Handshake Data Scientist Interview Questions & Guide 2026

The 60-second brief

As a Data Scientist at Handshake, you will play a pivotal role in shaping the career journeys of millions of college students and early-career professionals. Handshake is not a traditional job board; it operates as a complex, three-sided marketplace connecting students, higher education institutions, and employers. Your work will directly influence how these distinct groups interact, ensuring that students find relevant opportunities, universities support their cohorts effectively, and employers discover diverse, qualified talent.

Learn the economics of the product category before the loop. Marketplaces, subscription products and ad-supported products turn on different core quantities (match rate and liquidity, retention and churn, fill rate and yield) and fail in different characteristic ways.

Focus areas

• Turn a vague request into a measurable question
• Pick a randomisation unit that respects interference
• Size an experiment before anyone launches it

Overview

The data challenges at Handshake are unique and highly strategic. Unlike linear platforms, the ecosystem relies heavily on matching algorithms, search relevance, recommendation engines, and trust and safety models. As a Data Scientist, you will dive deep into network effects, marketplace liquidity, and user engagement metrics. You will design and implement the models that power personalized job recommendations, optimize employer outreach strategies, and prevent fraudulent activity on the platform.

Your contributions will have a direct impact on product development and business strategy. You will collaborate closely with product managers, software engineers, and executive stakeholders to translate complex data insights into actionable product features. Whether you are optimizing the search interface or developing predictive models for student engagement, your analytical rigor will ensure that remains the most trusted and effective platform for early-career recruiting.

The Handshake interview process

Handshake candidates report 5 rounds · ≈ 4-6 weeks. The stages below are what candidates describe, not a published process.

Recruiter Phone Screen

Most candidates lose this call inside the first two minutes, during the walkthrough of their own background. The account runs chronologically, sits at the level of tools and titles, and never arrives at a decision anyone could have disagreed with. Anchor on a problem instead of a timeline: what the team could not answer, what you did about it, what happened next. Ninety seconds is enough, and stopping on time leaves room for the half of the call that belongs to you. What you ask about how work gets prioritised signals your level more reliably than the walkthrough does.

What to demonstrate

• Whether your background summary has a shape (problem, decision, consequence) or is a chronological list of tools and employers
• Whether you can account for gaps, short stints and the reason you are looking, unprompted and without hedging
• The substance of the questions you ask back, which an experienced screener reads as a level signal

How to prepare

• Time your opening walkthrough against a clock. If it runs past two minutes, compress the earliest role into a single clause and spend the recovered time on the most recent one
• Write one honest sentence for every gap or short stint visible on your resume and offer it before being asked about it
• Prepare questions about how work arrives and gets prioritised: who writes the request, how often priorities change, and what happens to an analysis after it is delivered

Technical Screening

A handful of shapes account for most of what gets asked in this format: a ranking or deduplication inside groups, a running or rolling total, a period-over-period comparison, and a cohort tracked forward over time. Recognising the shape quickly is most of the speed here; deriving it from scratch while a clock runs is where the time goes. Know that a window function keeps every row while a GROUP BY collapses them, and know which one the question needs. If the exercise is in Python instead of SQL, the same shapes arrive as groupby with transform, shift and merge, and the same grain mistakes are available.

What to demonstrate

• Whether you reach the right construct without a detour, such as ROW_NUMBER over a partition to deduplicate instead of a self-join against a MAX subquery
• Whether you know what your window frame actually is, since adding ORDER BY inside OVER changes the default frame and silently changes a running total
• Whether the thing runs. A near-miss that throws an error scores below a plainer query that returns the right rows.

How to prepare

• Write each of the four shapes once from memory against a small schema and keep the working version somewhere you will reread it: dedupe with ROW_NUMBER, a running total, a month-over-month change with LAG, and a retention table
• Compute one running total twice on data with tied timestamps, once on the default frame and once with ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW, and look at where the two disagree
• If Python is on the table, rebuild the dedupe and the running total with groupby and cumsum, then assert the two implementations return identical rows

Take-Home Project

The clock is part of the test. Three to six hours is not enough to do everything the dataset supports, so the submission mostly reveals how you spend a fixed budget against an open question. A reviewer sees which paths you took and, by absence, which you abandoned. Work that runs out of time inside the analysis ships a thin conclusion, while work that cuts scope early protects the last hour for writing. The most reliable way to lose here is to leave the scoping decision implicit, so it reads as something you missed rather than something you chose.

What to demonstrate

• Whether the scope you settled on is presented as a decision with a reason, rather than left for the reader to infer from what is missing
• Whether the depth of the work is consistent with the stated time budget, instead of several half-finished directions left open
• Whether the closing section reads as something written on purpose rather than assembled from whichever cells survived

How to prepare

• Run a timed rehearsal on a public dataset with a hard stop, holding the final sixty minutes for writing no matter where the analysis has got to
• Before opening the data, list the questions it could plausibly answer, pick one, and keep the discarded ones as a short note on what you did not attempt and why
• Commit a one-line finding after each analysis step so the writeup is assembled from recorded results rather than from memory at midnight

Hiring Manager Session

Underneath the questions about your past work sits a resourcing question. Given four things worth doing and one of you, which gets done and what happens to the rest? Managers ask because that is the daily texture of the job, and because the answer shows whether you rank work by effort or by what it changes. The weak version sorts by personal interest or by whoever asked most insistently. The strong version ties each candidate piece of work to a decision somebody downstream is waiting on, and then names the one you would drop and who you would tell.

