EPI cheat sheet

Governance triad: a steering committee sets priorities and funds, an improvement office (PMO/CI office) manages the portfolio and methods, and process owners are accountable for results in their own value streams.

Strategy deployment (Hoshin Kanri) cascades 3-5 annual breakthrough objectives into divisional targets, then into daily management metrics; each level has one owner, one metric, and one review cadence.

Portfolio triage rule: rank candidate projects by (impact on strategic goal x probability of success) / resource demand; fund the top tranche and hold the remainder in a backlog with defined entry triggers.

Project charters must state problem, baseline metric, target, scope boundaries, sponsor, team, milestone dates, and resource commitment; a charter without a quantified baseline is not approvable.

Governance review cadence: weekly team stand-up, monthly sponsor review of metric trajectory, quarterly portfolio review for rebalancing and benefit realization, annual strategy refresh.

Benefit realization is tracked 6-12 months after handoff; a project is not closed until the control plan owner signs acceptance and the metric has held for at least three reporting periods.

Maturity models (e.g., five-level capability stages from ad hoc to optimized) are used to set the improvement roadmap, not to score individuals; assess both capability and deployment breadth.

Decision rights matrix (RACI or DACI) prevents the common failure of an unowned metric: every KPI needs exactly one accountable owner even when many contribute.

Measure definitions must specify numerator, denominator, data source, inclusion/exclusion rules, calculation frequency, and owner; undocumented definitions are the leading cause of disputed dashboards.

Operational definition test: two trained observers measuring the same event should agree; inter-rater reliability below roughly 90% agreement signals a definition or training problem, not a performance problem.

Run charts: a shift is 6 or more consecutive points on one side of the median; a trend is 5 or more consecutive points all rising or falling; astronomical points and cycles are also non-random signals.

Control charts: choose attribute charts (p, np, c, u) for count/proportion data and variable charts (X-mR, Xbar-R, Xbar-S) for continuous data; subgroup size and constant opportunity determine the correct chart.

Control limits are computed from the data (typically mean plus or minus 3 sigma) and are not specification limits or targets; specification limits belong on a capability study, not a control chart.

Common cause variation requires process redesign; special cause variation requires investigation of the specific event. Reacting to common cause as if special (tampering) increases variation.

Balanced measurement families: outcome (patient/client result), process (cycle time, first-pass yield), and balancing (unintended consequences such as cost, access, or staff workload) measures.

Sampling decisions trade precision for cost: larger samples narrow confidence intervals by the square root of n, so quadrupling sample size only halves the margin of error.

SIPOC defines Suppliers, Inputs, Process, Outputs, Customers at a high level; complete it before detailed mapping to fix scope and prevent mapping a process that has no clear customer or output.

Value stream mapping distinguishes processing time from lead time; flow efficiency = value-added time / total lead time, and typical transactional processes run below 10%.

Eight wastes mnemonic (TIMWOODS): Transport, Inventory, Motion, Waiting, Overproduction, Overprocessing, Defects, Skills underuse.

Little's Law: average work in process equals arrival rate multiplied by average cycle time (WIP = throughput x cycle time); use it to predict the queue effect of adding work or cutting capacity.

Queueing rule of thumb: as utilization approaches 100%, wait time grows nonlinearly; systems above roughly 80-85% utilization experience sharply rising delays and are fragile to variation.

Bottleneck identification: the step with the highest utilization or the longest queue is the constraint; improving a non-constraint does not increase system throughput (Theory of Constraints).

Root cause tools: 5 Whys for single-path causal chains, fishbone for divergent cause categories (people, method, machine, material, measurement, environment), Pareto to rank causes by frequency or cost.

Redesign sequence: eliminate non-value-added steps, then simplify, then combine, then automate last; automating a wasteful step locks in the waste.

PDSA (Plan-Do-Study-Act) is for rapid small-scale tests of change; DMAIC is for problems needing data-driven root cause analysis and sustained gains; Kaizen events compress focused improvement into 3-5 days.

DMAIC phases: Define (charter, voice of customer), Measure (baseline, measurement system validation), Analyze (root cause, hypothesis tests), Improve (pilots, designed experiments), Control (control plan, monitoring).

Lean tools: 5S for workplace organization, kanban for pull replenishment, standard work for documented best method, poka-yoke for error-proofing, SMED for setup reduction.

Six Sigma logic: a process at six sigma has 3.4 defects per million opportunities; sigma level is derived from defects per opportunity and assumes a 1.5 sigma long-term shift in common practice.

Measurement system analysis precedes capability analysis; if repeatability and reproducibility (gage R&R) consume more than about 10% of tolerance, fix the measurement system before drawing conclusions.

Process capability: Cp compares specification width to process spread (6 sigma); Cpk additionally accounts for centering. Cpk below 1.0 means the process produces out-of-specification output.

FMEA scores severity, occurrence, and detection on 1-10 scales; risk priority number = S x O x D. Prioritize high severity regardless of RPN, since severity cannot be reduced by detection controls.

Improvement sustainment requires standard work, visual management, daily tiered huddles, and an escalation path; gains decay when the control plan lacks an owner and a review cadence.

