Lamoda — крупнейший fashion-маркетплейс РФ (часть Yandex). Собес туда отличается от обычного e-comm: фокус на fashion-specific метриках (return rate by size, fit prediction, customer lifecycle longer than e-comm staples). В этом гайде разберу 25+ реальных вопросов с 5 раундов собеса аналитика Lamoda — с разбором сильного и слабого ответа.
Грейды аналитика в Lamoda (2026)
| Грейд | Compensation/мес РФ | Опыт | Что спрашивают |
|---|---|---|---|
| Junior | 130-200K ₽ | 0-1 год | SQL, pandas, базовые e-comm метрики |
| Middle | 200-310K ₽ | 1-3 года | ClickHouse, A/B, fashion-specific метрики |
| Senior | 310-470K ₽ | 3-6 лет | ML / size prediction, RecSys, cohort deep |
| Lead | 470-630K+ ₽ | 6+ лет | Multi-team, fashion strategy, mentorship |
5 раундов собеса Lamoda
| Раунд | Что | Длительность |
|---|---|---|
| 1. HR-скрининг | Мотивация, fashion опыт | 30 мин |
| 2. SQL live (ClickHouse/PostgreSQL) | 2-3 задачи + fashion cases | 60 мин |
| 3. Python + статистика | pandas + A/B + cohort | 60 мин |
| 4. Fashion-кейс | Return rate / size / pricing | 60 мин |
| 5. Финал | Behavioral + culture | 45 мин |
7 SQL-вопросов с собеса Lamoda
Return rate by size — найди размеры с high return rate.
✅ Сильный ответ:
\\\sql
SELECT
product_id,
size,
count() AS total_orders,
countIf(status = 'returned') AS returned,
countIf(status = 'returned') * 1.0 / count() AS return_rate
FROM orders
WHERE order_date >= today() - 90
GROUP BY product_id, size
HAVING total_orders >= 100
ORDER BY return_rate DESC, total_orders DESC
LIMIT 100;
\\\
Senior follow-up: «Why \total_orders >= 100\? — statistical significance. Иначе sizes с 5/10 returns =50% return rate скрывают noise.»
Cohort analysis: M+1 return rate.
✅ Сильный ответ:
\\\sql
WITH first_order AS (
SELECT user_id, toStartOfMonth(min(order_date)) AS cohort
FROM orders
WHERE status IN ('completed', 'returned')
GROUP BY user_id
)
SELECT
fo.cohort,
count() AS new_buyers,
countIf(o.order_date >= addMonths(fo.cohort, 1) AND o.order_date < addMonths(fo.cohort, 2)) AS m1_active,
100.0 * countIf(o.order_date >= addMonths(fo.cohort, 1) AND o.order_date < addMonths(fo.cohort, 2)) / count() AS m1_return_pct
FROM first_order fo
LEFT JOIN orders o ON fo.user_id = o.user_id
GROUP BY fo.cohort
ORDER BY fo.cohort;
\\\
Category mix: % выручки по категориям за квартал.
✅ Сильный ответ:
\\\sql
WITH cat_revenue AS (
SELECT category, sum(amount) AS rev
FROM orders
WHERE order_date >= '2026-01-01' AND order_date < '2026-04-01'
AND status = 'completed'
GROUP BY category
)
SELECT category, rev,
rev * 100.0 / sum(rev) OVER () AS pct_of_total
FROM cat_revenue
ORDER BY rev DESC;
\\\
AOV by user cohort (recency-based segmentation).
✅ Сильный ответ:
\\\sql
WITH user_recency AS (
SELECT user_id,
dateDiff('day', max(order_date), today()) AS days_since_last,
avg(amount) AS aov
FROM orders
WHERE status = 'completed'
GROUP BY user_id
)
SELECT
CASE
WHEN days_since_last <= 30 THEN '0-30d (recent)'
WHEN days_since_last <= 90 THEN '31-90d (warm)'
WHEN days_since_last <= 180 THEN '91-180d (cold)'
ELSE '180d+ (dormant)'
END AS recency_bucket,
count() AS users,
avg(aov) AS avg_aov
FROM user_recency
GROUP BY recency_bucket
ORDER BY days_since_last;
\\\
Brand performance: топ-10 brands по revenue per active customer.
✅ Сильный ответ:
\\\sql
SELECT
brand,
uniq(user_id) AS unique_buyers,
sum(amount) AS total_revenue,
sum(amount) * 1.0 / uniq(user_id) AS revenue_per_buyer
FROM orders
WHERE order_date >= today() - 90
AND status = 'completed'
GROUP BY brand
HAVING unique_buyers >= 500
ORDER BY revenue_per_buyer DESC
LIMIT 10;
\\\
Window function: разница между orders.
