KPI Hierarchies and Cascades
Board number to team dial without drift
KPIs nest: corporate metrics decompose into divisional, team and initiative metrics. Cascades keep line-of-sight so a widget in a team equalizes to the board's number without drift.
What you will learn
- Decompose a corporate KPI into contributory ones
- Describe a cascade and its reconciliation property
- Avoid the 'cascade equals same-KPI-shrunken' trap
What is this?
KPI hierarchies organize metrics from board-level outcomes down to team-level contributions. A cascade is that chain made explicit: corporate revenue decomposes (not shrinks) into regional revenue, which decomposes into win rate × pipeline × ASP. The design rule: each level's total must reconcile to its parent, and lower levels carry genuinely different-owner metrics, not tinier copies.
Why do you need it?
- Teams act on team-level inputs, not corporate outcomes.
- Reconciling cascades prevent 'my number disagrees with yours'.
- Line-of-sight keeps individual work visibly tied to company goals.
- Hierarchy design is a senior BI analyst's deliverable.
Mental model
Accounts' parent-and-child ledger
KPI hierarchy
The income statement rolls up: sales minus COGS = gross profit; subdivisions must total to the family totals or the books don't balance. A KPI cascade is the same bookkeeping spirit — each level balances to its parent, and the totals agree at every rung.
How it works
- Step down by denominator: Revenue = Orders × AOV; compose the chain.
- Reconciliation: parent equals the sum/integral of its children at the right grain.
- Different-owned: child metrics belong to distinct teams (win rate ≠ revenue).
- Cadence cascade: corporate monthly, divisional weekly, teams daily.
- Line-of-sight: each level shows its parent, itself, and its delayers.
Corporate → division → team chain.
Examples
A decomposable revenue hierarchy
L0 Revenue = SUM(region revenue)L1 Region Revenue = Σ units × price (each region)L2 Regional drivers = win rate × pipeline × ASPL3 Team dials = leads, win %, deal size, velocity Reconciliation checks in CI: L1 sums == L0 (per period/region) L2 product ≈ L1 (within tolerance) each L3 owned by a named team No level repeats the parent; every level changes behavior.The cascade decomposes rather than copies, reconciles at every rung, and hands a distinct dial to each owner.
In production
- A regional P&L cascade: corporate revenue = Σ region; region = Σ segment; the sums are CI-checked.
- A churn cascade: corporate churn → per-cohort churn → engagement dials per team.
- A delivery cascade: SLA comply rolls from team to platform, reconciled weekly.
Best practices
- Decompose by denominators (Orders × AOV), not shrink-copies.
- CI-check reconciliation at every level of the cascade.
- Give each rung a distinct owner-appropriate set of dials.
- Vary cadence down the tree: longer at the top, tighter at the teams.
Common mistakes
- Re-slicing the same KPI smaller instead of decomposing by cause.
- No reconciliation check — levels drift apart silently.
- Cascades that own nothing new at the team level.
- Corp KPI changing the middle level's truth (grain chaos).
Debugging tips
- Team dials move, corp number doesn't → the decomposition is mythical; verify the chain math.
- Two regions sum ≠ corp → grain mismatch; fix the L0 def before blaming teams.
- Line-of-sight lost → cadence/rung mapping unclear; label parent–child links.
Interview checkpoint
Quick quiz
A good cascade decomposes a parent KPI by:
Cascade levels must reconcile to their parents.