AI for Business & SMEs
Build, buy, or partner for AI: which path fits your SME?
Build, buy, or partner: the three paths in brief
Every AI initiative confronts Swiss SMEs with the same basic question: do we adopt ready-made software, develop our own, or bring in an implementation partner? The three paths are not mutually exclusive — many successful solutions combine them. The key is to separate the options cleanly before you talk budget.
- Buy: Ready-made AI features inside standard software or SaaS — AI in your CRM, office suite, or a specialised tool. Fast to launch, low entry cost, little differentiation.
- Build: Develop and operate a bespoke solution with your own team. Full control and data sovereignty, but high effort, scarce expertise, and permanent operating responsibility.
- Partner: An external provider builds or integrates a custom solution. Faster than building alone and gives access to expertise — but raises questions of knowledge transfer, IP, and dependency.
When buying, building, or a partner makes sense
The guiding question is not «what is cheapest?» but «is this use case a competitive advantage or a commodity?». Anything many companies need in the same way — summarising minutes, drafting emails, searching documents — you buy. What makes your business unique and relies on your own data is a candidate for building or a partner.
- Buy when: the use case is standard, proven vendors exist, time-to-value matters, and no highly sensitive data has to leave your house.
- Build when: the solution creates real differentiation, relies on proprietary data, you can retain an AI/development team, and you intend to operate it long-term.
- Partner when: you need a custom solution but lack the know-how (for now), or building an in-house team does not pay off — with clear knowledge transfer as the goal.
Total cost of ownership: the real cost of each option
The purchase price or project quote is only the tip of the iceberg. Calculate across the full lifecycle — setup, operation, maintenance, model updates, and exit. A seemingly expensive solution can be cheaper over three years if running and upkeep costs stay low.
- Buy: predictable operating costs (subscription per user/usage), minimal entry, little maintenance — but recurring costs scale with usage, and the vendor sets price and roadmap.
- Build: high upfront investment (talent, infrastructure, data preparation), plus permanent costs for operation, monitoring, and model upkeep (MLOps). Operation is often underestimated.
- Partner: project cost plus an operating or maintenance agreement. Watch IP ownership, the operating model after go-live, and the cost of future changes.
- Always count: internal time, data preparation, change management, and training — these hidden items hit all three paths.
Assessing risk realistically: data, dependency, operation
Every path carries its own risks. For Swiss SMEs, data protection under the revised FADP (revDSG/nFADP) is central: where is data processed, who has access, and is a data-processing agreement in place? Clarify this early, not at go-live.
- Buy: vendor lock-in, data flowing to third parties (often outside Switzerland/EU), dependence on the vendor's price and roadmap decisions.
- Build: key-person risk (what if your specialist leaves?), technical debt, and full responsibility for security and operations.
- Partner: dependence on the provider, unclear knowledge transfer, IP and exit questions. Without a solid contract you risk a «golden cage».
- Data protection throughout: data residency, a data-processing agreement, purpose limitation, and transparency towards data subjects — accountability always stays with you as the SME.
A decision framework: seven questions before the budget
Answer these questions before you commit. They lead almost automatically to the right option — and often to a combination.
- 1. Differentiation: is the use case a competitive advantage or a commodity? Commodity → buy.
- 2. Data: does the solution rely on your proprietary data, and do you have the rights and the quality for it?
- 3. Sensitivity: how sensitive is the data (revDSG, professional secrecy)? The more sensitive, the more data residency and control matter.
- 4. Talent: can you build and retain the needed expertise — or would it be orphaned after the project?
- 5. Time-to-value: do you need results in weeks (buy), or is a longer build acceptable?
- 6. Total cost of ownership: what does the solution cost over three years including operation, not just entry?
- 7. Exit: how easily can you get out — data export, IP rights, switching vendors?
Common mistakes and the pragmatic Swiss path
Most failed AI projects fail not on technology but on the initial choice. The realistic path for most SMEs is hybrid: buy standard tasks, start one or two differentiating cases with a partner, and build competence internally step by step. Begin with a small, measurable proof of concept rather than a large investment.
- Building a commodity: re-coding standard features burns budget without differentiation.
- Buying without integration: a tool with no data and process strategy stays a useless island.
- Partnering without knowledge transfer and an exit clause: leads to permanent dependency.
- Ignoring TCO: operation, monitoring, and model upkeep are almost always underestimated.
- No success criterion: without a measurable goal, no option can be honestly evaluated.
- Starting too big: a focused proof of concept lowers risk and builds a learning curve.
Frequently asked questions
Should an SME buy or build AI?
For most Swiss SMEs, buying is the right start: standard software with AI delivers value quickly at low, predictable cost. Building only pays off when the solution creates a real competitive advantage, relies on your own data, and you can retain a team to operate it long-term.
When is it worth building AI in-house?
When the use case differentiates your business, relies on proprietary data, has no suitable off-the-shelf product, and you can build and retain the needed know-how. Budget for operation (monitoring, model upkeep, security) over several years — not just development.
What does a custom AI solution cost?
It can only be costed honestly per case. Think in cost structure rather than a flat price: one-off development (talent, data preparation, infrastructure) plus ongoing operation (hosting, monitoring, model updates, support). Over the lifecycle, operation is often the larger share — and the most underestimated.
How do I recognise a good implementation partner?
By transparency about costs and limits, a clear plan for knowledge transfer, your ownership of the result (IP and data), and a fair exit option. A good partner makes you gradually more independent, not permanently dependent — and knows Swiss data-protection requirements.
What does the revDSG mean for the decision?
The revised Data Protection Act (revDSG/nFADP) requires you to know where and by whom personal data is processed. For bought cloud tools you need a data-processing agreement and clarity on data residency. For sensitive data, data sovereignty and control argue for a self-operated or Swiss-hosted solution. Accountability always stays with you.
Can the three paths be combined?
Yes — and for SMEs that is usually the best route. Buy standard tasks, implement one or two differentiating cases with a partner, and build competence internally step by step. This combines fast value with long-term independence.
Key terms in the glossary
Practical AI for your business
From idea to implementation – we show you what is concretely possible in your case.
Request a demo