The middle option almost nobody picks¶

A plain-language companion to the draft working paper "Open Source Back-End Services: A Migration Guide for Directory, Storage, Mail, and Real-Time Communication" (v0.9, July 2026). The guide carries the product assessments, the scoring template, the reference architectures, the migration mechanics, the break-even model, and the contract chapter; this text carries the ideas.

Read the full draft working paper (PDF)

The short version. Organisations tend to frame the collaboration back end as a choice between staying with the incumbent and migrating off it. There is a third position between them, it is under-used, and for a large share of organisations it is the right answer: build credible optionality without migrating. Federate identity, move to open formats, run a working reserve. That buys a real exit threat, knowledge of your own dependencies, and continuity cover, for a fraction of a migration, and it is also the correct first phase of a migration, so choosing it costs nothing if you go further later. The guide adapts three public studies to organisations generally, and it is candid about the headline: no source in its corpus supports a claim of cost savings. The case rests on exit cost, escalation exposure, and optionality.

Three positions¶

Stay and negotiate. Nothing structural changes; you use the threat of the other two. This buys nothing unless the threat is credible, and an incumbent can tell. It is cheap, and it decays, because leverage you never demonstrate stops being believed.

Build credible optionality. Federate identity, move to open formats, run a working reserve. No full migration. This buys most of the benefit of migrating for a fraction of the cost and a few months of work.

Migrate. Replace the stack, domain by domain, over years. This buys the above, plus removal of the licence line and the escalation exposure. It is investment-shaped: worse before better.

What you are actually buying out of¶

The guide treats exit cost rather than licence cost as the object, following the Schleswig-Holstein strategy's observation that the price of client access licences is best understood as the price of a dependency. The licence line is what you see; the dependency is what you are paying for.

Two more departures from the source studies shape the recommendations. Supplier depth is treated as a first-class selection criterion alongside function and maturity, because a single-vendor product replaces the identity of a dependency without changing its structure, and only a plural supplier base turns an open licence into an actual ability to change your mind. And directory of record, authentication, and device policy are separated into three procurement decisions rather than one search for a product covering all three.

What it costs, without guessing your contract¶

The model derives the figure that needs no assumption about your agreement: the incumbent cost per seat per year at which migration breaks even.

Euros per seat per year for the back-end stack. Under an incumbent price rising at the rate implied by the only published series in the corpus, every figure falls by roughly a quarter at three years and by half at ten.

Two things follow. Migration cost is close to independent of estate size, so the fixed cost of doing it at all dominates and small organisations face the hardest arithmetic. And the three-year comparison usually favours the incumbent, correctly, because the investment phase sits inside that window. A board that evaluates on three years should be shown the ten-year column beside it, with an explanation of why.

The phasing runs 6 to 12 months of foundation work (inventory, identity federation, open formats, the reserve, all reversible and valuable on their own), 12 to 30 months of substitution (files, editing, mail and groupware, which is where the disruption lives), and 12 to 24 months of consolidation (real-time communication, telephony, print, decommissioning). Three to five years for a mid-sized estate, against four to seven observed in the two documented field cases.

The five things that actually decide it¶

The incumbent agreement's renewal date and exit terms, frequently a harder constraint than any technology, since reducing seats mid-term often saves nothing.

Whether you keep Outlook, which determines the groupware choice and the sequencing of the largest domain.

Where you will run it. Open-source software on infrastructure you were trying to become independent of changes the licence bill and little else.

Whether you can staff a platform capability or will buy the stack operated. Both are legitimate; pretending is not.

Whether the sponsorship outlasts a change of leadership. This is the only risk the source corpus rates catastrophic, and it is the mechanism that ended the best-known programme in this field.

What improved, and what stays hard¶

Two positions changed materially after the source studies were written: groupware with native Outlook compatibility, and integrated sovereign suites, an option that did not exist in 2024. Credible open-source options now exist in every domain the guide examines.

Four problems remain hard, and each is a reason to sequence around it rather than a reason not to start: tenant-wide AI assistance grounded on your own content and permissions; secure pull-printing with badge release; enterprise telephony as a whole; and calendar free/busy across the boundary during coexistence.

Where the work actually is¶

Product selection is a third of the guide. The rest is what the source studies leave out, which is most of the engineering: the hosting decision, on which most of the sovereignty turns and which a guide discussing only software would miss entirely, then reference architectures and sizing, data-migration mechanics, coexistence patterns, backup and continuity, security baselines, operations, and what to put in a contract. A migration fails in those chapters, and succeeds in them.

The small print¶

This is a draft working paper, a practitioner reference rather than an economics paper. Product statuses are stated as of a given date, and entries the guide marks as moving need re-checking before any procurement decision. The break-even figures rest on stated assumptions and are meant to be re-run with your own numbers. Three of the six source documents contain no cost figures at all, and the most rigorous concludes cost parity over a long horizon. The companion volume covers the end-user workstation, the desktop, and the managed Linux estate.