Full SKILL.md
KPI Architecture
Builds a strategy's measurement layer — financial anchor and value gap, Kaplan-Norton strategy map with linkage hypotheses, a 20-to-25-measure scorecard with lead/lag pairs, owners and definition sheets, a value-realisation ledger with L0–L5 stage gates reconciled to the P&L, and a review cadence with decision rights. Use whenever a plan needs targets, baselines or benefits tracking, or the user says "what KPIs should we track", "build us a scorecard", "how will we know the strategy is working", "the savings never show up in the P&L", "benefits tracking", "strategy map", "define this metric properly", "too many metrics". Not for sizing value at stake before a plan exists (value-driver-modeling) or a product KPI tree and decision gates inside a product strategy (product-strategy, which may call this for definitions and the ledger).
Measures the strategy's causal chain, not what is available, and reports value only when finance has signed it. Kaplan-Norton and the L0–L5 gate model are in references/frameworks.md; read it only when a step needs defending.
When not to use
- No strategy yet:
strategic-choice, then return. - Initiatives undefined (no owners, no charters):
execution-planningfirst. - Sizing value at stake rather than tracking value delivered:
value-driver-modeling. - Product-level KPI tree and gates:
product-strategyowns them; call this skill for definition sheets and the ledger.
Inputs
Ask for, or infer and label: the chosen strategy (proposition, three or four things to excel at, horizon); financial anchor (baseline P&L and dated target); initiative list with owners, sponsors, business-case sizes, current stage; what is measured today and which systems hold order, customer, cost and HR data; governance in place (transformation office, finance validator, existing forums); audience and frequency; and the thing rarely volunteered: any measure currently tied to bonus.
Ask one round of questions if the anchor or the initiative list is missing. Otherwise proceed on a plausible labelled baseline and a small transformation office with a part-time controller.
Workflow
-
Financial anchor and gap. Four lines (revenue, gross margin, opex, EBITDA) for baseline and target year; the gap as one number with a date; the bridge (initiative run-rate, organic, inflation, headwinds). Take the
value-driver-modelingbaseline if it exists. Freeze the baseline: period, source, sign-off, named finance owner. Test: if every initiative delivers at business-case value, does the bridge reach the target? Say so now. -
Strategy map (
references/templates.md§2). Top-down: financial objectives split productivity vs growth, saying which leads; customer proposition in customer words with two to four outcomes; three to five internal processes as outcomes not projects (nine means the strategy is unchosen); learning-and-growth foundations with specific gaps (human, information, organisation capital). Arrows last, only where an executive will state the if-then; cut the first draft by half; about 15. Three or four vertical themes. Draw it with executives in the room where possible. Test: pick any arrow; who would be embarrassed if it proved false? -
Linkage hypotheses (
references/templates.md§3). Per arrow: "if <lead> reaches <x>, then <lag> reaches <y> within <n quarters>". Lead at the tail, lag at the head; the same measure can be lag for one arrow and lead for the next. The lead must be observable at least one review period before the lag can move. Test: for the highest-value arrow, what number in two quarters would make us abandon it? -
Select measures. Candidates are the Stage 3 pairs; dashboard measures join only if they sit on an arrow. Rules in order, rejections recorded: few (20–25, four to seven per perspective); causal (on an objective); owned (one person); actionable (owner can move it in the review period with resources they control); definable; balanced (each lag has a lead; each financial outcome has a health measure beside it); gaming-resistant (ask the cheapest way to hit it without the real work, add the exposing pair). Sort into tiers: milestone (work done; initiative plan, not scorecard), operational KPI (behaviour changed; leading evidence), value (money in the P&L; ledger). Retirement rule for measures that will read 100% or zero, successor named. Test: cover the names; can every owner say unprompted why their number is there?