What to demonstrate

• Whether you rank work by the decision it unblocks or by how interesting the method is
• How you describe a request you declined, and whether you can say who you said it to
• Whether your sense of how long something takes survives one follow-up question about the messy part
• How you decide something is good enough to hand over unfinished

How to prepare

• Write out your current queue and, next to each item, the decision that stays stalled until it lands. Anything with no waiting decision becomes your example of work you would cut
• Rehearse turning down a plausible stakeholder request out loud, including the smaller alternative you offered instead
• Have one case where you shipped a rough answer early and one where you refused to, with the reason that separated them

Virtual Onsite Panel

Where a loop includes a partner from outside the data team, that conversation usually carries the same weight as the technical ones and gets the least preparation. The person opposite you will not follow a derivation and does not need to. They are working out whether having you involved would make their decisions better or slower. The failure mode is not being too technical. It is answering a question about a decision with a description of your method, leaving the translation to them. What they carry into the debrief is the sentence you handed them, not the analysis underneath it.

What to demonstrate

• Whether a statistical result arrives as something the partner could act on, with the one caveat that would change their decision kept and the rest left out
• Whether you can state what you need from their side, in their terms: instrumentation that does not exist yet, a definition they own, or a holdout they have to agree to
• Whether uncertainty is given as a range someone can plan against, rather than as hedging that invites them to ignore the result
• Whether you ask what decision is actually on the table before explaining anything

How to prepare

• Take a result you know well and write the version for someone who stops reading after one sentence, then the three-minute version, and check the short one is not the long one with the qualifications stripped out
• For a past project, list everything you asked a non-technical partner for and how you phrased it, then rewrite each ask so it names what goes unmeasured without it
• Practise saying where a result does not apply, out loud, in one sentence that a partner could repeat accurately to someone else

Where candidates lose points

Watching an experiment daily and stopping when it crosses significance

A fixed-sample test controls type I error at one pre-declared look. Checking repeatedly and stopping at the first p < 0.05 inflates the false positive rate to roughly 0.15 to 0.20 for ten looks, and it rises further with more frequent checks, because the p-value takes a random walk that will eventually dip below the threshold under the null. The usual defences are a fixed horizon declared before launch, group-sequential boundaries such as O'Brien-Fleming that spend alpha across a planned number of looks, or always-valid confidence sequences that are correct under continuous monitoring. Compounding it, the effect size reported conditional on having crossed the threshold is biased away from zero, and the bias is larger the lower the power was, so an underpowered test that 'won' typically overstates the lift it found.

Comparing cohort retention curves of different maturities, or building the curve from users who are still present

A cohort four weeks old has no week-8 value, so an average taken across cohorts silently drops young cohorts from the later columns and keeps them in the earlier ones. The curve then bends upward at the tail, and the reading that 'retention is improving over time' is an artefact of which cohorts survived to be measured. The same error appears in the denominator when retention is computed over users active in the current period rather than over the full original cohort, which conditions on survival and guarantees a flattering number. The fix is a triangle: fix the cohort at signup, bound every window on both sides, and only compare cells where every cohort has had the full elapsed time, publishing the rest as blank rather than as a partial average.

Comparing periods without accounting for seasonality or day-of-week

Compare whole weeks against whole weeks and check whether the same swing appeared in prior cycles or prior years before attributing it to anything you changed. Weekday and weekend populations often differ enough that a Tuesday-to-Saturday comparison is meaningless.

Solving silently instead of narrating the reasoning

Say which branch you are taking and why you chose it over the alternative, for example checking the denominator first because it changes what the comparison means. A correct answer that arrives with no visible path scores below a rigorous one that needed a hint.

Questions & practice

Rebuild per-visitor ordering without groupby convenience methods

Easy · pandas · vectorisation · window logic

You have a DataFrame of 2 million fct_event rows with visitor_id, occurred_at_utc and event_id, unsorted and containing duplicate timestamps within a visitor. Produce three new columns: event_rank, the 1-based position of the event within its visitor ordered by occurred_at_utc; seconds_since_prev, the gap to that visitor's previous event, NULL for the first; and is_first_for_visitor. You may use sort_values, shift, cumsum, numpy and boolean masking. You may not use groupby.transform, groupby.apply, groupby.cumcount, groupby.rank or merge_asof. Break timestamp ties on event_id.

