LEARN. PRACTICE. GROW.1. Adoption is observed behavior
A launch can have strong attendance and little effective use. If 20 people trained, 18 logged in and 12 demonstrated the task, these are three distinct milestones. With adoption defined as individual demonstration in the 20-person group, evidence is 60%. Do not remove people who have not logged in from the denominator to improve the percentage. Also examine why demonstration is missing. Someone may understand the method yet be unable to execute it because of access, hours or approval. Organizational change addresses working and learning conditions; it is not limited to technical release authorization.
2. Address barriers with the people doing the work
The night shift retains the previous procedure because new approval is available only by day. Resending a launch message does not solve the dependency. Work with the shift and control owner to prepare an authorized path, test the task and record the result. Do not label the limitation as personal resistance without investigation. If the method works only while its champion is present, rehearse that person’s absence and identify knowledge or support still centralized. Define maintenance, updating and help so improvement survives the pilot and personnel changes.
3. Separate implementation from effectiveness
A postmortem action may be coded without its control being demonstrated. If the criterion is blocking an invalid request before production, relevant proof includes that rehearsal and its observed result. Record “implemented, effectiveness to verify” when that matches the evidence. A blameless analysis still needs concrete actions: two near-identical forms without destination validation may justify rehearsed validation rather than a vague reminder to pay attention. Identify ownership, completion conditions and later review. The aim is to reduce the condition enabling failure, not merely close a record.
4. Represent unknown data
Of ten eligible runs, eight have results: one failure and seven successes. Two are unknown. The population rate may range from 10% to 30%; 12.5% describes only the known subset. Do not treat unknown as success to obtain 10%, or as certain failure to assert 30%. With a boundary between those extremes, evidence is insufficient to conclude. Check origin, cutoff and collection completeness. If repeating assessment is necessary, explain that a new rehearsal supplies new evidence; it does not automatically make old outcomes known. Keep population and rules explicit in reporting.
5. Compare without inventing causality
Fewer errors after improvement is an observation, but simultaneous version or volume changes limit causal attribution. Record concurrent factors and seek a consistent comparison before claiming the intervention produced the whole reduction. If failure definition itself changed, mark the series break and reconstruct an equivalent basis only with sufficient data. Across a flow, compare complete effects: three preparation hours, eight waiting hours and five execution hours total 16. Reducing preparation to one saves two; reducing waiting to four saves four under stated assumptions. Selection must still respect quality and admissibility boundaries.
6. Decision and continuity workshop
Prepare a decision note about 20 runs: two failures, 14 successes and four missing outcomes. A 15% boundary is not proven met because the possible range is 10%–30%. Explain which evidence must be recovered and who owns that action. Add the situation of 30 people with 18 demonstrations and unavailable night approval: 60% adoption does not authorize removing the approval condition. Propose a support path and the next rehearsal. Summary: observe behavior, resolve barriers, distinguish delivery from effectiveness, keep unknowns visible and prepare maintenance. Sustained benefit needs review after initial launch enthusiasm.
Eligible population: 20
Confirmed failures: 2
Confirmed successes: 14
Unknown: 4
Lower bound: 2 / 20 = 10%
Upper bound: (2 + 4) / 20 = 30%
Target: at most 15%
Conclusion: insufficient evidence to confirm compliance.20 runs: 2 failures, 14 successes, 4 unknown. Minimum rate: 2/20 = 10%. Maximum: 6/20 = 30%. The 15% boundary cannot be confirmed.
Common pitfalls
Training as adoption; merge as effectiveness; missing data as success; before/after as causal proof; maintenance without ownership.
Related topics: Organizational change · Measurement and learning
An improvement demonstrates its intended effect only with adequate evidence within agreed scope and conditions.
Reference: SRE Workbook: Organizational Change Management in SRE · ITIL 4 DPI; observed JA v1.3.1, current detailed revision comparison pending