Background
I have ~3.5 years as analytics at a large e-commerce company. There I worked on supply chain and most recently on metric definitions and SLAs for a new business vertical. I applied for the Lead Data Analyst (L6) role in Target's Digital Data Analytics team. The role supports Target Plus, Target's third-party marketplace, and sits between business analytics and product analytics.
How I Got the Interview
I applied through a strong referral. HR called me within about two weeks, and all the interviews were set up within the following week.
Timeline
Rounds
Round 1 — Analytics Case Study & SQL
Format: Video call | Duration: ~60 min | Interviewer: Lead, Product & Item Analytics (Target Plus)
The interviewer opened with the role: a marketplace-focused lead analyst who works on both business and product problems. Then I introduced myself.
Q1 — Walk me through a recent project end to end
I talked about an AI knowledge orchestrator: a RAG system over each analytics team's context documents (metric definitions, SQL snippets, tables, past analyses). He kept interrupting with follow-ups:
- Leadership wanted "ask a question, get the answer". Why did we scope it down to giving context instead?
- If the LLM must retrieve SQL rather than generate it, will you ever have 100% coverage of the queries people ask? (No. We generate from retrieved snippets, then run a syntax check and a check that blocks queries pulling large amounts of raw data.)
- When the query runs, does the data come back into the LLM? How do you control token and cloud cost?
- Were the end users analysts only, or stakeholders too?
- Did you build a semantic layer on the tables, or query the raw tables?
- How did you evaluate it? (Retrieval: recall@5 and precision@5 against a golden question set per team. Generation: a separate LLM as judge.)
Q2 — Gross revenue is down 8% over the last two weeks in the WBR. How do you find the root cause?
- First, check whether 8% is outside the metric's normal band, and check the data source (row counts, nulls, duplicates).
- Clarify how gross revenue is calculated. He said to keep it as overall revenue, with no cancellation or return netting.
- Set a funnel baseline from the two weeks before (traffic → product page views → add to cart → checkout → order) and find which step dropped.
- Hypotheses per step: banners and recommendations, price, serviceability, delivery time, payment failures, promotions.
- External factors: coming off a sale period, competitor promotions, seasonality.
Q3 — Only ~20% of marketplace items have ratings and reviews. Size the GMV opportunity of raising coverage to 25%.
He gave a hint: you need to tie reviews to conversion. I proposed comparing similar items (same category × price band, similar delivery speed and brand tier) with and without reviews, normalizing for traffic, and measuring the conversion lift with a significance test or diff-in-diff. Ideally I'd confirm it with an A/B test on a small share of traffic. He called this a synthetic-control approach.
Q4 — Rate yourself on SQL, Power BI/Looker, Python and statistics
I gave SQL 4/5, Python strong, statistics 3–3.5/5. For dashboarding I was honest: I've designed the views and metrics but haven't built many dashboards hands-on.
Q5 — Have you worked on A/B tests? How do you pick primary, secondary and guardrail metrics? What is MDE?
I've evaluated tests (test vs control on user IDs), but the product team set them up. For a bigger add-to-cart button I said: primary metric add-to-cart rate; guardrails conversion and cart abandonment; secondary metrics GMV and orders per customer.
Q6 — A table has columns A, B, C, D. (A, B) should be unique, D is the insert timestamp, and (A, B) is duplicated. How many duplicates are there, and how do you keep only the latest row?
- Count:
GROUP BY A, B HAVING COUNT(*) > 1. I first went to concatenating A and B plus a CASE flag; he nudged me to the simpler version.
- Dedupe:
ROW_NUMBER() OVER (PARTITION BY A, B ORDER BY D DESC) and keep row 1.
Q7 — Difference between RANK, DENSE_RANK and ROW_NUMBER?
Where I got stuck: Some of my answers on the SQL optimization and on RANK vs DENSE_RANK came out muddled. RANK skips after a tie (1, 1, 3); DENSE_RANK does not (1, 1, 2). The A/B testing and MDE questions also showed I had more theory than hands-on experience.
Verdict: Cleared
Round 2 — Problem Solving, Case Studies & Guesstimate
Format: Video call | Duration: ~60 min | Interviewer: Head of Target Plus Analytics
He asked me to think out loud, and to drop my supply-chain lens and think like business and product.
