AI returns follow ambition, not activity.
PwC’s AI Performance Study, published on 13 April 2026, finds that a fifth of the companies surveyed capture three-quarters of the returns executives attribute to artificial intelligence. The Swiss National Bank’s own company talks suggest most firms in Switzerland are still in the phase that comes before returns.
Insights bi îngilîzî tên weşandin.
PwC published its AI Performance Study on 13 April 2026. One line has travelled widely: 20% of the 1,217 companies surveyed capture 74% of the returns their executives attribute to artificial intelligence. The more useful part is the explanation attached to it, because it is not a budget.
The respondents were senior executives at director level and above, in 25 sectors across Africa, Asia, Europe, the Middle East, North America and South America; 91% of the companies were publicly listed and 76% reported revenues of US$1 billion or more. Fieldwork ran in October and November 2025. A sample of that shape says little directly about a Swiss machine-tool firm; what it establishes is which management practices travel with measured returns.
Ranked by management practice, the top fifth stands on a different step
PwC measured what it calls AI-driven performance: the proportion of revenue and of efficiency or cost gains respondents attribute to AI, adjusted so each company is measured against its sector median. It then scored each company across 60 areas of AI management and investment practice, grouped into nine factors — six foundations, three measures of use — forming an AI fitness index.
Ranked by that index, the five quintiles record average AI-driven performance scores of −2.6, 0.3, 4.6, 6.4 and 15.7 percentage points relative to the median. The first four are separated by a few points each; the fifth by more than nine: the top 20% average 15.7 against 2.2 for the remaining 80% — the 7.2-fold difference the study reports.
Two limits belong with that figure: these are executives’ own assessments, not audited results, and PwC describes its index as significantly linked to performance — an association, not a demonstrated cause.
Efficiency is where everyone lands; the distance opens on what AI is aimed at
Asked how far their AI portfolio had improved a list of outcomes, 65% of the strongest performers report improved employee productivity — and so do 41% of other companies. That is the narrowest gap in the table: productivity is not what separates the two groups but what they have in common.
The wide gaps sit elsewhere: 59% of the leading group report that AI has transformed their business model against 23% of the rest, 56% against 20% report more decisions taken without human intervention, and 52% against 21% report faster speed to market for new products and services.
The same asymmetry appears in what AI is pointed at. Half of the leading group use AI to a large or very large extent to collaborate outside their own sector, against 17% of the others; 60% against 33% use it to sense emerging value pools. PwC identifies growth captured from industry convergence as the single strongest of the nine factors influencing AI-driven financial performance, ahead of efficiency gains. PwC’s global chief AI officer, Joe Atkinson, put it this way: “The leaders stand out because they point AI at growth, not just cost reduction, and back that ambition with the foundations that make AI scalable and reliable.”
Money is not absent: leading companies invest 2.5 times as much of their revenue in AI, about 5% of revenue in software, banking, and media and entertainment. But the mechanism the study describes is a conversion rate rather than a spending level. Companies with strong foundations that increase their AI use record nearly double the improvement seen by those with weaker foundations.
The Swiss National Bank’s company talks show firms in the phase that precedes returns
Each quarter the SNB’s delegates for regional economic relations meet company managements across Switzerland and the Bank aggregates what it hears. In the first quarter of 2026, 241 talks were held between 14 January and 3 March. Many companies are investing in AI projects and expect efficiency gains in the coming years; the majority report that AI is not currently reducing their need for staff.
Just under 80% of the managers interviewed said AI was not changing their staffing requirements. Around a fifth reported slightly lower demand, concentrated in consulting, software development and support functions such as accounting, legal and translation. Companies described preparing the ground — building the data basis, testing applications — and where they named an initial aim it was to improve service quality, deploy staff more systematically, or reduce outsourced work; reductions were expected only in the coming years. The second-quarter report, based on 243 talks held between 15 April and 2 June, records a somewhat higher share reporting reduced staff needs.
The cost side moves faster. In the second-quarter report, services companies name rising prices for software licences, cloud services, AI applications and IT security as their main issue, and IT service providers report longer delivery times for the products AI infrastructure requires, which is pushing up purchase prices for data storage. In the first quarter, ICT demand was supported by clients’ own efforts to cut costs through digitalisation, increasingly with AI.
Set beside the PwC data, the sequence is recognisable: efficiency named as the aim, the gain in the future tense, the invoice in the present one. The material is qualitative — discussions aggregated by the Bank, not a statistical sample — but it records where Swiss management effort is pointed.
European statistics show AI pointed mostly at functions any competitor can also automate
Eurostat’s 2025 survey on ICT usage in enterprises, extracted in December 2025, records that 19.95% of EU enterprises with ten or more persons employed used at least one AI technology, against 13.5% in 2024: 17% of small enterprises (10 to 49 employees), 30.4% of medium-sized ones and 55.0% of large ones.
Where they are applied is more instructive: among enterprises using AI, marketing or sales is the most common purpose at 34.7%, followed by business administration at 31.1%. Eurostat records the largest gap between small and large enterprises for ICT security — 47.5% against 14.5% — and 33.5% against 19.0% for production processes, both dependent on a firm’s own systems. The distribution is consistent with what PwC measured, without proving it.
Five decisions a company can take before its next licence renewal
Nothing here suggests the leading companies bought better technology. What distinguishes them is a short list of choices available at any size.
- Write the growth question next to the efficiency list: alongside “which tasks can this speed up?”, record the offer, market or customer problem AI now puts within reach. In PwC’s analysis, growth from convergence outranks efficiency as a driver of returns.
- Attach one business metric to each initiative and review it monthly. Leading companies are 80% more likely to track business impact systematically; stopping is rarer, with only 28% of leaders reviewing the portfolio to terminate initiatives to a large or very large extent.
- Redesign one workflow end to end rather than adding a tool per department: leading companies are twice as likely to do so instead of simply adding AI tools.
- Fix the single foundation blocking repeatability. Leaders are 1.7 times as likely to provide the high-quality data their priority applications need, and 2.4 times as likely to keep reusable, catalogued components.
- Treat governance as an adoption instrument, not a brake. Leaders are 1.7 times as likely to run a documented Responsible AI framework and 1.5 times as likely to have a cross-functional governance board; their employees are 2.1 times as likely to trust AI-generated insights and act on them.
Swiss conditions add a sixth discipline: track cost per outcome, not licences purchased. The SNB’s reporting shows input prices for AI rising while measured effects remain slight: the denominator of any return calculation is moving.
What the next four quarters will show
Three markers belong in the corporate calendar. The SNB submits its business cycle signals report with each quarterly monetary policy assessment, the third-quarter edition next; its chart on AI’s effect on demand for staff is now a recurring series, and a durable move away from “no change” would be the first Swiss sign that AI has shifted from a cost line to a capacity one. Eurostat’s enterprise survey runs annually. PwC judges that, without a change of approach, the gap it measures is likely to widen further.
Executives’ self-assessment remains a real limitation. What the evidence does establish is narrower and firmer: the returns a company can measure are not distributed in proportion to the activity it generates. They follow what that activity was aimed at.