How to Codify Tacit Knowledge and Institutional Judgment
Most of what makes experienced staff effective lives in their heads, not in process documentation. When agents take over workflows, these undocumented exceptions and contextual judgments become critical gaps. This playbook gives you structured methods for extracting tacit knowledge, documenting it in agent-usable formats, and keeping it current as organizational practices evolve.
This playbook covers the how. For the why and what, see the
skill definition
.
Developing Start here. Build the foundation.
- Pick one workflow that an agent currently handles or will soon handle. Schedule a 60-minute session with the most experienced person who used to do that work. Use scenario-based questioning rather than open-ended interviews: walk through 5 real cases from the past month, including at least 2 that involved exceptions or unusual circumstances. For each case, ask: 'What did you consider that is not in the process documentation? When would you have handled this differently?' Record the answers as specific decision rules and add them to the agent's knowledge base.
- Create a two-column document for the workflow: 'What the documentation says' in one column and 'What people actually do' in the other. Observe 3-5 instances of the workflow being executed by experienced staff and note every deviation from documented procedure. These deviations are not errors. They are accumulated wisdom about how the process actually works. Document each deviation as a decision rule with the condition that triggers it and the action it requires.
- For each process you document, specifically ask experienced staff: 'When do you not follow this process?' Create an exceptions list that captures the circumstances, the alternative approach, and the reasoning. Most process documentation describes the happy path. Agents need the exception paths too, because those are where the most consequential judgment calls happen.
Proficient Build consistency and rhythm.
- Build decision trees for the 3 most complex judgment calls in each agent-handled workflow. Start with the initial decision point and branch for each factor the expert considers. Keep each tree to a maximum of 5 levels deep. If a branch gets more complex than that, split it into a sub-tree. Test each decision tree by walking through 10 real historical cases with an experienced staff member. If the tree produces a different answer than the expert would have given, refine the branching logic until it matches at least 90% of cases.
- Conduct a gap analysis between your documented tacit knowledge and agent behavior. Pull a sample of 20 agent decisions from the past month and evaluate each one against the decision rules and exception handling you have documented. Identify cases where the agent made a decision that an experienced human would have handled differently. For each gap, determine whether the cause is missing documentation, ambiguous documentation, or documentation the agent cannot interpret. Fix the root cause, not just the individual decision.
- Establish a monthly review cadence for your most critical knowledge documents. Assign a specific owner for each document. During each review, the owner checks: have any processes changed? Have new exceptions emerged? Are there decision rules that no longer match current practice? Update the documents and notify the team of changes. Track how many updates each review produces. If a document consistently produces zero updates, either the workflow is unusually stable or the reviewer is not looking carefully enough.
Mastered Operate at the highest level.
- Create a structured knowledge capture program that runs continuously rather than as a one-time project. Train 2-3 people on your elicitation methods and assign each one a set of workflows to monitor. They should conduct monthly spot-checks by observing the workflow and comparing observed practice to documented knowledge. When they find gaps, they update the documentation and flag the change for agent configuration review. This distributed approach scales knowledge capture beyond what a single person can maintain.
- Build a feedback mechanism where agent decision outcomes feed back into tacit knowledge documents. When an agent handles a case and the outcome is reviewed by a human, capture whether the human agreed with the agent's decision. When disagreements cluster around specific types of cases, those clusters identify tacit knowledge gaps. Use these patterns to target your next elicitation sessions rather than trying to capture everything at once.
- Develop metrics for tacit knowledge coverage. Track: percentage of agent-handled workflows with documented exceptions, number of decision rules per workflow, recency of last documentation review, and agent-human agreement rate on sampled decisions. Report these metrics quarterly to leadership to make the investment in knowledge codification visible and to identify workflows where coverage is dangerously low.
Unlock Skill Progression
Coaching Personalized to your current level
Progress Tracking Across every skill area
Mastery Validation Evidence-based, not guesswork