--- name: training_needs_analysis description: Run a full training needs analysis: define the performance gap, interview sponsors and performers, pull performance data, measure skills gaps on a fixed scale, diagnose whether the cause needs training or a non-training fix, rank gaps by priority, and produce a costed, owned recommendation report. Use when someone asks for a training needs analysis or assessment, a skills gap analysis, help deciding what training is actually needed, or a diagnosis of why a team is underperforming before training budget gets spent. --- # Training Needs Analysis ## Purpose Find out what performance is required, what performance exists, what causes the difference, and what to do about it, in that order. The output is a prioritized set of training and non-training recommendations tied to a business outcome, with owners, costs, and a way to check whether the fix worked. A list of course titles is not an analysis. The discipline that matters most: training is one possible fix, and it is checked after the cheaper causes. A request for training is a symptom report, not a diagnosis. ## Workflow ### 1. Frame the problem before collecting anything - Get the sponsor to name the business outcome that is off target, by how much, and since when. Write the performance gap in observable terms: "Reports take 3 reviews to pass; the standard is 1" beats "report quality is poor." - Apply the worth test: what does the gap cost per month in rework, delay, errors, lost sales, or risk? If nobody can name a cost and the gap is small, say so and stop. Not every gap is worth an analysis. - Agree the scope in writing: which roles, which tasks, which sites, and the decision date. Note what is out of scope. - Ask the sponsor the hard question early: if the analysis finds training is not the fix, will they act on that? Record the answer. ### 2. Analyze at three levels Work top down so individual plans only get built for gaps that matter. - **Organization:** Which goals, strategies, and upcoming changes (new systems, new products, new rules) create new performance requirements? Where should effort go first? - **Task:** For each role in scope, list the tasks that carry the outcome, the standard for each task, and the knowledge, skills, and attitudes each task requires. Use job documentation plus a top performer, because documents alone miss the judgment calls. - **Person:** Compare each person or group against the task standard. Decide who needs what, and who needs nothing. Sending everyone to training when a third of the team needs it wastes money and signals that the analysis was decoration. ### 3. Collect evidence from at least two source types Use the guides in `references/interview-guides.md`. - Sponsor interview: outcome, cost of the gap, what changed, what was already tried. - Performer interviews: walk the task step by step, find where it breaks, ask for the last time it went wrong. - Top performer interview: extract the real standard and a fast way to test for it. - Observation against a checklist built from the standard. - Performance data pulled before interviews where possible: error and rework rates, cycle time, complaints, audit findings, change dates lined up against the trend. Rule: one source is a claim, two agreeing sources are a finding. Where sources conflict, report the conflict. ### 4. Measure the gaps Use the scale and evidence hierarchy in `references/gap-measurement.md`. - Rate required and current proficiency 0 to 5 per skill, same anchors for both, current rated from the strongest evidence available. - Score and rank every gap with the tool: ```bash python3 ~/workspace/skills/training-needs-analysis/bin/gap_tool.py --sample > gaps.csv # edit gaps.csv with your measured rows, then: python3 ~/workspace/skills/training-needs-analysis/bin/gap_tool.py gaps.csv python3 ~/workspace/skills/training-needs-analysis/bin/gap_tool.py gaps.csv --format json ``` - The tool ranks by gap size, criticality, and people affected, flags the priority band, and routes each gap to training, a non-training fix, or diagnose-first based on the cause label you supply. It does not invent the cause. That comes from step 5. ### 5. Diagnose the cause of each gap Use `references/cause-diagnosis.md`. For each measured gap: - Run the skill test: could they do it if their job depended on it right now? Did they do it before? Is it done often enough to stay fluent? - Check the environment before the person: information and feedback, tools and resources, incentives and consequences. Then knowledge, capacity, motives, in that order. - Label each gap with one cause: knowledge, information, tools, incentives, capacity, motives, process, or unknown. Unknown is honest. It gets a named owner and a date for finishing the diagnosis, not a guess. - Feed the labels