✅ Сильный ответ:
\\\sql
WITH order_gaps AS (
SELECT user_id, order_date,
lag(order_date) OVER (PARTITION BY user_id ORDER BY order_date) AS prev_order,
dateDiff('day', lag(order_date) OVER (PARTITION BY user_id ORDER BY order_date), order_date) AS days_between
FROM orders
WHERE status = 'completed'
)
SELECT user_id,
avg(days_between) AS avg_days_between_orders,
count() AS total_orders
FROM order_gaps
WHERE prev_order IS NOT NULL
GROUP BY user_id;
\\\
Senior follow-up: «Average 60-90 days between orders typical для fashion (vs daily для grocery). Long cycle = больше внимания retention vs frequency.»
Top-3 categories per user (window function).
✅ Сильный ответ:
\\\sql
WITH ranked AS (
SELECT user_id, category, sum(amount) AS spent,
row_number() OVER (PARTITION BY user_id ORDER BY sum(amount) DESC) AS rn
FROM orders
WHERE status = 'completed'
GROUP BY user_id, category
)
SELECT user_id, category, spent
FROM ranked
WHERE rn <= 3
ORDER BY user_id, rn;
\\\
5 Python/ML-вопросов
Size prediction: что features использовать?
✅ Сильный ответ:
«Features для size prediction:
- Historical purchases: что юзер покупал размером (история S/M/L)
- Returns: какие размеры возвращал (signal что size off)
- Brand: same brand → consistent size (Brand A's M ≈ Brand A's M)
- Item type: dresses vs t-shirts size differently
- Demographic: age, gender, region (russian sizes vs european)
- Self-reported: profile size info if provided
Model: gradient boosting на 30+ features → predict probability of each size being correct. Top-2 sizes recommended.
Lamoda specifically: \Size Calculator\ (на сайте) collect explicit signals. Combined ML + user input.»
Pandas: AOV trend по cohort + brand.
✅ Сильный ответ:
\\\python
orders['cohort'] = orders.groupby('user_id')['order_date'].transform('min').dt.to_period('M')
agg = orders.pivot_table(
index=['cohort', 'brand'],
columns=orders['order_date'].dt.to_period('M'),
values='amount',
aggfunc='mean'
).fillna(0)
\\\
Lookalike modeling для cross-sell.
✅ Сильный ответ:
«Approach: найди juniors look-alike to top customers (LTV или basket size).
- Define seed: top-10% customers by LTV
- Features: demographic + behavioral (categories visited, brands purchased, frequency)
- Train: binary classifier (top vs not top)
- Score all users → probability of being like top
- Action: target top-scoring users с lifestyle marketing campaigns
In Lamoda: используется для premium customer acquisition — найти аудиторию для luxury brand promotions.»
Customer lifetime value (LTV) calculation для fashion.
✅ Сильный ответ:
«LTV для fashion complicated by long cycle. Approaches:
Simple: LTV = AOV × purchase frequency × customer lifespan
- Fashion AOV ~3000-5000 ₽
- Frequency ~4-6 purchases/year
- Lifespan ~3-5 years average
- LTV ~50,000-150,000 ₽ typical
Cohort-based: track cohort spending over time, project forward.
ML-based: BG/NBD model для frequency + Gamma-Gamma для basket → expected LTV.
For Lamoda specifically: разделяем on customer type (casual vs premium) — premium customers могут иметь LTV 500K+ ₽.»
→ LTV/CAC/Payback для SaaS — методология
Recommendation system для cross-sell.
✅ Сильный ответ:
«Levels of recommendations:
- «You may also like» — content-based (similar items)
- «Customers also bought» — collaborative filtering (item-item)
- «Complete your look» — outfit completion (top + bottom + accessory match)
- «Trending in your size» — personalized + popular
For fashion specifically:
- Style consistency (минимальный vs maximalist customers)
- Color coordination (RGB / color name matching)
- Season-appropriate (spring/summer vs autumn/winter)
- Body type compatibility (size + fit preferences)
Lamoda uses hybrid: ALS for collaborative filtering + content-based image embeddings для visual similarity.»
4 fashion-кейса
Return rate jumped from 35% to 42% — что делаешь?
✅ Сильный ответ:
«Декомпозиция:
Return rate by:
- Category — fashion (clothes) vs accessories vs shoes (each different baseline)
- Brand — premium vs mass-market
- Size — какие sizes drive returns
- Reason (если tagged): too big / too small / quality / not as described
- Time — sudden jump (incident) vs gradual (trend)
Hypotheses:
- New brand с poor sizing onboarded
- Manufacturing issue с specific batch
- New marketing campaign attracted wrong audience
- Seasonal shift (winter to spring inventory mix)
- Pricing strategy (impulse buys → more returns)
Action:
- Drill-down to root cause
- If brand-specific → escalate to category manager
- If marketing-specific → adjust ad creative
- Always: monitor weekly trend post-intervention»
Size Calculator вырос → return rate должен упасть. Не упал. Почему?
