L&D in Healthcare: A Playbook for Better Outcomes
Make the learning function accountable for workforce performance: diagnose the function honestly, connect every investment to an outcome, and report contribution in the board's language.
Ravinder Tulsiani, DBA · Enterprise Capability Executive
Executive abstract
Your organization spends serious money on learning: mandatory modules, onboarding, competency checks, leadership programs, and the hours clinicians spend away from patients to complete it all. What did any of it produce? In most healthcare organizations, that question has no good answer. This playbook describes what a learning function built around outcomes looks like: diagnose the function before fixing it, start every investment from the outcome it must move, rebuild the operating model around performance rather than course delivery, prove contribution with honest evidence, sequence the first twelve months, and give artificial intelligence a defined, governed place.
The central argument
The problem is not that healthcare L&D lacks effort or professionalism. It is that the function is managed as a training producer when it should be managed as a performance function. Completion rates, hours delivered, and satisfaction scores cannot answer the question executives actually ask: did the investment change workforce performance? Until the function can answer that, with baselines, comparison points, and stated limitations, it cannot defend its budget, stop weak work, or earn the board's confidence.
Key findings
Mandatory education crowds out capability building.
Required modules consume the largest share of staff learning time, while the clinical and role-based education that actually changes performance is smaller, funded last, and measured least.
Training is prescribed for non-training problems.
Staffing gaps, broken tools, unclear accountability, and weak management are routinely answered with courses. A function that cannot say "do not train" keeps solving the wrong problem.
Completion is reported as if it were impact.
Mandatory modules record completions, not behavior change. Without transfer measurement, the function cannot distinguish learning that worked from learning that merely happened.
Governance decides what the function is.
Whether the learning council can approve, condition, redirect, or discontinue investments on evidence determines whether the function is managed or merely funded.
AI needs a decision filter, not a strategy.
Every proposed use of AI in learning should be placed on a spectrum: where it adds real value, where a simpler solution wins, and where risk outweighs benefit. Pilots stay small and governed, with clear kill criteria.
What the evidence supports
- The College of Nurses of Ontario's Quality Assurance program requires nurses to reflect on practice annually, identify learning needs, and complete learning goals under the Regulated Health Professions Act: the professional accountability the playbook builds on.
- Accreditation Canada's Qmentum program has moved Required Operational Practices toward measurable safety improvement, including patient-safety education and training as a required practice.
- Canadian Institute for Health Information data links nurse staffing levels to hospital harm, including pressure-ulcer rates: the clinical anchor for the playbook's worked example.
- Statistics Canada job-vacancy data and Ontario's Fixing Long-Term Care Act (O. Reg. 246/22) establish the workforce-shortage and mandatory-training context the diagnosis starts from.
- Ontario's Information and Privacy Commissioner (AI scribes guidance), the federal Privacy Commissioner (AI principles), and the CPSO (AI in clinical practice) frame the governance expectations applied to AI in Chapter 6.
What this paper adds
The concepts and decision principles developed by Ravinder across twenty-plus years leading learning functions, including eight and a half years leading learning and development at one of Canada's largest healthcare organizations. Practitioner synthesis, distinct from the external evidence cited above.
- The CLEAR diagnosis model: Clarify success, Locate current performance, Explore causes, Assess transfer, Recommend interventions.
- The Domino Map: a seven-step chain from outcome to evidence. Aim, Define, Map, Performance Blueprint, Solution Design / Minimum Effective Dose, Measurement, Scaling.
- The Performance Blueprint: the operating model that rebuilds the function around outcomes, from its reporting line and governance to its roles and daily workflow.
- The IMPACT measurement framework: Identify the Business Problem, Map the Capability Pathway, Pick the Measures That Matter, Assemble Credible Evidence, Communicate Contribution Clearly, Track, Test, and Refine.
- The AI Value Spectrum: a three-zone decision filter for every proposed AI use.
- Four real examples from the author's own record: a governed technology kill decision, a platform switch saving over $400,000 annually in recurring costs, an onboarding redesign associated with turnover roughly halved, and a VR empathy program measured at application rather than reaction.
Models in this paper
1. CLEAR diagnosis.
Protects investment before any redesign by establishing what the function must accomplish, where resources go, whether learning transfers, and which activities warrant protection, improvement, or discontinuation.
2. Domino Map.
Forces every investment to start from a named outcome and trace backward through required performance, capability gaps, and the smallest intervention likely to close them. Produces falsifiable claims and a short list executives can judge like any other investment.
3. Performance Blueprint.
Rebuilds the function's authority, operating practices, and roles around outcomes: a council with real decision rights, a single learner record, measurement owned inside the function, and a catalog managed as inventory rather than protected as territory.
4. IMPACT measurement.
Turns evidence into a quarterly board account in four parts: regulatory assurance, workforce performance, business contribution with limitations stated, and investment decisions with validated benefits.
5. AI Value Spectrum.
Converts AI enthusiasm into governed pilots: every proposal placed where it belongs, with pre-set kill criteria and no scaling without evidence.
Questions for leaders
- What did our learning investment produce last year, stated in workforce performance rather than completions?
- Which of our mandatory modules have evidence of behavior change, and which are maintained on faith?
- When a performance problem appears, who decides whether training is the answer, and whether they can say no?
- What would it take for our board to see the learning function the way it sees any other material investment?
Illustrative examples
Clearly labeled throughout. Real examples from the author's experience are marked as such; worked decision examples and board-report exhibits are hypothetical and labeled hypothetical. No example invents outcomes.
Implications for leaders
- Diagnose before you restructure: a reorganization without a diagnosis rearranges the same unmeasured work.
- Give the learning council real authority to approve, condition, redirect, or discontinue investments. Or stop pretending governance exists.
- Start every material investment from the outcome, and require a falsifiable claim before funding.
- Report contribution with limitations stated; one honest report buys more trust than three glossy ones.
- Keep AI pilots small and governed, with preset kill criteria; scale only on evidence.
Recommended actions
- Commission a CLEAR diagnosis of the learning function within the first ninety days.
- Map two or three priority outcomes through the Domino Map with clinical leadership before approving new programs.
- Charter the learning council with discontinuation authority and a quarterly evidence cadence.
- Replace activity reporting to the board with the four-part IMPACT account.
- Run every AI proposal through the Value Spectrum before any pilot is approved.
Evidence boundaries and limitations
The frameworks in this playbook (CLEAR, Domino Map, Performance Blueprint, IMPACT, AI Value Spectrum) are the author's practitioner models, developed across his career and described in his books. They are not regulatory, accreditation, or standards-body frameworks. External sources support the direction of travel and selected design considerations; they do not prescribe this operating model. Worked examples are hypothetical and labeled as such. Real examples are the author's own account, with limitations stated where they apply.
The complete 40-page playbook is available as a PDF: six chapters, the framework visual, the Domino pathway exhibit, the illustrative board dashboard, and source notes included.