Approach

1. Sort once by ['visitor_id', 'occurred_at_utc', 'event_id'] and reset the index. The whole exercise reduces to row arithmetic on a sorted frame, and the tiebreak on event_id is what makes the result reproducible across runs.
2. Mark visitor boundaries with is_first = df['visitor_id'].ne(df['visitor_id'].shift()). This is the single fact every other column derives from.
3. Compute seconds_since_prev as the diff of the timestamp column, then overwrite it with NaT/NaN wherever is_first is True. The shift crosses the boundary between visitors and will otherwise hand the first row of each visitor the last event of the previous one.
4. Build event_rank from a running counter that resets at boundaries: take a global cumulative position (np.arange(len(df))) and subtract, per row, the global position at which that visitor started. Get the start position by forward-filling the positions where is_first is True, which is a cumsum-free reset and is O(n).
5. Verify against the forbidden method once, as a test rather than as the implementation, and confirm the two agree on every row.

Follow-up

• The frame does not fit in memory. How does your approach change if you can only process one visitor-partitioned chunk at a time?
• occurred_at_utc is client-supplied and sometimes runs backwards within a visitor. Does your seconds_since_prev go negative, and should it?
• How would you extend this to reset the counter at every change of surface as well as visitor?

Implement seven-day activation from its written definition

Medium · metric definition · cohorts · pandas · joins

Implement the seven-day activation rate. Inputs: dim_user with user_id, account_created_at_utc and is_internal; fct_event with user_id, occurred_at_utc and is_core_action. A user activates when core-action events carrying a non-NULL user_id fall on at least two distinct UTC dates inside [account_created_at_utc, account_created_at_utc + 7 days). The denominator is every non-internal user whose account_created_at_utc lands in the cohort week, including users with no events at all. Return one row per cohort week with numerator, denominator and rate, publishing only weeks whose last signup is at least eight days old.

Approach

1. Build the denominator first, from dim_user alone, filtered on is_internal = False. Deriving it from the join is the standard way to lose every user who never fired an event, which is exactly the population the metric is about.
2. Join events to users on user_id with a left join from the user side, then apply the window as a half-open interval: occurred_at >= created AND occurred_at < created + 7 days. The right bound is exclusive, so an event at exactly created + 7 days does not count.
3. Count distinct UTC dates per user, not distinct events. Floor occurred_at_utc to date before the nunique, and do it in UTC rather than local time so the threshold does not move with the user's country.
4. Apply the >= 2 threshold, aggregate to cohort week, and compute the rate by re-summing numerator and denominator per week rather than averaging any per-user or per-day rate. Fix the week anchor explicitly: cohort_week is the Monday of the signup week in UTC, which is what Postgres DATE_TRUNC('week') returns and what any SQL version of this metric will produce. In pandas, subtract dt.weekday days from the floored timestamp. If you reach for periods instead, the anchor that matches is to_period('W') (equivalently 'W-SUN'), whose weeks end Sunday and therefore start Monday; to_period('W-MON') labels weeks that end on Monday, so it runs Tuesday through Monday and its start_time is a Tuesday. Mixing the two shifts every cohort label by one day and silently moves Mondays into the previous week.
5. Suppress immature weeks: drop any cohort week whose maximum account_created_at_utc is within 8 days of the data cut, and return them as absent rather than as a partial number.

Follow-up

• The threshold is 2 distinct days. What changes in the reported history if someone moves it to 3, and how would you publish that change?
• Invited seats and SSO-provisioned users have no pre-signup session. Should they be in this denominator at all, and what does including them do to the rate for sales-assisted accounts?
• How would you produce the same metric at account grain, and which of the two would you put on the dashboard?

Audit a one-day event extract for structural defects

Easy · data quality · late arrival · pandas

You receive a one-day extract of fct_event as a DataFrame with event_id, occurred_at_utc, received_at_utc, visitor_id, user_id, account_id, event_name, is_bot_flagged and surface. Write a function returning one row per data-quality rule with the rule name, the failing row count and the failing share of the extract. Cover at minimum: duplicate event_id, received_at_utc earlier than occurred_at_utc, occurred_at_utc later than the extract's maximum received_at_utc, account_id present while user_id is NULL, and rows whose occurred_at date differs from their received_at date. Do not drop rows; report only.

Approach

1. Compute the extract's own reference clock first: max(received_at_utc). Wall-clock now() is wrong here because the extract may be replayed days later, which would turn every row into a future-dated failure.
2. Express each rule as a boolean Series over the same index so the checks compose, then aggregate with .sum() and divide by len(df). Building a list of (name, mask) pairs keeps the rule set extensible and keeps one code path for counting.
3. For the duplicate rule, decide and state the convention: df.duplicated('event_id', keep=False).sum() counts every member of a duplicated group, df.duplicated('event_id').sum() counts only the surplus copies. Either is defensible; an unstated choice is not. The rest of this item assumes keep=False.
4. Treat received_at < occurred_at as clock skew, not corruption: occurred_at is client-supplied. Separate it from the date-mismatch rule, which is the one that actually breaks a daily metric keyed on occurred_at.
5. Know which rules imply which before you read the counts. A row whose occurred_at exceeds max(received_at_utc) has its own received_at no later than that maximum, so it is necessarily a clock-skew row as well: the future-dated mask is a subset of the skew mask, always. Neither is a subset of the date-mismatch mask, because skew of a few minutes inside one UTC date mismatches nothing.
6. Return a tidy DataFrame sorted by failing_share descending, and add a boolean column saying whether the rule should block publication, so the output is a decision rather than a list of numbers.