Kotter's eight steps: urgency, guiding coalition, vision, communicate, remove barriers, short-term wins, build on gains, anchor in culture; skipping urgency is the most common failure point.

ADKAR is an individual change model (Awareness, Desire, Knowledge, Ability, Reinforcement); change succeeds only when each person passes all five, and the binding constraint is usually Desire, not Knowledge.

Lewin's model: unfreeze, change, refreeze. Refreezing corresponds to embedding new standard work, orientation content, job descriptions, and performance metrics.

Readiness assessment covers sponsor commitment, capacity of affected staff, prior change fatigue, union or regulatory constraints, and clarity of the case for change.

Resistance is diagnostic: identify whether it stems from lack of information, lack of capability, loss of status, or genuine technical disagreement; each requires a different response.

Culture is revealed by what gets measured, rewarded, and tolerated; changing the metric and reward system changes behavior faster than communication campaigns alone.

Diffusion of innovation: innovators and early adopters (~16%) adopt without proof; the early majority (~34%) needs peer evidence; target respected frontline opinion leaders rather than formal managers alone.

Communication planning rule: repeat key messages across at least three channels and seven touches; senders systematically overestimate how much of their message was received and understood.

Analytics ladder: descriptive (what happened), diagnostic (why), predictive (what will happen), prescriptive (what to do); most organizations stall at descriptive reporting.

Data quality dimensions: completeness, accuracy, timeliness, consistency, validity, uniqueness. Profile data against each before using it for decisions.

Statistical test selection: continuous outcome comparing two groups uses t-test; three or more groups use ANOVA; categorical associations use chi-square; non-normal data uses Mann-Whitney or Kruskal-Wallis.

Correlation does not establish causation; causal claims require a design such as randomized assignment, difference-in-differences, interrupted time series, or instrumental variables.

Interrupted time series is the workhorse design for improvement evaluation: it tests level and slope change after intervention while accounting for pre-existing trend.

Statistical process control distinguishes signal from noise; a p-value alone does not tell you whether a change is large enough to matter clinically or operationally.

Dashboards should follow a hierarchy: strategic (monthly, executive), tactical (weekly, manager), operational (daily, frontline), each with thresholds and a defined action when a threshold is breached.

Predictive models require validation on held-out data and monitoring for drift; report discrimination (e.g., AUC) and calibration, because a well-ranked but miscalibrated model misleads decisions.

Stakeholder analysis plots influence against interest: high influence/high interest are manage closely, high influence/low interest keep satisfied, low influence/high interest keep informed, low/low monitor.

Voice of customer methods: interviews, focus groups, surveys, complaint and grievance logs, shadowing, and observation of actual work; each has different bias profiles.

Survey design rules: avoid leading and double-barreled items, keep scales consistent, pre-test, and report response rate and nonresponse bias alongside results.

Closing the loop: inform participants what was heard and what changed; unreported engagement erodes future participation more than not engaging at all.

Communication structure for updates: situation, background, assessment, recommendation (SBAR) or headline-first pyramid; lead with the decision needed and the ask.

Meeting discipline: published purpose and decision rights, pre-reads, timeboxed agenda, documented decisions and owners with dates; meetings without decision rights become information sessions.

Engagement with frontline staff requires protected time and leadership presence; asking for input without authority to act on it is a recognized driver of cynicism.

Escalation path must be explicit and time-bound: define who resolves conflicts between process owners, and at what threshold an issue moves to the steering committee.

Improvement work involving human subjects (research intent, novel interventions, publication) requires ethics review; quality improvement intended for local learning typically follows a separate determination process.

Privacy safeguards: minimum necessary data, de-identification before analysis where possible, role-based access, audit logs, and a documented retention and destruction schedule.

Informed consent applies when participation is voluntary and research-oriented; withholding standard care or randomizing patients requires consent and ethics approval.

Risk register entries need likelihood, impact, owner, mitigation, and review date; residual risk must be accepted by a named accountable leader, not by the project team.

Compliance obligations commonly include billing and coding integrity, anti-kickback and conflict-of-interest disclosure, safety event reporting, and mandatory regulatory reporting timelines.

Equity check: stratify outcome metrics by demographic and access subgroups; an improvement that raises the average while widening a gap is a partial failure.

Whistleblower and non-retaliation protections require a confidential reporting channel and a documented investigation process independent of the implicated manager.

Algorithmic and AI tools used in decisions require bias testing, explainability appropriate to the stakes, human review of adverse decisions, and periodic revalidation.

Placeholder bullet 1: governance charters define decision rights, quorum, escalation thresholds, and conflict-of-interest recusal rules for committee members.

Placeholder bullet 2: portfolio capacity planning uses available trained facilitator hours per quarter as the binding constraint on project intake.

Placeholder bullet 3: stage-gate reviews at charter, baseline, root cause, pilot, and control phases allow early termination of non-viable projects.

Placeholder bullet 4: benefit tracking distinguishes hard savings (verified cost removal) from soft savings (avoided cost, capacity released) and reports them separately.

Placeholder bullet 5: project closure requires updated standard work, training records, control chart ownership, and transfer of monitoring to the operational dashboard.

Placeholder bullet 6: strategic alignment scoring uses weighted criteria agreed before scoring to prevent sponsor-driven selection bias.