Target Data Scientist Interview Guide

Certain Target teams are hiring for recommender systems, reinforcement learning, and LLM-powered search ranking. That makes specific pods within this org feel more like applied ML shops than what you'd expect from a retailer. If you're prepping only for SQL and basic classification questions, you'll be underprepared for the ML depth some of these teams demand.

Target Data Scientist Role

You're embedded with business teams across personalization, supply chain, search, and pricing. The work ranges from building production ML models (recommender systems for Target Circle, ranking models for Target.com search) to designing experiments and translating results for merchandising partners. What "success" looks like depends heavily on your team and level: a P3 on the MarTech team might ship a production recommender model and show measurable lift in click-through rate, while a P3 on an experimentation-focused team might own the measurement framework for a major product launch.

A Typical Week

The split between writing code and writing words is closer to even than most candidates expect. Target's engineering culture treats documentation (feature design docs, experiment readout decks, Confluence writeups) as a first-class artifact, so your ability to narrate results matters as much as your ability to produce them. If you're someone who ships a model and moves on without writing anything down, you'll feel friction here.

Projects & Impact Areas

The MarTech applied ML team works on recommender systems and reinforcement learning to personalize what Target Circle members see across app and web. Search and ranking is a separate group applying NLP and LLMs to product discovery on Target.com, a problem that compounds in difficulty every time the assortment expands. Supply chain optimization (including last-mile delivery) and pricing/promotions round things out, and job postings for these teams emphasize production-grade, cost-efficient systems rather than just offline model accuracy.

Skills & What's Expected

ML expertise is the headline (rated expert in the skill dimensions), but the skill that quietly separates successful candidates is software engineering discipline. Target's postings call out unit testing, CI/CD, modular architecture, and code reviews alongside the ML requirements. You're expected to write production Python, build data pipelines in PySpark and Hive, and maintain your models rather than prototype in a notebook and hand off. Communication is rated high too, and the source data defines it broadly: clear narratives, appropriate visualizations, and storytelling that drives action with non-technical partners.

Levels & Career Growth

The widget shows scope and responsibilities at each level. The key insight it doesn't capture is what blocks promotion. P3 to P4 is the hardest jump because it requires shifting from "I delivered this model" to "I set technical direction for a domain and influence partner roadmaps." One realistic growth lever at Target: the retail scale means lateral moves across domains (personalization to pricing to supply chain) are a common way to build the breadth P4 and P5 roles demand.

Work Culture

According to internal culture notes, the hybrid policy has most teams in-office around three days per week at the Minneapolis headquarters on Nicollet Mall, though some senior roles (like the last-mile supply chain lead) are listed as remote-eligible. The pace, from what candidates and employees report, is steady rather than frenetic, with an emphasis on documentation, design reviews, and knowledge-sharing through internal paper reading groups. If you want rigor and a Minneapolis cost of living that makes your comp stretch further than a coastal equivalent, it's worth a serious look.

Target Data Scientist Compensation

Target's RSU vesting is a flat 25% per year across all four years. That's unusually even compared to the backloaded schedules you'll find at many big tech companies, which means your Year 1 TC won't look dramatically different from Year 2 or 3 (though stock price swings and bonus variability will still move the number). The real risk is that TGT stock has been volatile alongside broader retail, so your actual realized comp could drift meaningfully from the grant-date estimate.

No refresh grant details appear in any public source for Target's data science roles, which makes the offer stage your only window to ask about refresh cadence, size, and eligibility criteria. Push on this directly, because the answer determines whether your comp grows or stagnates between promotions.

The single biggest negotiation lever most candidates miss is level calibration. If your experience puts you on the boundary between two levels, anchoring the conversation on the higher level changes the entire comp structure (equity weight shifts dramatically at senior levels, as the widget shows). Beyond that, sign-on bonuses are a realistic ask when you're holding a competing offer from a company recruiting for similar applied ML work in recommender systems or search ranking. Bundle everything into one coordinated counter: level, base, sign-on, and relocation if you're moving to the Minneapolis area.