-
Definition sheets (
references/templates.md§5). Per measure: formula precise enough to code (numerator, denominator, period, inclusions, exclusions); source with join keys and versioning; capture and review frequency; owner and data steward; baseline with period and sample; dated targets; pairing measure; ledger link; double-count rule where two initiatives touch one P&L line, written on both sides; known limitations; change control. Test: an analyst who has never met the owner computes the same number. -
Stage gates and ledger (
references/templates.md§6–7). L0 idea logged (TO); L1 validated with owner, mechanism, baseline source (owner and TO); L2 business case with frozen baseline, run-rate, phasing, one-off costs, double-count check, lead measures (finance); L3 in execution with dated milestones (TO and sponsor); L4 implemented, milestones closed, leads near target (TO); L5 realised in management accounts against the frozen baseline, sustained three months, budget adjusted so the money cannot be spent twice (CFO). Realised means L5 only; everything earlier is assessed, early or forecast, stage split always shown; stages move backwards with dates. Ledger: one row per boulder, one aggregate for sand below a threshold; columns owner, validator, stage, run-rate target, in-year target, YTD in P&L, forecast, variance, one-off costs, next gate and date, last stage move. Beneath: YTD split by stage, headline on L5 only; pipeline by stage with coverage (L2+ run-rate over required value) interpreted against attrition, refill action on the same page; one line of arithmetic per boulder (volume × unit effect × realisation factor). Test: show the headline, then how much a sceptical CFO would sign today. -
Reconcile to the P&L (
references/templates.md§8). Frozen baseline EBITDA, plus in-year ledger value, minus one-off costs on their own line, plus or minus headwinds and tailwinds independent of the programme, equals forecast. Compare with FP&A's independent forecast; trace differences to causes. Test: is the no-programme EBITDA written down anywhere? -
Cadence and hypothesis-test agenda (
references/templates.md§9). Forums with decision rights: weekly TO review (chief transformation officer: stage moves to L4, unblocking, owner changes); monthly value review after close (CFO: L5 sign-off, baseline restatements, re-phasing); quarterly scorecard and strategy review (CEO: confirm or revise hypotheses, add or retire measures, reallocate); quarterly board value page; annual map refresh. Quarterly agenda per arrow due: did the lead move, did the lag follow in time, and if not is it execution (lead did not move) or hypothesis (lead moved, lag did not)? Board page, one page: action title, value bars by stage, coverage and refill, four to six scorecard headlines, hypotheses tested, decisions needed. Test: name one decision the monthly value review took or would take.
Output
The gap, coverage and L5-realised headline in the first 150 words. Then, each once, 2,000 to 3,500 words:
- Financial anchor and bridge
- Strategy map (text form) with themes
- Linkage hypothesis table
- Scorecard table (ref, objective, measure, tier, owner, baseline, targets, frequency) and rejected-measures list
- Definition sheets
- Stage-gate criteria and value-realisation ledger with stage split, pipeline and coverage
- Ledger-to-P&L reconciliation
- Cadence table and quarterly agenda
- Board value page
No method vocabulary, framework authors, skill names or the worked example's client in the deliverable. Full example: references/worked-example.md; xlsx renders ledger and sheets.
Quality bar
Run silently; fix, do not report.
- Gap is one number with a date; bridge adds up; baseline frozen with a named owner.
- Every objective is an outcome; every arrow has a stated if-then and an owner.
- Every arrow has a lead/lag pair with a lag time.
- 20–25 measures; every measure on an objective, with one owner and a pairing measure where gameable.
- No milestone metrics on the scorecard.
- Every measure has a definition sheet; double-count rules are on both sides.
- Headline value is L5 only; stage subtotals sum; every boulder has its arithmetic.
- Coverage is stated and interpreted, with a refill action.
- Ledger reconciles to the P&L and to FP&A, differences explained.
- Every forum has a decision right.
- No leaked method vocabulary.
Anti-patterns
- Measuring what the dashboard already has instead of what sits on an arrow.
- Thirty arrows; a map presented finished by advisors.
- Activity metrics on the scorecard as evidence of value.
- Committee owners; measures the owner cannot move.
- Value reported "delivered" at L3 or L4.
- A ledger that only moves forward; a re-baselined baseline.
- One-off costs netted invisibly; the programme credited with market tailwinds.
- Status meetings called reviews.