Measuring autonomy: four metrics that don't lie

Our thesis is falsifiable: autonomy over tools can be measured, and four metrics are enough. The real starting points of our projects are a reminder: an Excel file updated by hand every Monday, a process that exists only in people's heads, six systems that never talk to each other — nobody had measured them (Ownward internal data, 2026). Not a single score — the SPACE framework (ACM Queue, 2021) settled that: “Productivity cannot be reduced to a single dimension”. These four autonomy KPIs, collected without heavy machinery, are the foundation of an autonomy-first transformation: measuring a transformation before declaring it a success — it is what shapes how we build.

This metric transposes DORA, the longest-running research program on software delivery (more than 39,000 professionals over ten years), which measures change lead time “from committed to version control to deployed in production”. In a business-tools context, the boundaries change — better to say so plainly.

The measurement: two dates per request — submitted, in service — in a shared table. When the architecture allows it, the gap collapses: on our synchronization engine connecting six CRM/ERP systems, a new source is a new mapping, not new code; the finance team publishes its dashboards in minutes, through Claude (Ownward internal data, 2026). Trap: comparing heterogeneous processes — DORA forbids it.

The measurement, over a quarter:

• changes shipped in total;
• of which shipped by the team alone, with no outside intervention.

The ratio tells you who really runs your tools. Context makes it strategic: 57.5% of EU companies struggle to recruit ICT specialists (Eurostat, 2023 data), and Gartner forecast in late 2022 that at least 80% of low-code tool users would sit outside IT by 2026 — this year. This ratio does not rise by decree: it requires the framework described in citizen developer governance. Trap, inherited from SPACE: confusing activity with autonomy — ten automations built prove nothing if every modification waits for a third party. Our reference point: 17 custom Airtable extensions built in one year, inside the tool the team already works in (Ownward internal data, 2026).

The truck factor: the minimum number of people whose departure would cripple a system. Across 133 popular open source projects, about 65% sat at 2 or fewer (Avelino et al., 2016). Critical knowledge does not live only in code: the “Bus Factor In Practice” study (ICSE 2022, 269 engineers at JetBrains) also finds it in reviews and meetings.

Transposed to business tooling, the minimal measurement, per critical process:

• who knows how — the irreplaceable people;
• who has access — admin accounts, workbooks, licenses held by a single person.

We took over a weekly dashboard maintained by hand every Monday — bus factor of 1 — now a multi-tenant SaaS (Ownward internal data, 2026). Trap: counting only code.

The FinOps Foundation formalizes unit economics: connecting technology spend to the value created — cost per transaction, per customer, per resolved case — rather than a global budget line. Per automated process: subscriptions, maintenance time and external interventions, divided by the volume handled. The mechanics are detailed in the real cost of a SaaS subscription.

Measurement: one quarterly review, three lines per process. The CNCF platform engineering maturity model legitimizes this: rough metrics are a maturity level, not a fault — the mistake is staying there. Trap: a global TCO, which hides drifting unit costs.

DORA 2024: an internal platform brings +8% individual productivity and +10% team performance, but first −8% throughput and −14% change stability. DORA 2025: “AI's primary role is as an amplifier” — without measurement, it amplifies dependency. Hence: measure before you transform.

What doing nothing costs — A bus factor of 1 comes due the day someone leaves: 57.5% of EU companies struggle to recruit ICT specialists (Eurostat, 2023), and Switzerland projects a shortfall of 54,400 ICT specialists by 2033 (BSS / ICT-Berufsbildung Schweiz, 2025). Every change handed to a third party renews that dependency without ever making it visible.

Adapt it to your processes — and never compare them with each other.

The DORA transposition has not been validated by research for business tools: the figures cited come from software teams, not from Airtable or ERP processes. Four metrics capture neither team satisfaction nor deliverable quality — SPACE recommends crossing several dimensions. Without history, allow two or three quarters before drawing any trend. The bus factor remains partly qualitative. Finally, the Gartner forecast dates from December 2022: its order of magnitude is still useful, not its precision.