AI in Switzerland: The CHF 15 Billion Opportunity for Business

The estimate describes potential additional annual economic output by 2034, rather than a ten-year total. It concerns AI-assisted research and development (R&D) across the economy, including companies beyond small and medium-sized enterprises (SMEs). Implement Consulting Group produced the report for Google and digitalswitzerland. ¹

Its model assumes widespread adoption and R&D efficiency gains of 10–15%. It extrapolates from leading companies to sectors, with the largest estimated gains in pharmaceuticals, biotechnology and food production. Wider spillover benefits are excluded. A company's returns still need to be measured locally. ¹

Look for work where expertise is difficult to access, an idea takes too long to test, or a customer need goes unanswered. The opportunity may be a better existing service, a faster development cycle or a new offer.

The Swiss AI Playbook makes this distinction explicit. Its framework covers personal productivity, process improvement, customer service and new business. It recommends defining a business ambition before selecting use cases. ²

For your company, that means asking a specific question: could you respond to technical enquiries sooner, test more product ideas with the same team, or package an existing capability into a service customers would buy?

The following are proposed experiments, not reported client results.

Engineering: turn technical knowledge into a faster proposal

Use approved specifications and previous project records to prepare a source-linked response to a customer's requirements. An engineer checks feasibility and approves the proposal.

Measure time to an accepted proposal, corrections and the margin on work won. A fast draft has little value if an engineer spends longer fixing it.

Product development: move customer evidence into a test

Combine customer feedback and approved development records to propose a small set of product hypotheses. Specialists choose which to test and define what would count as success.

Measure time from evidence to a tested concept and how often concepts pass the agreed test. The useful output is a learning decision, rather than a longer list of ideas.

Services: investigate an offer you could deliver repeatedly

Identify a recurring customer problem that currently needs expensive manual research. Test whether AI-assisted knowledge management can support a repeatable service, with human review and checks in the languages your customers use.

Measure customer acceptance, delivery cost and repeat purchases. This separates a convincing demonstration from an offer people will pay for.

Generative AI produces content such as text, images or code. AI agents combine model reasoning with tools to carry out steps towards a goal, with varying levels of autonomy. Workflow automation connects systems and executes defined steps. A useful business process can combine all three. ²

In the engineering example, generative AI drafts the response. An agent could retrieve permitted records and prepare a review task. Automation could record the approved outcome in the customer system. The engineer retains authority over the technical commitment.

“Agentic AI” is relevant to the Swiss discussion, but adoption is still developing. PwC's 2025 survey found 72% of its Swiss respondents were exploring or piloting agentic AI, while 15% were implementing and scaling it. The Swiss sample included 71 executives; 60% represented organisations with revenue of at least USD 1 billion. These are not SME adoption rates. ³

When planning AI automation, choose the amount of autonomy the job needs. A source-linked answer may be enough. A process that changes records or sends a customer commitment also needs explicit permissions, review and a way to handle exceptions.

SWISS's Operations Decision Support Suite combines operational data and proposes changes to aircraft rotations. Google Cloud's customer case reports more than CHF 1 million in savings during the first 14 weeks. Controllers review and approve the proposed changes. This is a supplier/customer account of operational optimisation, rather than evidence of a typical SME return. ⁴

The pattern is useful to investigate at a smaller scale: bring the relevant information together, prepare a recommendation, and let the responsible person act. Judge the system by the decision it helps complete.

Measure accepted work against a baseline, then include the cost of review, corrections and ongoing operation. Record whether returned capacity leads to extra delivery, a shorter backlog or a demonstrable reduction in spending.

For an illustrative pilot, suppose a recurring task takes 40 minutes per case. An AI-assisted version takes 10 minutes to prepare and 15 minutes to review. Across 100 cases, the difference is 25 hours of capacity. That is useful only if quality holds and the team uses the capacity. It is not automatically a cash saving.

Revenue needs separate evidence. A faster proposal process could help sales, but you need to observe conversion and margin before claiming that result.

Start with one business problem, a named owner and a test you can evaluate. The OECD's SME adoption work emphasises that companies differ in digital maturity, skills and resources. Their starting points should differ too. ⁵

The playbook supports this progression through prototypes, usable products and sustained operation. It also treats document preparation, data ownership and employee learning as part of adoption. ²

For a practical ranking method, use our automation opportunity scorecard. For the wider adoption picture, read AI adoption in Swiss SMEs.

1. Define the decision. Choose one delay, knowledge gap or customer problem and record its current cost.
2. Check the inputs. Identify approved sources, document owners, access permissions and missing information.
3. Test with real users. Compare accepted outputs with the baseline, including difficult cases and reviewer effort.
4. Decide from evidence. Continue, change or stop. Expand only after the process demonstrates value in ordinary work.

An AI investment should help your company answer a customer, make a sound decision or deliver something valuable. Start where you can observe that change.

Swiss Product Studio builds AI agents, workflow automation and internal tools around existing teams and systems. Bring one product, service or knowledge bottleneck to an operations review. We can examine the process, data and review requirements with you and identify a scope worth testing.

1. Implement Consulting Group. Accelerating innovation with AI in Switzerland, January 2026. Commissioned by Google and digitalswitzerland. Pages 25 and 49: economic estimate, assumptions and commissioning disclosure. Report overview. ↑1 ↑2 ↑3
2. Implement Consulting Group. The Swiss AI Playbook, first edition, 2026. Pages 17 and 53: generative AI and agents; pages 29–31: business ambition and impact areas; pages 38–41, 58–59 and 79: prototypes, operations, document preparation and employee learning. Playbook overview · AI Action Plan initiative. ↑1 ↑2 ↑3
3. PwC Switzerland. Cloud Business Survey 2025: Swiss findings, published 2026. Pages 4 and 33: sample composition and agentic AI adoption stages. ↑1
4. Google Cloud. SWISS customer case study. Operational decision support, reported savings and controller approval. Supplier/customer account. ↑1
5. OECD. AI adoption by small and medium-sized enterprises, December 2025. Page 18: differences in SME digital maturity, complexity and adoption scope. ↑1