Worked solution (20 min)

1. Parse both timestamp columns with utc=True and assert the dtype, since a silently-object column makes every comparison string-wise and wrong.
2. Set ref = df['received_at_utc'].max() and build the five masks against it.
3. Assemble results as pd.DataFrame(rows) with columns rule, failing_rows, failing_share, blocks_publication.
4. Keep the masks addressable (a dict of name to Series) rather than only their sums, so the overlap between rules can be asserted rather than assumed.
5. Verify the function is pure: assert the input frame's shape is unchanged after the call.

Expected result: A DataFrame with one row per rule, failing_share equal to failing_rows divided by len(df) for every row, and the input frame returned unmodified. The rules overlap rather than partition the extract, and one containment is structural: every future-dated row is also a clock-skew row. The counts therefore must not be summed or presented as a total.

Verify it yourself

• Plant violations deliberately and assert an exact expected vector rather than assuming one hit per rule. On a 10-row frame carrying two rows that share an event_id, one same-date skew row, one row whose occurred_at sits an hour past max(received_at_utc), one row with account_id set and user_id NULL, and one row whose occurred_at falls on the previous UTC date: duplicate_event_id is 2 under the keep=False convention, clock_skew is 2 (the planted skew row plus the future-dated row, which cannot avoid tripping it), future_dated is 1, orphan_account is 1, date_mismatch is 1.
• Assert the containment directly rather than by counting: (future_dated_mask & ~clock_skew_mask).sum() is 0. A non-zero value means the reference clock was not max(received_at_utc).
• Feed a clean frame and confirm all counts are 0 and no rule errors on empty masks.
• Confirm len(input) is identical before and after, since the deliverable is a report and not a cleaned frame.

Follow-up

• The date-mismatch count is 2.1 percent on this extract. What late-arrival rule would you write for a daily metric, and how many days would you hold the number open?
• Duplicate event_id values appear only on the 'core_action_completed' event. What upstream cause would you check before deduplicating?
• Which of these rules should fire an alert at the pipeline, and which should only appear in a weekly review?

Write a SQL query to calculate the month-over-month growth rate of active students on the platform.

Medium · sql and data manipulation

Approach

1. Check whether any join is one-to-many before aggregating, or the sums inflate.
2. Compute rates by summing numerator and denominator separately, never by averaging rates.
3. State the window function and its partition and ordering out loud before writing it.

Follow-up

• What breaks if events arrive late or out of order?
• How would you verify this result without re-running the same query?

Given a table of student job applications, write a Python script to identify the top three industries each student interacts with.

Medium · sql and data manipulation

Approach

1. Check whether any join is one-to-many before aggregating, or the sums inflate.
2. Compute rates by summing numerator and denominator separately, never by averaging rates.
3. Handle the rows that do not match: a LEFT JOIN with a NULL check is usually the question.

Follow-up

• How would you verify this result without re-running the same query?
• What breaks if events arrive late or out of order?

Write a SQL query to identify "dormant" employers who have posted jobs but received zero applications in the last 30 days.

Medium · sql and data manipulation

Approach

1. Handle the rows that do not match: a LEFT JOIN with a NULL check is usually the question.
2. State the window function and its partition and ordering out loud before writing it.
3. Compute rates by summing numerator and denominator separately, never by averaging rates.

Follow-up

• How would you verify this result without re-running the same query?
• What breaks if events arrive late or out of order?

Explain how you would optimize a slow-performing SQL query that involves multiple joins on high-cardinality user tables.

Medium · sql and data manipulation

Approach

1. State the window function and its partition and ordering out loud before writing it.
2. Compute rates by summing numerator and denominator separately, never by averaging rates.
3. Check whether any join is one-to-many before aggregating, or the sums inflate.

Follow-up

• How would you verify this result without re-running the same query?
• How does the query change if the join becomes one-to-many?

Attribute account MRR to first-touch channel without seat fan-out

Medium · fan-out · joins · revenue · attribution

dim_user holds user_id, account_id, first_touch_channel, account_created_at_utc. dim_account holds account_id, lifecycle_status. fct_subscription_period holds account_id, period_start_utc, period_end_utc, period_status, mrr_cents_constant_fx. For a given as-of date, report active MRR by the first-touch channel of the account's earliest-created user. An account has many users and many periods. The channel totals must sum exactly to total active MRR on that date, and accounts whose label is NULL must appear as their own bucket.