Q1 — What AI use cases have you worked on?
The same knowledge orchestrator. He also asked how we got documentation out of teams in the first place.
Q2 — Conversion has dropped week over week for four weeks. How do you find the cause?
- Check the data, then the numerator vs denominator, then any festive period or promotion just before the drop, then the funnel against a baseline.
- Follow-up: users land on the homepage but don't reach product pages. Which metric shows that? I proposed tracking the position of the first click (how deep the user scrolls before clicking).
- Follow-up: what if nothing is shown at all? (The answer he wanted was null searches.)
- Follow-up: product, search and external factors are all fine. What else? (Mix shift: category mix on the homepage changes. Mobiles convert very differently from fashion, so the overall rate drops even when each category holds steady.)
Q3 — McDonald's wants its first outlet in a tier-2 city. How do you pick the location?
- Baselines from other tier-2 stores if any exist. If not, look at where competitors are (clustering), the target audience (colleges, IT parks), drive-in potential, and the share of orders coming through delivery apps.
- Follow-up: you've shortlisted 10 locations. How do you pick one? (I said weight each factor's impact on revenue, e.g. a regression. He pushed back that this needs training data; I suggested competitor data and surveys.)
- Follow-up: how do you set a realistic delivery time to advertise? (A floor from the fastest possible kitchen time plus travel time to the densest area, then trade staffing cost against what customers expect.)
Q4 — Guesstimate: how many four-wheelers pass through Silk Board junction in a day?
I went top-down: Bangalore population → people who can afford a car → car owners → minus carpooling → share commuting via the ORR → minus work-from-home. That gave ~7 lakh.
Where I got stuck: I left out taxis and cabs entirely. When he asked for a second approach to cross-check the number, I couldn't find one. He pointed to the one I'd joked about at the start: count from the junction's capacity (signal cycles × green time × lanes × car length).
Verdict: Cleared
Round 3 — Experience Deep-Dive
Format: Video call | Duration: ~30 min | Interviewer: Lead, Media Analytics
Q1 — Why this role, and why leave now?
Q2 — Walk me through one or two complex analytical problems you've solved
I started with the RAG project. He redirected me: he wanted something analytically complex that drove a business decision, not an AI build. So I covered a late-delivery prediction model:
- Which features? (Seller dispatch time, hub turnaround time, lane history, rolling facility performance, delivery-hub capacity and manpower variance.)
- How do you run it at that scale across all nodes? (A pilot on selected lanes, with hub metrics encoded as numeric features in each row.)
- Was it deployed? (Yes, to send customers a proactive delay message at the last leg.)
- How did you handle false positives? (The classes are imbalanced. Recall was ~90% and precision ~60–70%, and per-node thresholds kept early alerts from triggering cancellations.)
Q3 — Tech stack, how hands-on are you, and any A/B testing?
BigQuery, Looker Studio, Python. A/B work was mostly pre/post and diff-in-diff analysis.
Verdict: Unknown
Final Outcome
The process was ongoing at the time of writing.
Work Culture
- Target's online business is about 20% of total sales. The digital analytics team works like a smaller e-commerce company inside Target.
- There's no separate dashboarding team. Leads and senior analysts spend about 30–40% of their time building dashboards.
- Analysts own the AI work end to end: custom GPTs over datasets for business teams, automated drivers for the WBR, and market-trend scraping. They're moving to Copilot.
- It's an 80+-year-old company with a lot of legacy systems, so the AI push is "get the data house in order first".
- The lead is expected to own charters end to end (pricing, promotions, item analytics, seller onboarding), find white-space problems, and mentor 1–2 analysts.
What I'd Do Differently
I'd revise window functions properly. I'd have a hands-on A/B story ready (MDE, power, sample size). I'd keep a second, non-AI analytics project ready to lead with. And I'd practise triangulating guesstimates from both supply and demand.
Tips for Future Candidates
- Expect at least one dashboarding question. The role needs hands-on BI.
- For any metric-drop RCA: data quality → metric definition → funnel baseline → internal vs external factors → mix shift.
- Know A/B basics cold: primary, secondary and guardrail metrics, and MDE.
- In guesstimates, list every segment (taxis!) and have a second method to cross-check.
- Bring an analytics project that changed a business decision, not just a technical build.