SaaS versus custom development in Switzerland: what generative AI shifts in the trade-off

Note revised on 25 May 2026. Article originally published in April 2026 — full rewrite in the editorial register.

The software subscription — Software as a Service, or SaaS — has imposed itself for fifteen years as the default choice for almost all the support functions of a company. Customer relationship management, project tracking, document management, accounting, electronic signature, team scheduling: each function has found its publisher, each publisher its subscription model per user or per module, and the arithmetic has imposed itself as an obvious one: it was necessarily faster, cheaper and safer to rent a mature tool than to build the equivalent in-house.

This obviousness is giving way. Three forces are simultaneously shifting the trade-off: the structural pricing inflation of SaaS publishers, the regulatory tightening of data sovereignty in Switzerland, and the fall in the cost of software code under the effect of development tools driven by generative models. This note sets out the dynamics of this shift, and qualifies what it calls for in the software strategy of a Swiss SME.

First force: pricing inflation by SaaS publishers

Subscription software prices have risen significantly faster than general inflation in developed economies over the past few years, according to several converging sector studies. Among the most established publishers, price increases on the installed base now represent a substantial share of revenue growth, ahead of new customer acquisition.

Concretely, a Swiss company of a few dozen staff sees its annual software-subscription bill grow year after year, even as the effective use of paid features plateaus — the majority of staff use only a fraction of each tool's capacities. This inflation does not finance an asset belonging to the company. It finances the right to use a tool whose publisher can unilaterally modify the rate, remove a feature deemed unprofitable, or refuse a particular use case at renewal.

This pricing dynamic has a perverse effect known as vendor lock-in. The more a tool is used in depth within an organisation, the more the specific data formats, proprietary workflows, integrations with the publisher's other tools, and multi-year contracts make exit costly. This situation places the company in an unbalanced negotiating relationship at successive renewals.

Second force: regulatory tightening in Switzerland

In November 2025, the conference of Swiss data-protection commissioners — Privatim — published a structuring position restricting the use of international SaaS services by public authorities for processing sensitive data, except where end-to-end encryption is in place and the provider has no access to the decryption keys^[1]. This position fits within a broader trend of reaffirmation of data sovereignty.

The underlying issue concerns the extraterritorial application of US law to data hosted by companies under US jurisdiction, even when physical storage is located in Switzerland. This subject can raise questions of applicable law and data access that exceed the direct contractual framework between the user company and its publisher. For a country that has built its economic reputation in part on the confidentiality of its practices, this extraterritoriality produces a tension that thickens with the years.

Privatim's position concerns the public sector first. Observation shows that regulated industries — banks, insurers, the health sector, organisations linked to foreign affairs — progressively align with these standards, sometimes in anticipation of forthcoming regulatory developments. The Federal Act on Data Protection^[2], revised in September 2023, requires adequate technical and organisational security measures that are not mechanically satisfied by a subscription to a publisher under foreign law.

Custom development, hosted with a Swiss or European provider under a compatible jurisdiction, sharply reduces this category of risk if hosting, keys and operation remain under control. The company controls its code, its data, its encryption keys. The operational perimeter becomes documentable, which is not systematically the case in a subscription to a publisher under foreign law.

Third force: the fall in the cost of software code

This is the most recent of the three forces, and probably the most structuring. Development tools driven by generative models have collapsed the time separating a software idea from its functional implementation. The developer using a current coding assistant can produce significantly faster than two years ago, at comparable delivery quality, provided they maintain code-review rigour and architectural discipline.

This compression of production time has an immediate effect on the boundary between subscription and in-house development. A software project that would have demanded several months and a substantial budget from an agency five years ago can be delivered on a shorter horizon today, with a tightened team and a budget envelope that makes the economic trade-off different. The barrier to entry of custom development has been lowered.

This lowered barrier does not mean that everything must now be developed in-house. It means that the arbitration zone in which custom development becomes competitive has widened. Tools whose features were once too costly to reproduce in-house now become serious candidates for replacement by a proprietary equivalent. This dynamic is documented in the note Replacing a SaaS subscription with a local application driven by an open model, which sets out an observed case at a small scale.