Approach

1. Reduce the label side to one row per account before any join: ROW_NUMBER() OVER (PARTITION BY account_id ORDER BY account_created_at_utc, user_id) = 1 over dim_user, carrying first_touch_channel. Joining dim_user to the period table directly repeats each revenue row once per seat.
2. Reduce the money side to one row per account too: the period where period_start_utc <= :as_of AND period_end_utc > :as_of AND period_status IN ('active','past_due'). If more than one row survives that filter the account holds overlapping subscriptions, which must be deduped or deliberately summed, and either way the choice is stated rather than left to the join.
3. Join the two one-row-per-account CTEs and SUM(mrr_cents_constant_fx). Use the constant-FX column, since the same query run in two quarters otherwise reports currency moves as channel performance.
4. Bucket NULL first_touch_channel as 'unattributed' with COALESCE instead of letting it drop. Roughly the share of users with no observed touch is not small, and dropping it understates the total and inflates every named channel's share.
5. Reconcile: the SUM over the joined result must equal the SUM over the period CTE alone. That single equality check catches fan-out immediately, which is why it is worth writing before presenting the number.

Worked solution (30 min)

1. Compute total active MRR on the as-of date from the period table alone and write the number down as the target.
2. Build the label CTE and assert COUNT(*) = COUNT(DISTINCT account_id).
3. Build the as-of period CTE and assert the same, investigating any account with more than one surviving row.
4. Join, COALESCE the channel, aggregate, and compare the total to the target.
5. Deliberately run the naive dim_user join and record the inflation factor, which is the MRR-weighted mean seat count.

Expected result: One row per first_touch_channel plus an 'unattributed' bucket, whose MRR values sum exactly to the independently computed total active MRR for the as-of date.

Verify it yourself

• Channel total equals the standalone total from the period CTE, to the cent.
• Both CTEs satisfy COUNT(*) = COUNT(DISTINCT account_id).
• The unattributed bucket is present and non-zero rather than silently absent.
• The naive join version produces a strictly larger total, and you can state the ratio.

Follow-up

• Which user should own the account's label: earliest created, the 'owner' role, or the one who converted? What does each choice systematically favour?
• What does this table entitle you to say about the value of a channel, and what does it not?
• Some accounts have a parent_account_id. How does rolling those up change both the numerator and the channel mix?

A test that won on day four and lost by day twelve

Medium · peeking · sequential testing · multiple comparisons · winner's curse

A two-week test was checked every morning. On day four the primary metric crossed p = 0.03 and the team shipped. By day twelve, with logging still running, the estimate had fallen to a 0.2% lift at p = 0.61. The day-four deck also reported a significant 9% lift among paid_search visitors on mobile_web, one of twenty segments inspected. Explain both patterns quantitatively, say what the day-four evidence was actually worth, and specify the monitoring rule that prevents a repeat.

Approach

1. Price the peeking. A fixed-sample test controls type I error at one pre-declared look. Ten daily looks at a nominal 5% raise the false positive rate to roughly 0.15 to 0.20, because the running test statistic takes a random walk that will eventually cross the boundary under the null.
2. Price the winner's curse separately, because it is a different error. Conditioning an estimate on having crossed the threshold biases it away from zero, by a factor of roughly 3 when true power is 20% and only about 1.1 at 80% power. Day four sits at roughly 29% of planned information, which is exactly the low-power case.
3. Read day twelve as the better estimate rather than as a contradiction. Nothing regressed; the day-four value was an extreme draw that was selected precisely for being extreme.
4. Price the segment claim. Across twenty segments at a nominal 5%, at least one crosses about 64% of the time under a global null, since 1 - 0.95^20 = 0.64, and the slice that surfaces is by construction the one with the most favourable noise.
5. Specify the rule instead of scolding. Either fix the horizon and look once, or use group-sequential boundaries such as O'Brien-Fleming that spend almost no alpha early, or use always-valid confidence sequences that stay correct under continuous monitoring. Pre-register the small set of segments, correct across the ones tested, and treat anything found afterwards as a hypothesis needing its own powered test.

Worked solution (25 min)

1. Count the looks actually taken, twelve daily, and state the inflated alpha band of 0.15 to 0.20 for ten looks, rising with more.
2. Place day four on the information fraction, about 4/14 = 29% of planned sample, and apply the exaggeration factor for low power.
3. Compute 1 - 0.95^k for the number of segments actually inspected; with k = 20 this is 0.64.
4. Report the day-twelve interval as the estimate and say plainly that the experiment supports no effect at any size the team would ship for.
5. Write the rule into the experiment template: horizon and segments declared at launch, sequential boundaries if interim looks are needed at all.

Expected result: The day-four result is consistent with pure noise: ten looks inflate the false positive rate to roughly 0.15 to 0.20, and an estimate conditioned on crossing at about 29% information is exaggerated by roughly a factor of three at low power. The segment finding appears about 64% of the time across twenty slices under a global null. The supportable conclusion is a 0.2% lift whose interval covers zero.

