Microsoft 365 exit cost: what the Swiss study really says

The Microsoft 365 exit cost for a Swiss administration is mainly a one-off bill, with staff time in training the largest line and running costs broadly unchanged, according to a study for Digitale Verwaltung Schweiz (DVS), the Confederation and cantons’ joint body for digital administration, dated 18 September 2026. Written by the consultancy Eraneos, it costs a complete move to the open-source suite openDesk for two real administrations, anonymised: at least CHF 25 to 30 million one-off for about 6,000 staff and CHF 9 million for about 1,500, over four to eight years. It recommends exploring gradual paths but did not cost them; on 29 October 2026 the DVS political committee took note of its findings.

This post covers the study’s scope and costs, its “seven to eight times”, the question it leaves open and how to cost a gradual path. The study is published in German and French; we quote the German original, and English translations are ours. Facts are as of 3 November 2026.

What did the Swiss study on leaving Microsoft 365 measure?

“Alternativen zu M365 und Konsequenzen bei einem Wechsel” (PDF in German and in French) follows the DVS “Second Source” study of 2025. A group of cantonal IT heads asked for it, and the DVS office commissioned it (p. 3).

• Two real administrations, anonymised. A canton with about 6,000 staff and a central IT of about 200, and a mid-sized administration with about 1,500 staff and an IT of about 70 (pp. 37 and 38). Five other cantonal administrations checked the findings by questionnaire.
• One illustrative target. openDesk, from Germany’s ZenDiS, was chosen “solely for illustration” and is “expressly not a recommendation” (p. 4).
• One scenario: all or nothing. Mail, calendar, files, collaboration and meetings, with their identity, security and compliance integrations, all replaced (pp. 3 and 91).
• Indicative figures. Orders of magnitude from cost data, expert estimates and assumptions, not budgets (p. 25).

Its conclusion: a complete, short-term replacement is not realistic without political readiness for the extra costs, a clear rise in staff and significant functional gaps (p. 6).

The DVS political committee took note of the findings on 29 October 2026 and classed digital sovereignty as a permanent strategic task. Taking note is not a decision to stay or leave: the study leaves the choice to each administration as “a political decision” (pp. 6 and 7).

What does the Microsoft 365 exit cost, line by line?

From the study (Table 11, p. 61; pp. 79 and 80):

One-off cost, CHF million	Larger unit (6,000 staff)	Smaller unit (1,500 staff)
Adapting specialist applications and data	5.65	5.0 (data 2.5; applications and macros 2.5)
Training and adoption	18.37	3.9
External consulting and validation	1.85	Not listed
Recruiting engineers	Not listed	0.06
Direct one-off total	25.87 (headline: 25 to 30)	About 9.0
Productivity loss after training	21.45, outside the total	Not costed
Parallel running of the old platform	30.66, outside the total (six years)	Not costed

Training is 71% of the larger unit’s direct total and 43% of the smaller unit’s (our arithmetic). It is priced at five days per person for 5,500 staff in the larger unit and four days in the smaller one, at CHF 650 a person-day covering direct training costs and imputed working time lost (pp. 60 and 79). The study means “the staff time tied up on the administration’s side”, not primarily course material or trainers (p. 8): a real cost, but mostly not an invoice.

It also hangs on one assumption, the number of days (p. 8). The validators’ estimates range “from less than a day to several days” per person (p. 86). At one day instead of five, the staff-time part of the larger unit’s line (CHF 17.9 million of the 18.37) would fall to about CHF 3.6 million (our arithmetic). Validators note that formal training tends lower and hidden onboarding higher (p. 86), and the study warns that costs “können im Einzelfall signifikant höher ausfallen” (may be significantly higher in individual cases) (p. 6).

“Seven to eight times” what?