Target Data Scientist Interview Process

Target's loop moves quickly, so you'll want to be prep-ready before that first recruiter call lands on your calendar. Walk in already able to connect your past work to their specific domains: fulfillment route optimization, Target Circle personalization, or search ranking with LLMs. From what candidates report, the ML & Modeling round tends to be the steepest hurdle, with interviewers pushing past textbook classification into production concerns like cold-start in recommender systems, feedback loops in ranking, and retraining cadence for demand forecasting models that face weekly promo cycles.

The rejection reasons in the data tell a clear story: unclear problem framing and shallow experimentation thinking sink candidates just as often as technical gaps. Vague behavioral answers are a real risk too. Interviewers score independently, so a strong modeling round won't paper over a collaboration story where you can't name the metric you moved or explain how you navigated pushback from a merchant partner on a pricing model rollout.

Target Data Scientist Interview Questions

1WITH params AS (
2  SELECT
3    DATEADD(day, -60, CURRENT_DATE) AS start_dt,
4    CURRENT_DATE AS end_dt
5),
6-- Pre-aggregate to the grain needed for the window.
7daily_channel_revenue AS (
8  SELECT
9    CAST(o.order_ts AS DATE) AS order_dt,
10    o.channel,
11    SUM(oi.qty * oi.unit_price) AS revenue
12  FROM orders o
13  JOIN order_items oi
14    ON oi.order_id = o.order_id
15  JOIN params p
16    ON CAST(o.order_ts AS DATE) >= p.start_dt
17   AND CAST(o.order_ts AS DATE) < p.end_dt
18  GROUP BY
19    CAST(o.order_ts AS DATE),
20    o.channel
21)
22SELECT
23  order_dt,
24  channel,
25  revenue AS daily_revenue,
26  SUM(revenue) OVER (
27    PARTITION BY channel
28    ORDER BY order_dt
29    ROWS BETWEEN 6 PRECEDING AND CURRENT ROW
30  ) AS revenue_7d_rolling
31FROM daily_channel_revenue
32ORDER BY order_dt, channel;

1WITH first_order AS (
2  SELECT
3    o.guest_id,
4    o.order_id AS first_order_id,
5    o.order_ts AS first_order_ts,
6    o.channel AS first_channel
7  FROM (
8    SELECT
9      o.*,
10      ROW_NUMBER() OVER (PARTITION BY o.guest_id ORDER BY o.order_ts, o.order_id) AS rn
11    FROM orders o
12  ) o
13  WHERE o.rn = 1
14),
15app_cohort AS (
16  SELECT
17    guest_id,
18    first_order_ts,
19    DATE_TRUNC('month', first_order_ts) AS cohort_month
20  FROM first_order
21  WHERE first_channel = 'APP'
22),
23repeat_flag AS (
24  SELECT
25    c.guest_id,
26    c.cohort_month,
27    CASE
28      WHEN EXISTS (
29        SELECT 1
30        FROM orders o2
31        WHERE o2.guest_id = c.guest_id
32          AND o2.order_ts > c.first_order_ts
33          AND o2.order_ts <= DATEADD(day, 30, c.first_order_ts)
34      ) THEN 1
35      ELSE 0
36    END AS is_repeat_30d
37  FROM app_cohort c
38)
39SELECT
40  cohort_month,
41  COUNT(*) AS cohort_guests,
42  SUM(is_repeat_30d) AS repeat_guests_30d,
43  1.0 * SUM(is_repeat_30d) / NULLIF(COUNT(*), 0) AS repeat_rate_30d
44FROM repeat_flag
45GROUP BY cohort_month
46ORDER BY cohort_month;