Verify it yourself

• 1 - 0.95^20 = 0.6415. If your family-wise number lands near 0.05 you have computed the per-test rate instead.
• The day-twelve standard error should be roughly sqrt(12/4) = 1.7 times smaller than day four's. If the interval did not narrow, sample was not accumulating and there is a second problem.
• A Bonferroni threshold for twenty segments is 0.0025; check whether the mobile_web result would have survived it.

Follow-up

• Under O'Brien-Fleming with five planned looks, roughly what nominal p-value would the first look have needed to stop?
• The feature already shipped. What analysis can you still run, and what can it not tell you?
• If mobile_web paid_search were genuinely the only affected segment, how would you design the confirmatory test and how large would it have to be?

Halve the runtime of a flat test with pre-period data

Medium · cuped · variance reduction · stratification

A test on 28-day core actions per user is powered for six weeks and the team will not wait. You have fct_event (user_id, occurred_at_utc, is_core_action) covering the period before each unit's first_exposed_at_utc in fct_experiment_exposure, and the pre-period count correlates with the in-experiment count at r = 0.6. Apply CUPED, state precisely how much runtime it buys, and handle the units with no pre-period because they signed up after the experiment started.

Approach

1. Define the adjusted outcome: Y_cuped = Y - theta (X - Xbar), with theta = Cov(Y, X) / Var(X) estimated on the pooled sample so theta itself carries no treatment effect. It is unbiased because X is pre-exposure and therefore has equal expectation in both arms.
2. Quantify the gain exactly: Var(Y_cuped) = Var(Y)(1 - r^2) = Var(Y) x 0.64, so 36% of the variance is removed. Required sample scales with variance, so six weeks becomes 3.8 weeks, and at the original sample the MDE shrinks by sqrt(0.64) = 0.8.
3. State the precondition that actually bites: X must be measured strictly before first_exposed_at_utc, not before assigned_at_utc and not before the calendar start of the experiment. Any post-exposure information in the covariate lets the treatment effect leak in and biases the estimate rather than merely failing to help.
4. Handle missing pre-periods as a stratum, never by dropping rows. Dropping changes the population; setting X to the pooled mean leaves those units' adjustment at zero, which is unbiased but buys them no reduction, so report the covered share and expect the blended reduction to fall short of 36%.
5. Note that stratification is the same mechanism with a categorical covariate. Post-stratifying on device_type and referrer_channel removes the variance those stratum means explain, and it composes with CUPED provided the strata are also fixed before exposure.

Follow-up

• Here the covariate is the same metric as the outcome. When would you deliberately pick a different one, and what is the risk if that covariate is itself affected by treatment?
• The point estimate moves by 0.4 standard errors when CUPED is applied. Is that reassuring or alarming, and what would you check?
• How does CUPED combine with cluster randomisation on account_id, and at which grain is the covariate built?

Separate a definition edit from a behaviour change

Easy · definition-change · metric-versioning · cohorts

Seven-day activation rate is documented as core actions on at least 2 distinct UTC dates inside the first seven days after account_created_at_utc, and every cohort published through last week was computed on that definition. The dashboard now reads 38% for the signup cohort that started Monday, against 31% for the cohort before it. Two things landed in that week: an onboarding change, and a commit to the metric's SQL that you have not read yet and that came with no documentation update. You have dim_user, fct_event and the metric's version history. Deliver two numbers: how many of the 7 points are behaviour and how many are the definition. Then say which published cohorts are no longer comparable.

Approach

1. Diff the metric SQL across the commit boundary and enumerate every changed predicate individually: the distinct-day threshold, the is_internal exclusion, the is_core_action source, the window bounds, and the identity key. A single commit routinely changes more than one, and each has its own sign.
2. Dual-run both definitions over the same 26 cohort weeks. The definition effect is the level shift between the two series computed on identical data, which isolates it from anything that happened in the world.
3. Read the behaviour effect only from the old-definition series, comparing the new cohort to its own trailing cohorts. The new-definition series cannot answer the behaviour question because it has no pre-period.
4. Explain the mechanism of the changed predicate rather than only its size. Lowering a distinct-day threshold from 2 to 1 admits every user who acted once in a single session, which is a large and low-intent population, so a several-point jump is the expected magnitude and not a surprise.
5. State the restatement plan: pick one definition, backfill the full history on it, and annotate the break date. A series carrying two definitions is not a time series.

Follow-up

• The onboarding team wants the 1-day threshold kept because it is easier to move. What is your argument, and what would change your mind?
• How would you make a definition change visible to every consumer of this metric without relying on people reading a changelog?

Suggested practice checklist

Instead of guessing where the week should go, day one measures it under a fixed rubric and allocates the remaining hours in proportion to the gaps. The method is deliberately rigid: the allocation is written down before any studying starts and is not renegotiated when a topic turns out to be unpleasant.

