Какие харды нужны аналитику в 2026 | Подкаст «Доверительный интервал»

Main ideas & lessons

• Core debate: hard skills vs. analytical thinking

  • The discussion centers on whether analysts in 2026 need “hard skills” (e.g., SQL/Python, math/statistics) or whether analytical thinking and a broader analytical culture matter more.
  • One viewpoint: hard skills are not the main differentiator; conceptual/problem-structuring thinking is harder to develop and is more important long-term.
  • Another viewpoint: without hard skills it’s unclear how someone differs from other roles (e.g., product); hard skills make you an analyst, even though you still need thinking skills.

• Analytical thinking is treated as primary (but basics still required)

  • Even if libraries/LLMs can generate code or formulas, analysts must still understand:
    • cause–effect reasoning
    • critical evaluation
    • conceptual understanding of why methods/tests work (e.g., what a test is actually doing), not just how to call an implementation

• Hiring/testing philosophy: evaluate thinking with case studies

  • Instead of testing only SQL/Python knowledge, interviewers use a case to see how candidates reason:
    • how they clarify premises
    • what questions they ask
    • how they gather context
    • how they build a solution chain and reach conclusions
  • Cases are typically:
    • broad and “from life”
    • often tasks that require designing something that hasn’t been built yet (e.g., dispatch/courier assignment, dynamic pricing)
  • Preparation beforehand matters less than how the candidate thinks during the interview.
  • Technical sections still exist (often earlier/later) to check:
    • basic math/statistics and coding ability
    • but the analytical case is meant to assess reasoning

• Analyst vs product manager/product roles: different mindsets

  • Analysts tend to ask:
    • “How much?” and “How do we calculate it?”
    • which metrics to optimize and how to structure measurement
  • Product managers tend to ask more:
    • “How are we going to do this?” (process/implementation/product steps)
    • typically with less digging into numeric/statistical specifics (though conceptual understanding may still be needed)

• Career direction: generalist vs specialist

  • The speakers lean toward a T-shaped path:
    • broad versatility early
    • potential for deep expertise later
  • Fully expert tracks exist (e.g., deep A/B testing specialists or teams building AB platforms), but they’re rarer and in smaller quantities.

---

Methodology / practical guidance (as presented)

1) What to assess in interviews (analytical case approach)

• Use a realistic scenario relevant to the team’s work.
• Ask the candidate to:
  • outline how they would approach the task (even if they can’t provide the exact “final” algorithm)
  • explain how they set up assumptions/premises
  • request/gather necessary context
  • ask clarifying questions
  • reason through steps and build “chains” to a conclusion
• Ensure cases are broad enough that:
  • preparation doesn’t fully determine success
  • the candidate must reason “on the fly”
• Add additional stages to check minimum hard-skill literacy, so candidates can pass technical gates before/alongside the analytical case.

2) How to treat hard skills in 2026 (LLMs/data tooling context)

• Python / SQL may be less about writing everything from scratch and more about:
  • conceptual understanding
  • validating, reviewing, and critiquing generated code
  • understanding architecture and data context
• Analysts may rely on LLMs for code generation, but still must know enough to:
  • detect wrong assumptions
  • understand what the code/data pipeline is doing
  • ensure outputs and definitions align with business needs
• Practical need depends on the environment:
  • mature data engineering/DWH and self-service tooling may reduce daily coding needs
  • in a startup/small team where you’re the only analyst, you may need to build infrastructure and scheduled processes—so hard skills remain important

3) “Entry ticket” framing for juniors

• Hard skills act like a screening filter:
  • without foundational knowledge, it’s hard to enter/practice the profession
• For junior analysts:
  • basic competency in relevant hard skills (plus math/stat basics) is treated as necessary
  • analytical thinking is still the long-term differentiator

---

Specific hard skills discussed (2026 relevance)

• Python

  • Still relevant, especially for building pipelines/infrastructure when engineering support is limited
  • Not required to memorize every loop/function; instead, understand how to work at an architectural level and critique LLM output

• SQL

  • Syntax/function trivia (e.g., window functions) may be less emphasized if tooling/LLMs help
  • The core challenge remains:
    • understanding data structures
    • table meaning
    • column semantics
    • context
  • Different analyst “schools” exist:
    • SQL-heavy analysts
    • Python-heavy analysts who process more after extraction

• Excel

  • Framed as the “mother of all analytics” for many real-world workflows
  • Used heavily especially when culture/tools are less mature
  • Even in mature environments, mature reporting can still be exported to Excel for quick iteration

• LLMs / AI assistants

  • Used to speed up routine analytics
    • reported team survey: ~30–40% improvement on routine tasks
  • Common uses:
    • turning Excel/DataLens outputs into combined artifacts quickly
    • discussing methodologies and getting critique
    • generating SQL → PowerBI / processing scripts or similar
  • Example “agent” behavior:
    • LLM/internals calculate metrics like “orders in the food segment” using predefined classification rules, revenue, average-bill logic—highlighting the importance of business definitions

---

Speakers / sources featured

Speakers

• Dima (Дима) — team lead, analytics for Yandex logistics; host of the “Confidence Interval” podcast
• Ksyusha (Ксюша) — leads analytics for the workshop; discusses education background and analytics tooling; mentions programming/mathematics background and an internal library she wrote
• Sasha (Саша) — responsible for product marketing and customer analytics at YandexReal Estate and Yandex Rent; discusses conceptual/analytical thinking and education background
• “Dim” / “Dima” — referenced multiple times as the same person (Dima)

Sources / external references mentioned

• Yandex (logistics; product/analytics context)
• Yandex Music
• Telegram channel (Yandex Fone Analytics mentioned)
• Public domain materials: “Nahabr” (likely Habr) and the podcast/channel content
• Yandex DataLens (mentioned in an LLM-assisted workflow example)
• Internal LLM/agent (“Insight” / “IseatUs”-like referenced agent) for metric calculation
• LMKA / LLMs (large language model assistants) mentioned throughout