Capturing Sole-Source Commissioning Knowledge Before Nine Retirements
Caldon Water operates 62 pump stations across a drought-prone region. Nine senior engineers are eligible to retire within 36 months; four hold sole knowledge of commissioning logic for 31 of the stations. A previous capture effort produced 220 hours of video interviews. Median views per video are three, and new engineers still phone retirees at home. Last year a station startup was delayed 11 days because a pressure-surge workaround was known only to one retiree. The KM team has 600 hours of SME time and one knowledge engineer. The union agreement makes retiree participation voluntary. A consultant proposes transcribing the videos and layering a chatbot over them. Operations wants everything captured before the retirements. The head of engineering asks for a defensible capture plan.
1. How would you prioritize what to capture, given that 'everything' is impossible and participation is voluntary? Name the criteria and apply them.
Read the worked answer
Prioritize by consequence, frequency, and substitutability, not by seniority or tenure. The unit of analysis is the commissioning decision, not the person's career.
Apply the criteria:
- Consequence: a wrong decision causes safety, environmental, or multi-day startup delay. The 11-day pressure-surge delay is a reference case.
- Frequency: the decision recurs across stations or is likely during the next 36 months of capital work.
- Substitutability: the knowledge is held by one person and is not in drawings, manuals, or vendor documentation.
Start with the 31 sole-source stations and collect the last 24 months of commissioning deviations, near misses, delayed startups, and workarounds. Use critical-incident interviews to identify five to ten decisions per station family: what cues the problem, what options were considered, what rule or threshold applied, and what exception invalidated the standard procedure. Do not capture general career stories or repeat what is already in the OEM manuals.
Because participation is voluntary, frame the ask around specific decisions and pair each retiree with a successor on live work. That gives the retiree a defined scope and the successor a reason to participate.
2. Design the capture method. Why is a chatbot over existing videos unlikely to close the gap?
Read the worked answer
The method should produce decision-grade artifacts, not more narrative.
Use a structured elicitation sequence:
- Task analysis for each priority commissioning decision.
- Decision tables or if-then rules for stable, repeatable logic.
- Annotated P&IDs and asset-tagged walkthroughs for physical context.
- Paired commissioning simulations or tabletop exercises with the successor doing the work and the retiree observing and correcting.
- An exception log with owner, trigger, and escalation path.
Codify only what is stable. Leave judgment-heavy, context-dependent reasoning in a mentorship relationship with named backups.
The chatbot over 220 hours of video is unlikely to close the gap because video is narrative and not indexed to the decision point. A chatbot can retrieve a passage, but it cannot reliably infer the station, the operating state, the sensor reading, or the consequence of the workaround. It also cannot validate whether the retiree's account is current. The cost of a wrong answer during commissioning is high. A better use of the videos is as source material for structured extraction, with the retiree reviewing the resulting playbook. Validate by having a successor handle a non-routine scenario without phoning the retiree.
3. How would you validate retention and measure success? What would you not measure?
Read the worked answer
Validate retention behaviorally, not by content volume.
Validation:
- Successor completes a tabletop or live commissioning task for a priority station without assistance.
- Time to resolve a non-routine commissioning issue falls, and escalations to retirees decrease for covered decisions.
- Startup rework and delay reasons are tracked against the prior 24-month baseline.
Success measures:
- Percentage of sole-source station families with a validated playbook and a named backup.
- Percentage of priority decisions where a successor has demonstrated the decision, not just read it.
- Retiree willingness and successor confidence, reported alongside the behavioral tests, not instead of them.
Do not measure video views, video minutes, or number of documents produced. Those measure supply, not retention. Also avoid a counter-metric blind spot: track documentation burden on operators. If the playbook is too long to use in the field, it will fail. The defensible claim is that a named successor can perform a defined task under realistic conditions, not that 'everything was captured.'