A third path emerging: code delivered with its maintenance agent

The classical objection to custom development concerns maintenance over time. When the provider that built the tool disappears, changes its rates, or loses the person who mastered the code, the company finds itself in an operational dependence it thought it had avoided by leaving the subscription model. This objection has long justified preferring a stable publisher, despite its cost.

A third path has emerged in the past eighteen months. It becomes possible to deliver a custom software product accompanied by its own assisted-maintenance tooling. On delivery, the client receives not only the source code of their application — which they own in full property — but also a set of model-driven agents, calibrated on the architecture of that code, on the business vocabulary of the organisation, and on the operational workflows. An agent that knows the code and can generate new features. An agent that produces documentation of changes. An agent that ensures the coherence of integrations with other systems.

This configuration redefines the client's posture. They are no longer the tenant of a tool on which they depend to evolve. They are no longer merely the owner of code that they will have to entrust to a new provider to make it evolve. They are autonomous in the maintenance and evolution of their tool, with a reduced in-house team or a provider of their choice.

This path is not relevant for everything. It assumes that the operational need is sufficiently circumscribed for a custom tool to be definable, and that the organisation has at least one in-house — or contractable — competence to maintain the dialogue with the agent tooling delivered. But when these conditions are met, the combination rewrites the economic equation of the previous decade.

Five questions that structure the trade-off

The choice between SaaS subscription and custom development is not ideological. Some situations call for subscription, others for development. Five questions structure the trade-off more reliably than immediate pricing comparisons.

Does the software touch the company's competitive advantage? If yes, custom development protects what distinguishes the organisation from its competitors. If the function is a commodity — standard accounting, basic payroll management — subscription remains relevant.

How many users are concerned? The per-user pricing model of subscriptions becomes quickly expensive beyond a certain threshold. Custom development has a fixed cost largely independent of the number of users, which modifies the arithmetic for growing organisations.

What data passes through the tool? If it is sensitive, subject to professional confidentiality obligations, or covered by professional-secrecy rules, sovereignty imposes hosting and control that a subscription under foreign law does not provide.

What is the strategic horizon? Over a three-year horizon and beyond, the total cost of ownership of custom development becomes comparable, sometimes lower, than the accumulation of subscriptions. For a temporary or experimental need, subscription retains its advantage.

Has the organisation adapted to the tool, or does the tool adapt to the organisation? If the answer is the former, the company pays a hidden cost in productivity, team frustration, and loss of agility whose bill never appears in the financial statements.

Taking the trade-off afresh

The movement described here is not a return to the heavy IT of the 2000s. It is a shift in the boundary between subscription and custom development, under the effect of the three simultaneous forces identified above. This boundary has moved towards development, and will continue to move there in the coming years as software-production tools driven by models gain in maturity.

For a Swiss SME, the practical consequence holds in an arbitration discipline. Each software-subscription renewal deserves to be posed afresh, rather than renewed by habit. The functions where subscription remains the right decision remain numerous. The functions where custom development becomes competitive expand. And the intermediate zone — where the trade-off must be supported by a specific analysis, without prejudging one side or the other — also widens.

A software built for a company and owned by it remains an asset that enters its patrimony. A subscription remains a recurring expense that does not enrich the balance sheet. Over the long horizon, the gap between the two is measurable, and it is now measured in a wider zone than it was five years ago.

Sources

[1] Privatim, Resolution on international cloud solutions, November 2025. www.privatim.ch/de/privatim-verabschiedet-resolution-zu-internationalen-cloud-losungen/ [↩]

[2] Federal Act on Data Protection (FADP), revision of 25 September 2020, in force since 1 September 2023. www.fedlex.admin.ch/eli/cc/2022/491/en [↩]

Jérôme Deshaie is CEO of MCVA Consulting SA, a Swiss firm specialising in strategic consulting on artificial intelligence, based in Valais.