Autonomy-first transformation: the missing category

The industry's most quoted figure — "70% of transformations fail" — has no valid empirical basis. In 2011, Mark Hughes traced it back to its five founding publications: no reliable evidence. The myth suits everyone: it sells support services.

Our thesis, refutable: the problem is not the figure, it is the metric. The serious studies measure budget, schedule and scope — none measures what the client can do alone once the provider is gone. As long as that metric is missing, the market rewards dependency. The missing category is the one whose success criterion is the client's autonomy: autonomy-first transformation.

The solid data, meanwhile, is severe. Across 1,471 IT projects (Flyvbjerg & Budzier, 2011), the average overrun is 27%; one in six becomes a "black swan": +200% cost, nearly 70% delay. The database of 16,000+ major projects (Flyvbjerg, 2023) is even starker:

Three columns — budget, schedule, benefits — and still none that measures what your teams can do after the provider leaves.

The market's first model sells the recommendation: sound diagnosis, execution on you — you are left alone the moment difficulty begins. The second sells execution as an annuity: every change goes back through the provider, who has no structural interest in your learning.

Mazzucato and Collington (UCL, 2023) show how heavy reliance on external consulting captures the learning loop: the client stops learning. The UK National Audit Office put numbers on it in 2025: a state spending £14 billion a year on digital fields only 15 people to manage its 19 largest suppliers. The skills asymmetry is not a side effect: it is the mechanism of dependency.

Autonomy-first transformation defines itself against both models: at the end of the contract, your teams operate, modify and extend the system without us.

This practice has a name: the Ownward Method. Five principles, verifiable in a contract, extending how we build.

In our own work: 17 custom Airtable extensions in one year, including a full CRM, adopted immediately because the team stayed in its tool; a synchronization engine connecting six CRM/ERP systems through readable configuration — a new source is a mapping, not code (Ownward internal data, 2026). Switzerland already applies the second principle to its state: since 2024, art. 9 of the EMBAG act requires publishing the source code of software developed for the administration.

The EU Data Act has applied since September 12, 2025: open interfaces, standard exports, full removal of cloud switching charges — egress included — on January 12, 2027. Designing for reversibility means anticipating the norm; the five forms of vendor lock-in provide the reading grid.

Exposure keeps growing: 52.7% of EU enterprises used paid cloud services in 2025, up from 17.8% in 2014 (Eurostat). As early as 2016, 71% of surveyed businesses cited lock-in as a brake on cloud adoption (Opara-Martins et al.).

On the skills side, the arithmetic is stubborn: 39% of key skills will change by 2030 according to employers (WEF, 2025); Switzerland projects 54,400 missing ICT specialists by 2033 (ICT Vocational Training Switzerland, 2025). You cannot hire your autonomy: every project has to transfer it to your teams. That is what autonomy-driven transformation means.

What doing nothing costs — One IT project in six ends as a black swan: +200% average cost overrun and nearly 70% delay (Flyvbjerg & Budzier, 1,471 projects, 2011). The 2025 UK public audit prices five digital programmes at £3.29 billion in overruns and 29 years of combined delay. And every year of dependency worsens the asymmetry: the less your teams learn, the more the exit costs.

Ask them pre-sales: they gauge business models, not people.

A right-column majority guarantees nothing; a left-column one tells you what you are buying.

Autonomy-first is not always the right purchase. For a capability you do not want to internalize — payroll, highly specialized infrastructure — durable delegation is rational. Transfer has a cost: it mobilizes your teams during the project, and that time must be budgeted. Nor can autonomy be decreed from the provider's side: without an internal counterpart who wants to learn, the best mechanisms fail (Goyette et al., 2015). Finally, our examples are internal data, not an independent study: they illustrate the method without proving it.