✅ Сильный ответ:
«Hypothesis testing:
- Selection bias: только конкретные users use Size Calculator (uncertain users). Они и без него возвращают часто.
- Calibration: Size Calculator outputs не accurate (especially для нестандартных brands)
- Override: users still order их familiar size despite recommendation
- Category coverage: Size Calculator works для basics но not для new arrivals
Investigation steps:
- A/B test: users with vs without Size Calculator shown
- Accuracy audit: predicted size vs actually kept size
- Survey: users about why они choose differently
Action:
- If selection bias → segment analysis (without Calculator users behave differently)
- If calibration → retrain model
- If override → UX changes (highlight recommendation more prominently)»
Premium category retention падает — что делать?
✅ Сильный ответ:
«Premium fashion specific challenges:
- High AOV but lower frequency
- Quality sensitivity — small issues big complaints
- Service expectations — premium customers expect premium service
- Brand loyalty — usually стой к specific brands
Hypothesis:
- New competitor (e.g., entered premium space)
- Premium customers shifted to international (overseas shipping)
- Service quality declined (returns, delivery)
- Brand portfolio изменилось (favorite brands ушли)
Action:
- Survey of lapsed premium customers
- Premium account management (white-glove service)
- Exclusive launches для premium tier
- Loyalty program tier benefits»
New collection rollout: что мерим, что A/B test'им?
✅ Сильный ответ:
«Pre-launch metrics:
- Catalog page CTR (interest signal)
- Wishlist additions (intent signal)
- Search query share (organic discovery)
Post-launch (1 week):
- Sell-through rate (% of inventory sold first week)
- AOV vs forecast
- Return rate vs category baseline
- New customer acquisition rate
A/B tests:
- Pricing tier (full price vs discount intro)
- Marketing channel mix (paid social vs email)
- Featured placement (top of category vs dedicated landing)
- Influencer tie-ins (with vs without)
Long-term (1 month):
- Repeat purchase rate from launch buyers
- Reviews и sentiment
- Cross-sell impact (did launch boost overall sessions?)»
2 behavioral (раунд 5)
Расскажи случай когда твой анализ изменил решение.
✅ Сильный ответ (STAR):
«Ситуация: Marketing хотел запустить flash sale 50% off для clearance inventory.
Задача: анализировать impact на margin + brand reputation.
Действие:
- Pulled historical 30% / 40% / 50% sale data
- Showed elasticity: 50% sale increases volume only marginally vs 40% (diminishing returns)
- Showed reputation impact: returns higher после deep discounts (low-intent buyers)
- Recommended 35% with limited time window
Результат: Marketing согласился с 35% × 48 hours. Result: 80% of 50%-sale revenue achieved with 50% higher margin and 30% lower return rate.»
Самая сложная analytical задача.
✅ Сильный ответ:
«Задача: Build dashboard для C-level showing «true profit per customer» considering returns, support cost, marketing CAC, fraud.
Сложности:
- Data scattered across 8+ systems
- Definitions inconsistent (returns vs refunds vs chargebacks)
- Time-shifted costs (CAC spent now, profit realized over 2-3 years)
- Fraud detection probabilistic не binary
Approach:
- Stakeholder interviews (CFO, COO, CMO) для alignment on definitions
- Data integration layer (dbt models)
- Probabilistic model для fraud / LTV
- Cohort-based attribution для CAC
Result: Dashboard live, weekly C-level review. Insights drove changes in CAC allocation (cut underperforming channels) saving 15% marketing spend.»
Red flags (не делай)
- Не упомянуть size/fit specific challenges
- Generic e-comm answers без fashion context
- Не учитывать longer customer cycle (fashion ≠ daily grocery)
- Игнорировать brand portfolio dynamics
- Слишком ML-heavy без practical fashion examples
Как готовиться к Lamoda
Месяц 1: e-comm + fashion specifics
- AOV, return rate, cohort retention
- Size analytics, brand performance
- 521 SQL-задача
Месяц 2: cohort + ML basics
- Cohort analysis deep
- Recommendation systems
- Size prediction
Месяц 3: кейсы + behavioral
- E-comm кейсы из /cases
- AI мок-собес
FAQ
Lamoda vs Wildberries?
WB больше про operations + tech. Lamoda больше про fashion product (curation, brand portfolio, premium customers).
Можно ли пройти без e-comm опыта?
Junior — да. Middle+ — желательно retail/e-comm опыт.
Стек технологий?
ClickHouse + PostgreSQL + Yandex DataLens + Python + R (statistics).
Что дальше
- 521 SQL-задача
- 453 кейса фильтр e-commerce
- AI мок-собес
- Вопросы Wildberries аналитика
- Вопросы Ozon аналитика
- 150+ вопросов общего собеса
Сравнить Free и Pro → (1999 ₽/мес)
Источники
- Levels.fyi РФ 2026 (Lamoda grades, salary bands)
- Habr / lamoda.tech blog
- exp-platform.com — A/B methodology для retail