23 апреля 2026 г.

Как аналитику работать с AI

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

Зачем это знать

ChatGPT, Claude, Copilot — меняют работу аналитика. Те, кто использует — в 2-3x productive. Те, кто игнорирует — отстают.

На собесах в 2026+ «как использовать AI» — частый вопрос. Honest answer expected, не «I don't use».

Где AI помогает

1. SQL generation

«Напиши SQL: найди users, кто покупали в последние 30 дней и их средний чек».

AI пишет. Вы check и run.

Workflow:

  1. Describe задачу в natural language
  2. AI generates query
  3. Review logic
  4. Test on real data
  5. Iterate

Важно: always verify. AI sometimes hallucinates.

2. Debugging code

Stuck with error? Copy-paste stack trace в ChatGPT → explanation, fix.

3. Explain unfamiliar code

Inherit 500-line query. «Explain this SQL step-by-step» → saved hours.

4. Documentation

Write query comments / dashboard descriptions. AI drafts → вы edit.

5. Regex

«Regex для email validation» — AI быстрее Google.

6. Learning

«Explain window functions с examples» — personalized tutor.

7. SQL optimization

Paste slow query + EXPLAIN → «how to optimize?».

8. Python help

pandas tricks, numpy, scipy — quick reference.

9. Statistical concepts

«Explain bias-variance tradeoff simply» — great для onboarding topics.

10. Analysis brainstorm

«I need analyze churn. Какие angles могу посмотреть?» — get checklist.

AI tools

ChatGPT

Most popular. GPT-5 (2026) — strong SQL, analysis.

Free tier limited. Paid $20/mo.

Claude (Anthropic)

Strong на SQL / code. Long context window.

GitHub Copilot

In-IDE. Autocomplete SQL / Python. $10/mo.

Cursor

AI IDE. Deep code integration.

Perplexity

Research / search + AI.

YandexGPT / GigaChat

Russian models. Getting better.

Примеры

SQL generation

Prompt:

«Write SQL for this Postgres schema:

  • orders(id, user_id, created_at, total)
  • users(id, email, signup_at)

Find users whose first order было в January 2026, and их total spend с того момента».

Result: usually good query. Review → run.

Code review

Prompt:

«Review this query for correctness, performance, edge cases:

[PASTE QUERY]»

AI flags issues.

Python

Prompt: «pandas: как get для each category топ-3 values by count?»

Response:
df.groupby('category').apply(
    lambda x: x['value'].value_counts().head(3)
)

Analysis plan

Prompt:

«I need analyze why retention D30 dropped from 40% to 35% за last quarter. Give me comprehensive analysis plan с SQL queries где уместно».

Получаете outline, hypotheses, queries.

Pitfalls

1. Hallucination

AI makes up functions, column names. Verify.

«There's no function MAGIC_SQL_FUNC — might be hallucinated».

2. Outdated

Training cutoff — 2024/2025. Miss newest features.

3. Wrong assumptions

AI assumes schema. If your setup differs — wrong query.

Always verify against real schema.

4. Security

Don't paste sensitive data (customer names, financial data).

Anonymize first.

5. Dependency

Over-relying — skills atrophy. Still practice SQL / Python manually.

6. Context limits

Huge codebases / data — AI can't hold всё. Split chunks.

Prompt engineering

Specific

«Write SQL for Postgres» > «write SQL».

Context

«Given these tables: [schema]. Find X».

Examples

«Output в this format: ...».

Iterate

First answer imperfect. «Fix X, also consider Y». Refine.

Chain of thought

«Think step by step».

Workflow

Typical аналитик день в 2026:

Morning

  • Check AI-generated daily report summary
  • Respond к Slack с AI drafts

Work

  • Complex SQL: draft via AI, verify, run
  • New query: write yourself, AI optimizes
  • Blocked? Ask AI explain concept

Meetings

  • Quick questions → AI real-time

End of day

  • Summary — AI drafts
  • Code review с AI suggestions

Privacy concerns

Company data

Don't paste internal schema / queries в public ChatGPT без checking policy.

Use enterprise tools (Copilot for Business, Claude Enterprise, internal LLM).

Secrets

Никогда no passwords, API keys, PII.

Output verification

AI can produce plausible-wrong code. Always review.

Для собесов

Preparation

AI tutors: «Mock interview, SQL questions for middle analyst».

Practice problems

«Give me 10 SQL problems on window functions, hard level».

Solution explanation

Stuck on problem? Ask AI explain.

Code review mock

«Here's my solution. Critique».

Skills shift

В AI era, valuable скилы:

Less: memorizing syntax

AI remembers все.

More: problem framing

Asking right questions.

More: judgment

Which output use, which reject.

More: domain

AI general. You understand business.

More: communication

Explain к stakeholders. AI draft — вы polish.

На собесе

«Используете AI?»

Honest yes, с specifics:

  • SQL drafts для complex queries
  • Debugging
  • Code review
  • Learning

«Verify output, understand code» — critical add.

«AI заменит аналитиков?»

No. Amplifies good аналитика. Replaces low-level routine. Higher-level thinking still needed.

Связанные темы

FAQ

Copilot или ChatGPT?

Both. Copilot — in-IDE. ChatGPT / Claude — сomplex queries.

Free enough?

For learning — yes. Work — paid usually worth.

Можно заменить tutor?

Partially. Human mentor лучше для career advice.

Тренируйте с AI — откройте тренажёр с 1500+ вопросами для собесов.

Тренироваться в Telegram