23 апреля 2026 г.

Как развить data-intuition

Карьерник — квиз-тренажёр в Telegram с 1500+ вопросами для собесов аналитика. SQL, Python, A/B, метрики. Бесплатно.

Зачем это знать

«Number feels wrong» — sense, что metric неправильный до confirmed. Senior analysts имеют. Critical skill.

Без intuition — published wrong numbers. С intuition — catch errors проактивно.

Что такое data intuition

Ability быстро assess reasonableness numbers.

Examples:

  • «Facebook DAU 10 million» — плохо для Facebook.
  • «Our startup DAU 10 million» — wonderful для startup.

Context-sensitive judgment.

Components

1. Benchmarks

Know standard numbers вашего сektor.

2. Orders of magnitude

Fluent с scales (thousands → millions → billions).

3. Relationships

Understand how metrics connect (CR + traffic = conversions).

4. Anomaly sense

When looks wrong, know it.

5. Fermi estimation

Quick rough calculations.

Build process

1. Exposure

Look at many numbers. Many domains.

2. Comparison

Always «is this high or low?».

3. Benchmarks

Memorize key ones.

4. Question

«What would break this number?»

5. Practice

Daily exercise.

Сmart benchmarks

General SaaS

  • Monthly churn: 1-5%
  • LTV/CAC: 3+
  • Gross margin: 70%+
  • Activation rate: 40-60%

Mobile apps

  • D1 retention: 30-40%
  • D7: 10-20%
  • D30: 3-5%
  • Paying rate: 2-5%

Web

  • Bounce rate: 40-60%
  • Avg session: 2-5 min
  • Pages per session: 2-5

E-commerce

  • CR: 2-5%
  • AOV: varies $20-500
  • Cart abandonment: 70%

Ads

  • CTR: 1-5%
  • CPC: varies
  • CPA: depends vertical

Adapt к industry / product.

Magnitudes

Familiar

  • 1K = thousand
  • 1M = million
  • 1B = billion
  • 1T = trillion

Мир

  • Population: 8B
  • Russia: 145M
  • Moscow: 12M
  • Russian internet: 100M
  • Russian smartphones: 90M

Tech companies

  • Facebook MAU: 3B

  • YouTube: 2B

  • Instagram: 2B

  • TikTok: 1.5B

  • Yandex search: 95M MAU (RU)

  • VK: 100M MAU (RU)

  • Telegram: 800M global

Money

  • 1M ₽ ≈ $11k
  • 1B ₽ ≈ $11M

Currency conversions matter.

Estimation exercises

Fermi-style:

«How many taxis в Moscow»?

  • Moscow population: 12M
  • % who take taxi: 30% (3.6M users)
  • Trips per user per month: 3 → 10M trips
  • Active taxis: 100k cars
  • Trips per taxi per month: ~100

Orders of magnitude correct.

«DAU Yandex Music»

  • Russia internet users: 100M
  • % who stream music: 50% (50M)
  • % use Yandex (vs VK, Spotify restricted): 40% (20M)
  • DAU/MAU ratio: 50% → 10M DAU

Rough, but reasonable.

«Telegram daily messages»

  • Users: 800M
  • Active: DAU ~500M
  • Messages per DAU: 10
  • Daily: 5B messages

Roughly correct.

Sanity checks

Always

  • «Sum of parts = whole?»
  • «Percentages sum 100%?»
  • «Median < Mean для skewed?»
  • «Trend direction reasonable?»
  • «Compared к historical?»

Вопросы

  • Is it growing faster than industry?
  • Does it match my мental model?
  • What would explain?

Red flags

Too good

  • 95% retention
  • 50% conversion
  • 20x LTV / CAC

Usually wrong.

Too stable

Real-world fluctuates. Perfectly flat → maybe fake data.

Too round

10000 exactly — suspicious (sample, rounding).

Inconsistent

One metric says X, related metric Y. Contradict → investigate.

Training

1. Daily numbers

Notice numbers everywhere. Quantify estimates.

  • Walk street: «how many cars passing / min?»
  • News article: «does stat sound right?»

2. Back-of-envelope

Give yourself problems. Solve without calculator.

«How many pizzas sold daily Moscow?»

3. Compare

Read competitors' stats. Yours higher / lower? Why?

4. Practice

Kaggle datasets. EDA. Notice patterns.

5. Communities

Analytics chats. Others' questions / answers.

Common mistakes

Ignore context

«CR 5%» — which industry? Which product?

No benchmark

Isolated numbers meaningless.

Skip calculation

«Sounds good». Actual math важен.

Confirmation bias

Believe positive numbers. Skeptical negative.

Specific

Retention

Think curve. Sharp drop? Plateau? Survivorship?

Funnel

Where biggest drop? Industry position?

A/B lift

Effect size reasonable? Too good = check methodology.

Revenue

Units consistent? Different countries / currencies mixed?

Tools

Calculator mental

  • % change: (new - old) / old
  • Doubling time: ~70 / annual %
  • Compounding: small %s accumulate

Spreadsheet

Quick check calculations.

Python REPL

12_000_000 * 0.05 * 0.3  # Moscow × engagement × conversion

Fluent use.

Historical context

Data your company

Year-over-year. What's typical?

Industry history

Benchmarks evolve. Historic data points.

Major events

COVID impact, economic shifts. Adjust mental models.

Для decision-making

Intuition helps:

  • Catch wrong analyses quickly
  • Sanity check stakeholder claims
  • Identify valuable investigations
  • Communicate confidently

Не replacement для analysis

Intuition starts. Rigor confirms.

«Feels wrong» → investigate, не just dismiss.

Build habit

Weekly review

Key metrics. Notice trends, anomalies, surprises.

Team discussions

«Why this metric moved?» Discuss hypotheses.

Read widely

Various reports, blog posts. Different angles.

Write

Blog posts, internal notes. Forces thinking.

На собесе

Quick estimations demo intuition.

«How many taxis Uber processes India daily?»

Walk through:

  • Population 1.4B
  • Urban 50% → 700M
  • Adults 60% → 420M
  • Users % → 20% → 84M
  • Rides per user / month: 2
  • Monthly: 168M
  • Daily: 5.6M

Structured.

Danger

Over-confidence в intuition:

  • Miss real signals
  • Dismiss correct data
  • Bias investigation

Balance: skeptical но open.

Для aналитика уровней

Junior

Starting. Use benchmarks often.

Middle

Solid в вашем domain.

Senior

Broad intuition across domains.

Lead

Executive-level judgement. Big picture.

Связанные темы

FAQ

Years для build?

Decent intuition — 1-2 years. Deep — 5+.

Только для analyst?

Useful all data roles. PMs too.

Cheat sheet?

Benchmarks specific industry. Build your own.

Тренируйте — откройте тренажёр с 1500+ вопросами для собесов.

Тренироваться в Telegram