Как метрики связаны между собой?

Взаимосвязь метрик: как они влияют друг на друга

Введение

Метрики редко работают в изоляции. Изменение одной метрики часто влияет на другие. За 10+ лет анализа я выработал понимание типичных взаимосвязей и как их использовать для принятия лучших решений.

Основные типы метрик и их иерархия

1. North Star Metric (главная метрика)

Это одна метрика, которая отражает main business value.

Примеры:

• SaaS: Monthly Recurring Revenue (MRR)
• E-commerce: Gross Merchandise Value (GMV) или Revenue per user
• Marketplace: Total value of transactions
• Social network: Monthly Active Users (MAU) / Daily Active Users (DAU)

Почему North Star важна:

• Когда конфликтуют требования, North Star решает
• Все team должна быть aligned на эту метрику
• Это не operational metric, это business outcome metric

2. Key Result Metrics (как мы достигаем North Star)

North Star = функция нескольких ключевых метрик.

Пример для E-commerce:

Revenue = Users × Conversion Rate × Average Order Value
     ↓           ↓             ↓              ↓
North Star    Traffic      Purchase       Product price
                funnel       quality        mix

Если Revenue растет, это может быть благодаря:

• Больше users? (Growth)
• Выше conversion? (Product quality)
• Выше AOV? (Pricing, upsell)

Диагностировать problems:

• Если Revenue упала, но Users выросли → Problem в conversion или AOV
• Если Revenue упала и Users упали → Problem в acquisition

3. Diagnostic Metrics (что стоит за Key Result)

Для каждой Key Result есть несколько диагностических метрик.

Пример: Conversion Rate = Function of:

• Awareness: Do users know about the feature? (Page views, CTR)
• Clarity: Do users understand what they need to do? (Bounce rate, time-on-page)
• Friction: How easy is it to complete? (Completion rate, errors)
• Motivation: Do users have a reason to complete? (Reviews, social proof)

Если Conversion падает:

• If bounce rate высокий → Problem в clarity
• If CTR высокий но completion низкий → Problem в friction
• If CTR низкий → Problem в motivation или awareness

4. Leading vs Lagging Metrics

Lagging Metrics (результаты)

• Отражают то, что уже произошло
• Примеры: Revenue, Churn, Customer Satisfaction
• Меняются медленно (месяцы, кварталы)
• Полезны для оценки стратегии

Leading Metrics (предсказатели)

• Предсказывают будущие результаты
• Примеры: Feature adoption, NPS, Customer health score
• Меняются быстро (дни, недели)
• Полезны для быстрой feedback loop

Взаимосвязь:

Leading Metrics          Lagging Metrics
(Изменения в product) → (Business results)

Week 1: High feature      →  
adoption                  
Week 2: Growing user      →
engagement                
Month 2:                  → Revenue up 15%
                          → Churn down 5%

Если leading metrics good, lagging usually follow.

Реальные примеры взаимосвязей

Пример 1: SaaS Subscription Platform

North Star: MRR (Monthly Recurring Revenue)

├─ New MRR = New customers × Average contract value
│  ├─ Lead generation rate (website visits, signups)
│  ├─ Sales conversion rate (free trial → paid)
│  ├─ Deal size (depends on plan chosen)
│  └─ Time-to-first-purchase (how fast they buy)
│
├─ Expansion MRR = Existing customers × upsell rate × uplift
│  ├─ Feature adoption (are users using advanced features?)
│  ├─ Health score (how likely to upgrade?)
│  ├─ Customer success engagement (calls, trainings)
│  └─ Average contract value (which plans do they choose)
│
└─ Churn MRR = Churning customers × revenue lost
   ├─ Customer satisfaction (NPS score)
   ├─ Time-to-issue resolution (support responsiveness)
   ├─ Feature completeness (missing features → churn)
   └─ Competitive threats (competitor appears, churn risk)

Взаимовлияния:

• If feature adoption ↓ → health score ↓ → expansion MRR ↓ → churn ↑
• If NPS ↓ → customer satisfaction ↓ → churn ↑ → MRR ↓
• If time-to-issue ↑ → satisfaction ↓ → churn ↑

Пример 2: E-commerce Platform

Revenue = Traffic × Conversion Rate × AOV

Traffic (внешняя)
├─ Organic search (SEO)
├─ Paid ads (CAC)
├─ Direct traffic (brand strength)
└─ Referral (word of mouth)

Conversion Rate (product quality)
├─ Product-market fit (are we selling what people want?)
├─ Checkout friction (steps, form fields)
├─ Trust signals (reviews, guarantees)
├─ Page load speed (site performance)
└─ Mobile optimization (>60% traffic mobile)

AOV (pricing + behavior)
├─ Avg product price (product mix)
├─ Cross-sell rate (frequently bought together)
├─ Upsell rate (higher-priced variants)
└─ Bulk discount take rate (B2B orders)

Взаимовлияния:

• If page load speed ↑ (bad) → bounce rate ↑ → conversion ↓ → revenue ↓
• If we lower prices → AOV ↓ but conversion ↑, net effect depends on elasticity
• If we optimize for high AOV items → conversion ↓ (fewer people buy) but AOV ↑

Пример 3: Marketplace (Uber/Airbnb style)

GMV = Supply × Demand × Transaction size

Supply (供給方面)
├─ # of active sellers (drivers, hosts)
├─ Inventory per seller (cars, listings)
├─ Seller utilization rate (% of time used)
└─ Seller satisfaction (NPS, earnings)

Demand (需要方面)
├─ # of active buyers (riders, guests)
├─ Purchase frequency (rides/stay per user)
├─ Buyer satisfaction (NPS)
└─ Repeat purchase rate (loyalty)

Transaction Size
├─ Avg order value
├─ Distance/duration (ride length, stay nights)
├─ Surge pricing (peak time multiplier)
└─ Add-ons (tips, insurance)

Критичная взаимосвязь (Chicken & Egg problem):

• If buyers ↓ → sellers wait longer → earn less → churn ↓ supply → buyers ↑ wait time → churn ↑ demand
• Both sides must grow together
• This is "network effect" — harder to grow but stronger once you do

Correlation vs Causation: как я это различаю

Опасная ошибка: Думать, что correlation = causation.