1from __future__ import annotations
2
3from dataclasses import dataclass
4from datetime import datetime, date, timedelta, timezone
5from typing import Any, Dict, Iterable, List, Optional, Tuple
6
7
8Event = Dict[str, Any]
9
10
11def _parse_utc_date(ts: Any) -> Optional[date]:
12    """Parse event_ts into a UTC date.
13
14    Supports:
15      - ISO 8601 strings, including trailing 'Z'
16      - datetime objects (naive treated as UTC)
17      - POSIX seconds as int/float
18    Returns None if unparseable.
19    """
20    if ts is None:
21        return None
22
23    if isinstance(ts, datetime):
24        dt = ts
25        if dt.tzinfo is None:
26            dt = dt.replace(tzinfo=timezone.utc)
27        else:
28            dt = dt.astimezone(timezone.utc)
29        return dt.date()
30
31    if isinstance(ts, (int, float)):
32        dt = datetime.fromtimestamp(float(ts), tz=timezone.utc)
33        return dt.date()
34
35    if isinstance(ts, str):
36        s = ts.strip()
37        # Support Zulu time
38        if s.endswith("Z"):
39            s = s[:-1] + "+00:00"
40        try:
41            dt = datetime.fromisoformat(s)
42        except ValueError:
43            return None
44
45        if dt.tzinfo is None:
46            dt = dt.replace(tzinfo=timezone.utc)
47        else:
48            dt = dt.astimezone(timezone.utc)
49        return dt.date()
50
51    return None
52
53
54def _is_exact_duplicate(event: Event) -> Tuple[Tuple[str, Any], ...]:
55    """Create a hashable signature for exact-duplicate removal.
56
57    Exact duplicate means identical key-value pairs.
58    Sorting makes dict ordering irrelevant.
59    """
60    return tuple(sorted(event.items(), key=lambda kv: kv[0]))
61
62
63def compute_top_items_7d_trailing_revenue(
64    events: Iterable[Event],
65    end_date_utc: date,
66    top_n: int = 10,
67) -> List[Tuple[Any, float]]:
68    """Return top items by 7-day trailing net revenue ending on end_date_utc.
69
70    Rules:
71      - Skip events with missing unit_price.
72      - Negative qty is allowed and reduces revenue.
73      - Duplicates are exact same dict values.
74      - Timestamps are normalized to UTC date.
75
76    Returns list of (item_id, trailing_7d_revenue) sorted descending by revenue.
77    """
78    # 1) Deduplicate
79    seen = set()
80    deduped: List[Event] = []
81    for e in events:
82        sig = _is_exact_duplicate(e)
83        if sig in seen:
84            continue
85        seen.add(sig)
86        deduped.append(e)
87
88    # 2) Aggregate daily revenue per item_id
89    # key: (item_id, day) -> revenue
90    daily_rev: Dict[Tuple[Any, date], float] = {}
91    for e in deduped:
92        day = _parse_utc_date(e.get("event_ts"))
93        if day is None:
94            continue
95        item_id = e.get("item_id")
96        if item_id is None:
97            continue
98
99        unit_price = e.get("unit_price")
100        if unit_price is None:
101            continue
102        try:
103            price = float(unit_price)
104        except (TypeError, ValueError):
105            continue
106
107        qty = e.get("qty", 0)
108        try:
109            q = float(qty)
110        except (TypeError, ValueError):
111            continue
112
113        rev = q * price
114        daily_rev[(item_id, day)] = daily_rev.get((item_id, day), 0.0) + rev
115
116    # 3) Compute trailing 7-day sum ending on end_date_utc
117    window_days = [end_date_utc - timedelta(days=i) for i in range(7)]
118    trailing: Dict[Any, float] = {}
119    for (item_id, day), rev in daily_rev.items():
120        if day in window_days:
121            trailing[item_id] = trailing.get(item_id, 0.0) + rev
122
123    # 4) Top N
124    ranked = sorted(trailing.items(), key=lambda kv: kv[1], reverse=True)
125    return ranked[:top_n]
126
127
128# Optional quick self-check
129if __name__ == "__main__":
130    sample = [
131        {"order_id": 1, "item_id": "A", "event_ts": "2026-02-20T10:00:00Z", "qty": 1, "unit_price": 10.0, "promo_id": None},
132        {"order_id": 1, "item_id": "A", "event_ts": "2026-02-20T10:00:00Z", "qty": 1, "unit_price": 10.0, "promo_id": None},  # duplicate
133        {"order_id": 2, "item_id": "A", "event_ts": "2026-02-22T10:00:00+00:00", "qty": -1, "unit_price": 10.0, "promo_id": None},
134        {"order_id": 3, "item_id": "B", "event_ts": "2026-02-24T12:00:00Z", "qty": 2, "unit_price": 7.5, "promo_id": "P1"},
135        {"order_id": 4, "item_id": "C", "event_ts": "bad_ts", "qty": 1, "unit_price": 999.0, "promo_id": None},
136        {"order_id": 5, "item_id": "B", "event_ts": "2026-02-25T12:00:00Z", "qty": 1, "unit_price": None, "promo_id": None},  # skipped
137    ]
138
139    out = compute_top_items_7d_trailing_revenue(sample, end_date_utc=date(2026, 2, 26), top_n=10)
140    print(out)
141