Day 1 — Diagnostic, scored before you study anything

• Sit a 100-minute timed diagnostic in four blocks: 30 minutes of SQL across three prompts, 25 minutes of short-answer statistics, 25 minutes on one modelling or case prompt, and 20 minutes delivering one behavioural story aloud.
• Score each block from 0 to 3 on a fixed rubric where 3 is correct and fluent, 2 is correct but slow or prompted, 1 is partially correct, and 0 is stuck, grading the output rather than how the attempt felt.
• Allocate the hours for days two to five roughly in proportion to 3 minus the score in each block, write the allocation down, and commit to not revising it midweek.

Deliverable: A scored rubric and a fixed hour allocation for the rest of the week.

Day 2 — Largest gap: find the boundary rather than the subject

• Break the weakest area into five named sub-skills (for query work: grain control, window frames, date arithmetic, set logic with NULLs, and reading a query plan) and rate each one, so the rest of the week targets a sub-skill instead of a subject.
• Solve three problems chosen to sit just above where the rating drops off, and for each write the first move you failed to make.
• Re-solve one of them from memory four hours later, on paper, with nothing open.

Deliverable: A five-item sub-skill map with the two blocking sub-skills circled.

Day 3 — Largest gap: drill the blocking sub-skill

• Do eight short repetitions of the same shape rather than eight different problems, so what you practise is the pattern and not the puzzle.
• Write the rule you now hold in one sentence, then test it against a case built to break it: a ranking function over a column with ties, or a two-sample test on observations that are obviously dependent.
• Have someone else read your one-sentence rule and find the precondition you left out.

Deliverable: One rule statement with its preconditions attached and one counterexample that would have caught the incomplete version.

Day 4 — Second gap, plus maintenance on your strongest area

• Run the same sub-skill map and boundary protocol on the second-largest gap, compressed into half the day.
• Spend 25 timed minutes on your strongest area to stop it decaying, choosing the hardest problem you can still finish rather than an easy warm-up.
• Compare how the two areas fail: whether you lose time on recall, on setup, or on arithmetic, because the fix differs for each.

Deliverable: A second sub-skill map plus a one-line diagnosis of how each area fails you.

Day 5 — The gap that is not a skill

• Record yourself answering one technical and one behavioural prompt, then count two things in the playback: how many seconds before your first clarifying question, and how many sentences you started without knowing where they ended.
• Rewrite your three most-used stock phrases into shorter versions, and practise saying "I do not know, here is how I would find out" without softening it into a guess.
• Deliver one answer again with a hard 90-second limit to force structure before detail.

Deliverable: Two recordings with a counted improvement in time-to-first-question.

Day 6 — Retest under day-one conditions

• Sit the same 100-minute diagnostic structure with new prompts of comparable difficulty and score it on the identical rubric.
• Compare block by block, and for any block that did not move, change the method rather than adding hours: a block stuck at 1 usually means the practice was too varied, not too short.
• Write which single block you would still lose the offer on.

Deliverable: A second scored rubric placed next to the first, with one named remaining risk.

Day 7 — Full loop under interview conditions

• Run a 60-minute mock covering the two blocks that moved least, with an interviewer instructed to interrupt and change direction.
• Write your recovery script for the moment you go blank: restate the question, state your assumption, name the first thing you would check.
• Reduce the week to the rule statements you wrote, each with its preconditions attached, then say every one of them out loud without reading it and cut any you cannot state in a single sentence, since a rule you have to reconstruct mid-answer will not survive being interrupted.

Deliverable: A one-page card holding the recovery script and only the rules you could state from memory.

Getting an analysis acted on

Work that nobody used is a common and unflattering pattern in data careers, and interviewers probe for it. Have a story about an analysis that changed a decision, and be specific about how you got it in front of the person who could act. Also have one about work that went nowhere, with your reading of why.

Defend a flat experiment readout against a post-hoc segment

Medium · experiments · segmentation · pushback

A feature you evaluated is flat on seven-day activation: +0.05pp with a 95% interval of [-0.47pp, +0.57pp], from 61,000 exposed users per arm in fct_experiment_exposure joined to dim_user and fct_event. Baseline activation is 32%. The launch team asks you to drop every surface except mobile_web, where the point estimate is +1.1pp, and re-run. You have ten minutes in their planning meeting. Deliver a spoken position: what you will and will not do, and the decision you recommend.

Approach

1. Recognise what is being probed: whether you hold a statistical position under social pressure without becoming either rigid or apologetic. A generic answer says the segment is not significant; a strong one separates the request into a question that is answerable (is the mobile_web number real?) and one that is not (can we ship on it?), and answers both.
2. Price the multiplicity out loud. The slice was chosen after seeing the results, so its estimate is selected on favourable noise and is biased away from zero. With k independent looks at a nominal 5% level, the chance of at least one false positive is 1 - 0.95^k: 26% at six segments, 64% at twenty. Quote the k you actually inspected, not the k you reported.
3. Use the arithmetic already in front of you. On the point estimates, a +1.1pp mobile_web effect combined with a pooled +0.05pp implies the remaining surfaces average negative in proportion to mobile_web's share of exposures. State that as a testable implication of their story rather than as a rebuttal of it.
4. Ask the one question that settles the category: was mobile_web named in the analysis plan before launch? If it was, it is a planned comparison and gets a corrected reading. If it was not, it is a hypothesis, and the honest move is to size the test that would confirm it.
5. Convert the refusal into a cost. Size a mobile_web-only confirmatory test at the claimed effect, state the weeks of mobile_web traffic it needs, and close with the recommendation: do not ship this as a lift, and note that the interval already rules out anything at or above +0.6pp, which is itself a useful input to the roadmap.

