Methods and evidence

What the Sociotek diagnostic is built on, how it is being tested, and what it can and cannot claim today. We publish this so you can judge it for yourself.

Where it stands, 17 November 2026

The instrument is research-informed and under validation. The design is complete. Expert review and cognitive interviews with intended respondents are the next stage, followed by pilots with founding design partners. Until the evidence supports more, we describe it as a structured readiness conversation.

What it is built on

The diagnostic rests on sociotechnical systems theory, developed at the Tavistock Institute from the late 1940s: the people side and the technical side of work form one system, and improving one while ignoring the other makes the whole perform worse. AI agents test that principle directly, because they take on parts of the work and the decisions that used to belong to people.

From that starting point the diagnostic looks at nine dimensions: six describing the organisation that must absorb the change and three describing the conditions agents need to operate safely. It keeps two lenses apart: what the organisation can responsibly absorb, and how much autonomy the AI is expected to have. From the fourth of six autonomy levels upward, a short list of controls is treated as a condition of going ahead, which no strength elsewhere can make up for.

How it is being tested

  1. Expert review. Specialists in organisation design, work design, measurement, AI engineering, security, risk, HR and line leadership rate each statement for relevance, clarity and whether it can actually be observed.
  2. Cognitive interviews. People in the roles that will answer the questions talk through how they read each one, so that wording problems are found and fixed before any numbers are trusted.
  3. Pilots. Founding design partners use the diagnostic on real workflows. We look at completion, how often people cannot see enough to answer, how much groups disagree, and above all whether it changes a decision.
  4. Structural testing. Once several hundred responses exist across organisations, statistical testing checks whether the dimensions hold together as intended, using separate samples to explore and to confirm.
  5. Links to outcomes. Where clients allow, results are compared with what happened next: whether pilots scaled, incidents, trust, rework and decisions changed.
  6. Comparisons. Only after enough varied data exists will results be compared across organisations, with minimum group sizes to protect confidentiality.

What it does not claim

  • It is not yet validated, predictive or certified, and its results are not benchmarks.
  • Answers are people's own views of their organisation. One person's view is a starting point, which is why the full diagnostic compares several groups side by side.
  • The score bands used in results are provisional, for communication only, until data supports firmer thresholds.
  • No comparison with other organisations will be shown until a comparison group is large enough to mean something and to protect confidentiality. Our working floor is 30 organisations per group.

How changes are handled

Every statement carries a version number. Changes are recorded rather than made silently, the core statements are reviewed each year, and the statements about fast-moving agent capabilities are reviewed every six months.

Research it draws on

How the diagnostic works and the plain-language glossary