back into the gap tool so the ranking and the routing agree. ### 6. Prioritize and build the recommendation set Rank on four factors, and show your working: - **Risk:** compliance, safety, or legal exposure jumps the queue regardless of score. - **Impact:** size of the gap times criticality times people affected (the tool score). - **Cost of delay:** what another quarter of the gap costs. - **Cost and speed of the fix:** a job aid shipped this month can outrank a course shipped next quarter for the same gap. For each priority gap, recommend one primary fix: training, job aid or checklist, feedback loop, tools or access fix, process redesign, incentive change, reassignment, hiring, or a combination with the order spelled out. Non-training fixes that unblock performance come before training on the same gap. For training recommendations, specify: audience (named roles, not "all staff"), the task standard the training must produce, format matched to the gap (practice-heavy for skills, reference material for rare tasks), and how proficiency will be checked at the end. ### 7. Plan the check before anyone builds anything Decide now how each fix will be judged, using the Kirkpatrick levels as a menu, not a mandate: - Level 1 Reaction and Level 2 Learning: cheap, useful for delivery quality and knowledge checks. Never report these as proof of impact. - Level 3 Behavior: is the work itself different 60 to 90 days later? Measure with the same evidence type used in step 4. - Level 4 Results: did the business metric from step 1 move? - ROI (Phillips Level 5): reserve for expensive or contested programs. Convert the Level 4 impact to money, isolate the program effect from other changes as honestly as the data allows, tabulate fully loaded costs, and compute ROI (%) = net program benefits / program costs x 100. Proportionality rule: the cost of measuring should stay well under the cost of the fix. Most fixes get Levels 1 to 3 and a re-measured gap score. That re-measurement is the core proof: gap before, gap after, same scale. ### 8. Write the report and get decisions Deliver the report (see Output Contract), walk the sponsor through it, and get an explicit decision per recommendation: accept, modify, or reject, with the reason recorded. A report nobody decides on is a cost, not an analysis. ## Output Contract The deliverable is one report with these sections: 1. **Summary:** the gap in one paragraph, the top 3 recommendations, total estimated cost, and the decision needed. 2. **Problem framing:** business outcome, current vs required performance, cost of the gap, scope. 3. **Method:** sources used, who was interviewed, what data was pulled, dates, and the rating scale used. 4. **Findings:** measured gaps in a table (skill, required, current, gap, people affected, evidence source), plus findings that are not skills gaps. 5. **Cause diagnosis:** one cause label per gap, with the test or evidence behind each label. 6. **Recommendations:** ranked table with fix type, audience, owner, estimated cost, timeline, and the metric that will show it worked. 7. **Non-training fixes:** listed separately and scheduled before dependent training. 8. **Evaluation plan:** which level per fix, when re-measurement happens, who owns it. 9. **Risks and open questions:** unknowns, conflicting evidence, and what would change the recommendations. Supporting files delivered with the report: the scored gaps CSV and the tool output, so the sponsor can re-rank with different criticality ratings and see what changes. ## Operating Rules - Never recommend training for a gap whose cause has not been diagnosed. If time forces a recommendation anyway, label it provisional and name the diagnostic step still owed. - Never invent numbers. Every figure in the report traces to a source the report names: a metric, a work sample, an interview, or a rating with a named rater. Unverifiable claims get cut or labeled as opinion. - Self-assessment alone never proves a gap. Pair it with work evidence or label the finding as perceived need. - Protect interviewees. Report performer findings in aggregate or by role. A needs analysis that gets people punished for honesty will never get honest data again. - Watch for the classic failure modes: the request names a course before a gap; the same training ran before and the gap did not move; trained and untrained staff show the same gap; the gap dates to a system or process change. Each one points away from training as the first fix. - Re-run the analysis on a cycle: after a major fix lands, after a system or strategy change, and at least annually for roles tied to compliance or safety. - Keep the style plain in everything written for stakeholders. State findings as facts with sources, recommendations as decisions with owners, and uncertainty as uncertainty.