The summary puts one-off costs at “rund das Sieben- bis Achtfache eines M365-Jahresbudgets” (about seven to eight times an annual M365 budget) (p. 6). The base differs between the units:

• Smaller unit: CHF 9.0 million against direct annual M365 costs of about CHF 1.255 million (licences, backup and disaster recovery, engineering, external support): about seven times (the study says seven to eight) (p. 80).
• Larger unit: “7-9-fachen der jährlichen M365-Lizenz-Kosten von 3.3 Millionen CHF” (seven to nine times annual licence costs of CHF 3.3 million) (p. 61). Licences are 64% of that unit’s direct M365 costs (p. 50), so against all direct costs, CHF 25.87 million is about five times (our arithmetic).
• Validators: where they estimated at all, “mindestens ein bis zwei, teils mehr als zwei M365-Jahreskosten” (at least one to two years of M365 costs, some more than two) (p. 86). The study ties the spread to scope, notably whether everything migrates or new services are built in parallel, and reads the answers as confirming “substanzieller, dominierender Einmalkosten” (substantial, dominant one-off costs) (p. 86).

Quote the multiple with its base, or not at all.

Are running costs lower after leaving Microsoft 365?

Not in this study, which finds them “broadly in the same range” (p. 6). For the larger unit (pp. 56 to 59), the theoretical licence and hosting saving against E5 at CHF 47.50 per user a month is about CHF 1.6 million a year. Against it stand a new security stack (mobile device management, endpoint protection, data loss prevention and identity components, about CHF 1.25 million) and secure remote access (about CHF 148,000). The net is “a few hundred thousand francs a year”, within the study’s margin of error, before replacing Power Platform apps and Graph API integrations, which it did not cost and expects to shrink or cancel that net (pp. 58 and 59).

Two cautions:

• The hosting line is a placeholder. No binding Swiss price for openDesk hosting existed in April 2026, so the study’s Swiss prices are “fiktiv” (fictitious), and its CHF 10 per user a month is “ohne Anspruch auf Marktrelevanz” (with no claim to market relevance) (pp. 56 and 57). It expects Swiss hosting to cost more than German, mainly because providers then ran one instance per customer, until openDesk is multi-tenant.
• The validators spread wide, from somewhat higher to significantly lower running costs, most of them comparable (p. 85).

So this study shows no running-cost saving from leaving, and a case for leaving built on cheaper licences alone is unlikely to survive review.

The question the study leaves open: gradual paths

The study recommends that administrations “graduelle Pfade zu vertiefen” (explore gradual paths further) (p. 7), with three options chosen per function: controlled operation, a second source in reserve, or a partial or full switch (pp. 15 and 16). Its outlook sketches mixed scenarios, such as moving file storage first or reducing Microsoft 365 to core functions, then says they were “nicht im Detail analysiert” (not analysed in detail) but are the “naheliegende nächste Fragestellung” (the obvious next question) (pp. 91 and 92).

Four findings show how its costs would behave; the first two cut against optimism:

• A partial move may save less on licences. Parts of identity and endpoint management now covered by E5 would need new licences, typically standalone ones, so the net saving from a selective replacement “geringer ausfallen kann als bei einem vollständigen Wechsel” (may be lower than from a complete switch) (p. 59).
• Bought-in operations leave integration behind. The smaller unit’s experts expect a hoster to run the platform and basic monitoring, while “ein wesentlicher Teil der Integrations- und Abstimmungsarbeiten aber bei der eigenen Informatik verbleibt” (a substantial part of the integration and coordination work stays with the organisation’s own IT) (p. 78), and the unit’s engineering rises from 1.5 to 2.25 full-time equivalents even allowing for what ZenDiS and the hoster take on (p. 79).
• Training scales with what moves. The smaller unit assumes about one person-day of training per main component group (p. 79).
• Data decides the cost of the next step. With data classified, in open formats and stored in places you control, the application layer “can be added to, replaced or combined with much less effort” (p. 13).

So a gradual path keeps most of the Microsoft bill and adds each moved function’s running cost. It buys control over chosen functions, not a saving, in steps you can measure before taking the next.

How to cost a gradual path off Microsoft 365, and by when

The dates are our suggestion, not legal deadlines.