Follow-up

• The confirmatory test you sized needs nine weeks of mobile_web traffic and the team has three. What do you recommend instead?
• Suppose mobile_web was pre-registered. How does your reading change, and what correction do you apply?
• Your interval excludes +0.6pp. Is that the same as saying the feature does nothing?

Handle a request for numbers supporting a decision already made

Hard · integrity · framing · stakeholder

A senior leader has already decided to sunset a plan tier and asks you for the analysis showing it is the right call. Accounts on that tier carry 6% of MRR at constant FX and have the highest licensed-seat utilisation in the book. The leader's support matters to your next review cycle, and the decision is being presented in four days. Deliver what you produce, what you decline to produce, and the exact sentence you will say in the meeting where the number appears on a slide.

Approach

1. Recognise what is being probed: whether you can find the legitimate request inside an illegitimate framing instead of either complying or refusing on principle. The generic answer promises to push back; the strong one produces something genuinely useful and states its limits in the room, without ambushing anybody.
2. Separate the decision from the justification. Sunsetting the tier may be correct for reasons the data does not hold, such as support cost, roadmap surface area or sales motion. What you decline is a one-sided document. What you produce is the case read both ways, which also happens to be more useful to the leader.
3. Build the symmetric analysis: MRR at risk at constant FX, the share of affected accounts with a plausible migration path given seats_licensed and billing_term, the recovery rate assumed for that migration and where it came from, and the downside case in which high-utilisation accounts treat the sunset as a reason to re-evaluate the vendor entirely.
4. Surface the inconvenient fact privately and early. The highest seat utilisation in the book is a retention signal, and the leader should hold it before the room does, so they can incorporate it rather than be caught by it.
5. Agree the meeting sentence in advance with the leader, so that nobody is surprised. Something to the effect that the tier is 6% of MRR and its accounts are the most heavily used in the book, and that the case for sunsetting rests on cost and focus rather than on revenue. That is true, it supports the decision on its real grounds, and it stops the deck claiming the numbers endorse it.
6. Decide your own line before you need it: what you will not put your name to, and that the route if asked anyway is your own manager rather than a confrontation in the meeting.

Follow-up

• The deck circulates with your analysis included and the downside case removed. What do you do, and by when?
• What changes if the honest analysis says the sunset is clearly the wrong call?
• How do you write the same memo when the leader is your skip-level and the meeting is tomorrow?

Disagree with a product manager's roadmap claim using data

Medium · causal inference · selection bias · stakeholder

A product manager proposes building a feature on the argument that accounts connecting an integration in week one retain three times better at week four. The figure is correctly computed from dim_user and fct_event, and it has already been shown to leadership. You have one scheduled 1:1 before the roadmap locks. Deliver the specific analysis you would run to test whether the relationship is causal, the result that would change your own mind, and how you open the conversation so that the PM is not put in the position of defending the number in public.

Approach

1. Recognise what is being probed: whether you can separate a number being right from an inference being wrong, and do it without costing the PM face. The generic answer recites that correlation is not causation; the strong one names the specific confound and proposes the cheapest design that could distinguish the explanations.
2. State the alternative concretely. Accounts that connect an integration in week one are accounts that already have a workflow and a technical owner, so week-one intent plausibly drives both the connection and week-four retention. The selection is on intent, which no amount of post-hoc adjustment observes.
3. Order the discriminating analyses by cost. First, condition on pre-connection activity by comparing retention within strata of week-one core-action count, which removes the crude version of the confound but not unobserved intent. Second, look for variation in integration availability that was unrelated to intent, such as a staggered release or an outage window. Third, an encouragement design that randomises a prompt to connect and reads the intent-to-treat effect on week-four retention, which is the only version that identifies an effect.
4. Run the timing check, because it is nearly free and it is the most persuasive single piece of evidence. If the retention advantage among connectors is already visible before any of them connected, the causal story is largely finished.
5. Pre-commit to what would change your mind and say it before you show anything: if the gap survives stratification and the encouragement arm moves week-four retention at all, the feature has a case and you will say so.
6. Open the 1:1 by agreeing with the true part, that the correlation is real and worth chasing, then ask what effect size the roadmap plan assumes. That makes the size of the claim the topic instead of its authorship.

Follow-up

• The encouragement test needs six weeks and the roadmap locks in two. What do you recommend in the interim?
• Stratifying on week-one activity closes half the gap. What do you conclude, and what do you still not know?
• How would you word this in the roadmap document so the PM's original number is reframed rather than deleted?