Примеры:

• "Support response time коррелирует с churn, значит poor support causes churn"
• Но может быть: bad product → more issues → slow response AND more churn (both caused by product)
• Fixing support speed alone не поможет если product плохой

Как я определяю causation:

1. Исторический анализ

• Когда мы улучшили feature X, что произошло с Y?
• Отделяю ассоциацию от причинности

2. A/B testing

• Меняю metric X в тесте, смотрю на Y
• Если Y меняется только в тесте-группе → causation
• Если Y меняется и в контроле → external factors

3. Временной лаг

• Causa должна предшествовать effect
• Если feature adoption растет (week 1) и потом NPS растет (week 3) → может быть причинность
• Если они меняются одновременно → probably external factor

4. Domain knowledge

• Спрашиваю domain experts: "Имеет ли логический смысл, что X влияет на Y?"
• Если нет логики → probably spurious correlation

Diverging Metrics: когда метрики противоречат друг другу

Иногда при улучшении одной метрики другая падает.

Пример: Improving Conversion vs AOV

Scenario 1: Lower prices → More conversions but Lower AOV
- Revenue impact: depends on price elasticity
- If elasticity = 1.5 (1% price drop → 1.5% volume increase)
  Revenue goes up despite lower AOV

Scenario 2: Simplify checkout → More conversions but abandon higher value carts
- Revenue impact: unlikely, usually lower friction = better for all segments

Как я этот разрешаю:

1. Measure both metrics

Revenue = Conversion Rate × AOV

If ConversionRate ↑ 10% and AOV ↓ 5%
Revenue ↑ 4.5% — это win overall

1. Segment analysis

High-value users: Did their AOV change?
Price-sensitive users: Did their conversion change?
Often different segments respond differently

1. Monitor leading metrics ahead of time

Before: Launch "simplify checkout"
Expected: Conversion ↑, AOV might ↓
Monitor: Real-time conversion changes
Decision: Are enough conversions increasing to offset AOV loss?

Metrics Hierarchy: как я их организую

Tier 1 (North Star) — 1 метрика
└─ Revenue / MRR / GMV / MAU

Tier 2 (Key Results) — 3-5 метрик
├─ User growth
├─ Engagement / Feature adoption
├─ Monetization
├─ Retention
└─ Customer satisfaction

Tier 3 (Diagnostic) — 10-20 метрик
├─ For growth: CAC, LTV, Payback period
├─ For engagement: DAU, Session length, Feature adoption %
├─ For monetization: ARPU, AOV, Conversion rates
├─ For retention: Churn, NPS, Customer health score
└─ For satisfaction: CSAT, Support tickets, Bounce rate

Tier 4 (Detailed KPIs) — 50-100+ метрик
├─ Page views, Click-through rates, Time on page
├─ Bounce by segment, by traffic source
├─ Feature-specific adoption rates
└─ [Everything measured but not actively monitored]

Golden rule: Monitor Tiers 1-2 weekly, drill into Tier 3 when you see changes, use Tier 4 for debugging.

Как я документирую метрик relationships

Я создаю документ "Metrics Dictionary":

## Metrics Dictionary

### Revenue (North Star)
**Definition:** Total recurring revenue per month
**Formula:** New MRR + Expansion MRR - Churn MRR
**Related:** All other metrics (diagnostic purpose)
**Cadence:** Daily monitoring, Weekly review
**Target:** $1M/month by end of year

### Feature Adoption (Leading)
**Definition:** % of users who used feature X at least once in last 30 days
**Formula:** (Users with feature usage) / (Total active users)
**Related to:** Revenue (via expansion), Churn (via retention)
**Cadence:** Daily
**Insight:** If ↓ → need to improve onboarding or feature clarity

### Churn Rate (Lagging)
**Definition:** % of customers who cancel subscription in a month
**Formula:** Churned customers / Start-of-month customers
**Drivers:** NPS, feature adoption, support quality, pricing
**Cadence:** Monthly (reported in arrears)
**Target:** < 3% monthly churn
**Alert:** If > 4% → investigate immediately

Обычные паттерны корреляций, которые я наблюдаю

Key Takeaways

1. North Star drives everything — align все метрики к ней
2. Understand causation, not just correlation — drill deep to understand why
3. Use leading metrics for fast feedback — don't wait for lagging metrics
4. Diverging metrics are normal — measure both, optimize for North Star
5. Segment analysis reveals hidden truths — overall metrics hide segment-specific patterns
6. Document your metric relationships — helps new team members understand system
7. Monitor in hierarchy — drill down only when Tier 1-2 metrics change

Метрики — это язык, на котором data и business talks. Когда вы понимаете, как они связаны, вы можете predict consequences и make better decisions.