1. By 31 December 2026: rebuild the baseline for your estate. Split direct Microsoft 365 costs as the study does (licences, engineering, external support, backup), with contract end dates and integrations per function.
2. By 31 March 2027: choose functions by control and exit value, not licence saving. Start where data needs the most control and integrations are simplest: the study finds migration more feasible where applications use generic file operations, WebDAV or REST, and harder with SharePoint metadata, workflows or connectors (p. 76). Power Platform, macros and telephony can wait. For a fallback instead, see the federal approach.
3. By 30 June 2027: write the cost case for one function. Training as the main line (staff affected, times days, times a daily rate), integration hours, the Microsoft licences that remain (priced at the next renewal), an operations quote and the security integrations. Assume no licence saving. For openDesk, check who can supply it in Switzerland.
4. Before signing: buy operations, keep integration. Contract updates, backups, monitoring and incident response; keep an internal owner for how identity, files and specialist applications fit together. Device management can stay with your current tools; decide how classification and data loss prevention will cover the moved function (p. 75).
5. From July 2027: run it in production for one group, and measure. One real update, one backup restored and one full export tested, against success criteria written in advance (what a pilot needs). Record training days, tickets, integration hours and cost per user.
6. Before each renewal: decide the next function on measured figures, and cut licences only where the remaining plan really costs less.

In this study, open source changes the exit, not the bill

The study gives open source no running-cost advantage in this scenario, and it would be wrong to claim one. What it credits is control: “digital sovereignty is decided by the data, not the applications”, tested among other things by the “ability to change operator through open formats and standardised interfaces” (p. 13).

The study is also clear about what that does not buy. In its experts’ view, the ability to exit “does not improve automatically”, because much lock-in lies in working habits and processes, and even a modular suite would settle into a de facto lock-in, making a later switch again a major change project (p. 77). A switch brings “no automatic fall in risk, only a changed exposure” (p. 88): sovereignty risk may fall while operational and cyber risk may rise (pp. 88 to 90).

That holds for any change of tool. What open formats change is the operator, which is why the study tests for it (p. 13), and it expects security to depend more on “der Qualität des gewählten Betriebsmodells” (the quality of the chosen operating model) and its partners (p. 90). Part of that is operations work: patching across components, single sign-on, logs that reach the SIEM, backups that restore and someone who answers. None of it comes with an open-source licence; it can be specified, measured and written into a contract. The integration the study leaves with your own IT (p. 78) still needs an owner on your side.

How Pilae fits

Pilae operates open-source apps on your premises or on dedicated machines in Pilae Cloud. On a gradual path, that is one function at a time beside the Microsoft tenant you keep, with the operations ours and the split of every other task written down. Pilae does not operate mail, which stays with Exchange or your mail provider (Microsoft 365 comparison).

• One function beside Microsoft 365. The sovereign workplace can start with one function, such as files on Nextcloud, while Microsoft 365 runs the rest; from the Business plan, Keycloak can federate with Entra ID so staff keep one login.
• A pilot, a parallel run and a rollback. A migration follows a fixed-price written plan: inventory, a pilot team whose feedback adjusts the training and runbook, a parallel run, a cutover with a rollback, and training sessions for your champions.
• Operations bought, the split in writing. On your premises, from the Business plan, Pilae runs everything from the operating system up, and the agreed split of every task is an annex to your contract (shared responsibility). The Pilae Agent plans each update and waits for your approval.
• Security that plugs into what you run. From the Business plan, single sign-on through Keycloak or your Entra ID, with multi-factor authentication per group. Upstream advisories are tracked against your versions, and every login, approval and change goes to an audit log exportable to your SIEM.
• The exit in the contract. For a public body, the contract names Switzerland as the only data location, and a full export of your data and configuration in open formats is available at any time and at the end of the contract (public sector).

Our plans and prices are on the pricing page. To cost a first function, talk to an engineer: we go through the functions you would move first, the integrations they touch and what operating them properly would involve.

This post is general information, not legal or procurement advice. For the full text, read the study (PDF in German or in French).