI Act entered into force on 1 August 2024 and applies in phases: prohibitions from February 2025, obligations for general-purpose AI models from August 2025, and most obligations from August 2026. A Swiss SME that sells into the EU or relies on European suppliers cannot ignore it. US companies face a different shape of the same issue: without a single federal law, it is customer contracts and a patchwork of state rules that demand documentation.
Our reading is pragmatic. For most mid-sized companies, risk is managed through a simple inventory of uses, a classification by sensitivity and clear rules on which data may or may not be entered into a tool. This is a matter of weeks, not a large-corporate compliance programme.
What we recommend to leaders
Name one owner per use case. Not a technical project lead but a business owner who accepts being judged on the result. If nobody wants to carry the target, that is a signal in itself.
Measure before deploying. Two weeks measuring the current situation are worth more than six months of impressions. Without a starting point, no gain will ever be credible in front of a CFO.
Set a 90-day decision point. At the end, only three outcomes: scale, correct with a new short deadline, or stop. Stopping a use case is normal management, not failure.
Manage AI as a portfolio. Use cases compete for the same scarce resources: team attention, data and budget. Our use case on project portfolio rebalancing applies directly: rank by value, effort and risk, then accept that not everything will be done. This arbitration also needs a regular steering rhythm, as we describe in the three-year strategic plan that survives its first quarter.
Accept that AI is not always the answer. Sometimes an invoicing or planning process loses time because it is poorly designed, with three redundant approvals or data keyed in twice. Automating that disorder makes it faster, not better. In such cases the first step is common sense and organization, and our Operations & processes expertise is a better starting point than a new tool.
A discipline rather than a promise
None of this is spectacular, which is rather the point. The company that draws lasting advantage from AI in 2025 will probably not be the one with the most pilots, but the one that has tied a small number of use cases to decisions, owners and figures. This is the core of our approach to strategy and value creation: start from the strategy, deploy with method, then steer the results.
If you want to know which of your current use cases would withstand a financial review, our diagnostic gives you a clear first reading, in a few conversations, of what to scale, correct or stop.
A personalised first reading in three minutes.