The distribution skews hard toward applied modeling and experiment rigor, which means your ability to reason about retail-specific constraints (demand forecasting with leakage and drift, evaluating a Target Circle offer model at a 1% positive rate, designing guardrails for a checkout UI test) matters more than textbook recall. Where this gets tricky is the overlap between ML and experimentation: a question about a personalized deal badge in search results quickly becomes a question about whether a click-through lift without a purchase lift is a real win, forcing you to move fluidly between model evaluation and experiment interpretation. Candidates who drill SQL window functions but skip past ranking model validation or MDE reasoning for retail scenarios end up underprepared for the questions that carry the most weight.

Practice Target-style questions covering demand forecasting, Circle offer experimentation, and retail schema querying at datainterview.com/questions.

How to Prepare for Target Data Scientist Interviews

Target is planning to drive more than $15 billion in sales growth by 2030, and their open DS roles map directly to those bets. Job postings call out recommender systems and reinforcement learning for MarTech, search ranking with NLP and LLMs, and last-mile supply chain optimization. Their engineering blog posts on infrastructure showback and platform engineering tell you something else: DS teams here are expected to care about production cost and system reliability, not just model accuracy in a notebook.

Most candidates fumble "why Target" by making it a brand loyalty story. What tends to land better is connecting your ML skills to a specific initiative, like the largest-ever spring beauty assortment with 60+ new brands, and naming the technical challenge it creates (cold-start recommendations for dozens of new brands launching simultaneously, for instance). Read the 2024 annual report, pick two initiatives you'd want to work on, and sketch out what modeling approach you'd try for each.

Try a Real Interview Question

1WITH first_purchase AS (
2  SELECT
3    customer_id,
4    MIN(order_date) AS first_purchase_date
5  FROM orders
6  GROUP BY customer_id
7),
8retained AS (
9  SELECT
10    fp.customer_id,
11    fp.first_purchase_date,
12    CASE
13      WHEN EXISTS (
14        SELECT 1
15        FROM orders o2
16        WHERE o2.customer_id = fp.customer_id
17          AND o2.order_date > fp.first_purchase_date
18          AND o2.order_date <= fp.first_purchase_date + INTERVAL '7 day'
19      ) THEN 1 ELSE 0
20    END AS retained_7d
21  FROM first_purchase fp
22)
23SELECT
24  first_purchase_date,
25  COUNT(*) AS cohort_size,
26  SUM(retained_7d) AS retained_7d,
27  CAST(SUM(retained_7d) AS DECIMAL(18,6)) / NULLIF(COUNT(*), 0) AS retention_7d_rate
28FROM retained
29GROUP BY first_purchase_date
30ORDER BY first_purchase_date;

Retail data science interviews tend to reward candidates who reason about data modeling choices (why this join, why this grain) alongside writing correct queries. From what candidates report, Target's SQL round leans into that pattern. Practice more retail-flavored problems at datainterview.com/coding.

Test Your Readiness

Use datainterview.com/questions to pressure-test the areas where Target's posted roles put the most weight, particularly recommender systems, experimentation design, and causal inference.