Key figure hierarchy

Key figure hierarchy

The hierarchy of indicators connects the strategic goal of the company with the daily decisions of the teams. This is a managed "tree" from North Star Metric (NSM) and business totals to drivers, operational KPIs and guardrails (security restrictions). Below is how to design, document and operate such a hierarchy.

1) Why hierarchy is needed

Single focus: all teams pull in one direction, metric conflicts are minimal.
Transparent causality: it is clear why NSM moves (driver tree).
Speed of decisions: local metrics are directly related to the business effect.

Sustainability: guardrails prevent optimization "side effects."

2) Layers and terms

North Star Metric (NSM): main value metric (e.g. "Monthly Active Paying Users").
Business results: revenue, margin, GGR, market share, NPS - associated, but not always NSM.
Driver tree: NSM decomposition into factors: volume × conversion × frequency × average check, etc.
Team KPIs: indicators that are actually managed by the team (funnels, latency, PR-AUC, etc.).
Guardrails: restrictions: FPR≤x%, p95 latency≤y, the share of zhalob≤z - "brakes" with growth.
Process metrics: release speed, data quality, SLO updates - support the system.

3) How to choose North Star

Criteria: reflects value for the user, correlates with long-term income, is sensitive to product changes, understandable and resistant to manipulation.

• B2C product: 'WAU of payers' or 'Sessions with successful target activity/month'.
• Marketplace: 'Completed successful deals/month'.
• ML platform: 'Percentage of requests served by model with SLA ≤ X ms and calibration in tolerance'.

4) Driver Tree

1. Define the NSM formula.
2. Decompose into multiplicative/additive factors.
3. Add a layer of steerable arms.
4. Link sheet metal metrics to commands.
5. For each node - owner and target levels.

NSM: Активные платящие (MAPU)
= Активные пользователи × Доля платящих × Средняя частота платежей × Средний чек
├─ Активные пользователи = Трафик × Активация × Удержание
│  ├─ Трафик = Органический + Платный (CR кампаний, CAC)
│  ├─ Активация = Конверсия онбординга (шаги воронки)
│  └─ Удержание = D7/D30, когорты
├─ Доля платящих = Propensity-модели + Офферы (guardrail: RG/фрод)
├─ Частота = Миссии/квесты, контент-ротация
└─ Средний чек = Прайсинг, бандлы (guardrail: жалобы/возвраты)

5) Target cascading (OKR/KPI)

Enterprise level: NSM + 3-5 drivers (year/quarter).
Features/Products: driver subtrees (target Δ).
Commands: KPI on tree sheets; target setting in OKR (Commit/Stretch) format.
Individually: task metrics (through contribution to team KPI).

• One level - ≤7 metrics (overload kills focus).
• All targets are bound to a unique tree node (without duplication).
• For conflicting couples, explicit priority and/or compromise through guardrails.

6) Guardrails: how to set "sides"

• Quality/experience: p95 latency, crash-free%, fault tolerance.
• Ethics/compliance: anti-fraud FPR, share of complaints, RG indicators.
• Data: Freshness, Completeness, PSI drift.
• Finance: margin ≥ X, chargeback rate ≤ Y.

Guardrails - mandatory line in each OKR/metric passport; triggering blocks rollout.

7) Formulas, sources and semantic layer

All tree metrics are defined in the semantic layer (single definitions, rollups, calendar, currencies/timezones).
Versioning: 'METRIC _ NAME _ vN'; any formula edit = new version + backfill/reconciliation.
Passport metrics (short): code, definition, formula/SQL, sources, granularity, default segments, units/currency, SLO freshness and availability, owners, guardrails, change history.

8) Measurement frequencies and SLO

Operational metrics: min/h (SLO lag ≤ 10-15 min).
Product/Marketing: Day/Week (Mandatory MA/Seasonal Normalization).
Financial: week/month (reconciliations with accounting).
ML production metrics: stream + deferred offline assessments (OOT).

For each metric on the dashboard, show: "updated X min back" and planned SLO.

9) Dashboards and reviews

Executive layer: NSM + key drivers, alert-widgets guardrails.
Domain layer: funnels, cohorts, contribution of factors (waterfalls/decompositions).
Ops/ML layer: SLA, errors, drift, calibration, latency p95/p99.
Rituals: daily (operating system), weekly (tactics), monthly/QBR (strategy/OKR).
On reviews, use the tree: show the node, deviation, hypotheses, plan.

10) Responsibility Mapping (RACI)

• Responsible: KPI owning team.
• Accountable: Domain/Product Leader.
• Consulted: analytics/data/compliance.
• Informed: related teams/top management.

Scorecard conflicts are resolved at the shared tree level with Accountable.

11) Examples of "mini-trees" by domain

• 'Attracted quality users = Traffic × Share of qualified leads × CR in registration × Brand share'
• Guardrails: SAS≤targeta, Brand Safety, spam rate.

• 'WAU = DAU × Stickiness × frequency'
• Guardrails: crash-free ≥ 99. 5%, complaints ≤ threshold.

• 'Revenue = Active Paying × Payment Frequency × Average Check'
• Guardrails: refund%, chargeback rate, RG indicators.

• 'Business effect = Overlap loss − False lock costs'
• Key KPIs: PR-AUC, Recall@FPR≤x%, Coverage; Guardrails: latency p95, proportion of appeals.

12) Anti-patterns

Inconsistent definitions: "Retention" is considered differently - treated with a semantic layer and versions.
Proxy Optimization: Click Growth without NSM Growth - Protect via NSM and guardrails.
Too many metrics: "airplane panel" - leave ≤7 on the level.
No owner: The "draw" metric is rapidly degrading.
Quiet formula edits: always - new version + changelog.

13) Artifact patterns

A. Tree Node Passport

Node: 'RETENTION _ D30'

Tree Role NSM Driver

Formula (semantic object): reference/code

Segments: Country/Channel/Platform

Owner: Product Analytics

Goals (quarter): Commit/Stretch

Guardrails: churn complaints ≤ X, data freshness ≤ 1 h

Risks/assumptions: seasonality, traffic mix

Date/Version: 2025-Q4, v2

B. Target cascade map (example)

NSM (company): MAPU + 8% YoY

Product: Retention D30 + 2 p.p.

Team Onboarding: Activated registration + 5 pp

Marketing: Quality leads + 12% at CAC ≤ target

Monetization: ARPPU + 6% at refund% ≤ threshold

C. Deviation runbook

Event DAU drawdown − 10% DoD

Actions: check of freshness of data → incidents of releases → channels of traffic → regions/platform → RCA and plan

14) Hierarchy implementation checklist

• NSM defined and its relationship to value/revenue
• Built and documented driver tree (owners, formulas, sources)
• For each node - targets, guardrails, refresh rate and SLO
• Metrics are established in the semantic layer, versions and dictionary of terms are consistent
• Dashboards by layers and review rituals (daily/weekly/monthly/QBR)
• RACI is fixed; KPI conflicts and priorities described
• Runbooks for type deviations; alerts on NSM/guardrails
• Tree revision plan once a quarter; removing out-of-date nodes

Result

The hierarchy of indicators is the NSM → driver tree → command KPIs → guardrails, fixed in the semantic layer and connected to the control rhythms. This design makes goals transparent, responsibility explicit, and change measurable and controllable.