Skip to content
Tech Interview Prep
Technical interview guide

Self-Service Analytics Enablement

Letting business users answer their own questions safely, without a queue of ad-hoc requests to the data team.

Read
45 min
Practice MCQs
25
Interview Q&A
25
Edition
v4
Editorial status
Reviewed

Scope: Current Microsoft Power BI adoption roadmap, Tableau Blueprint, Looker, dbt MetricFlow, OpenMetadata, BigQuery, Snowflake, and WCAG 2.2 guidance reviewed 2026-09-04.

Curated: · Written: · Reviewed:

25 questions

MCQ-4A catalog entry for `fct_orders` carries a Certified badge. Its record reads: certified_by=alice@, certified_at=2024-03-11, criteria=[ownership, semantics, quality, security, freshness], freshness SLA=6h, last_verified=2024-03-11. Today is 2025-05-02 and the load monitor shows the table last refreshed 31 hours ago. A finance analyst wants to cite the badge as assurance for a revenue report. What does the badge assert at this moment?
MCQ-13The `orders_daily` dataset carries a certification contract with three clauses checked on every run: freshness (data lands by 06:00 UTC), volume (row count within ±5% of the trailing 7-run median), and key integrity (`order_id` null rate below 0.01%). The trailing 7-run median is 1,000,000 rows. This morning's run lands at 05:40 UTC with 1,090,000 rows, an `order_id` null rate of 0.00%, and a `customer_id` null rate of 4.1%. `customer_id` is not named in the contract, but three ML feature tables inner-join on it. Which certification outcome is correct for this run?
MCQ-15A data catalog lists 240 published tables. Ownership is recorded as a single field. Of the 240: 96 name an engineer still on the publishing team; 58 name an engineer who transferred to another team 8 months ago but still approves schema-change requests for those tables; 41 name an engineer who left the company; 27 hold a team alias such as #data-eng with no named individual; 18 are empty. How many of the 240 tables have no accountable individual who can approve a schema change today?
MCQ-2A 300-person analytics org runs a self-serve SQL exploration program on its warehouse. Domains: growth (events, subscriptions), support (tickets, customers), finance (transactions, invoices). The platform team approves exploration access only when four gates all pass: the requester's team owns the domain of every table in the request; the requester passed the SQL assessment within the last 12 months; the grant expires within 90 days; and the requester holds a current data-handling attestation. Five requests are pending. Which one should the platform team approve?
MCQ-9An active-users metric counts DISTINCT user_id per day from the events table. Monday reads 120,000; Tuesday reads 78,000, a 35% drop. Tuesday's partition now holds 3.5M rows against Monday's 3.6M, but 1.1M of Tuesday's rows arrived in a backfill at 08:00 Wednesday. Tuesday's ingestion job ran at 23:50 and loaded only events with ingested_at before 23:00. The metric job runs at 01:00 daily and was not re-run after the backfill. What does Tuesday's 78,000 most likely represent?
MCQ-7A 'Weekly Active Users' number is quoted in the product review from an analyst's personal workspace. The certified product metric counts a session of ≥ 10 s as active and reported 250,000 active sessions last week. The session log for that week shows 20,500 sessions lasting between 10 s and 30 s. The personal report reported 229,500. The personal report's gateway credentials are the analyst's personal service account; a refresh failure two weeks ago was fixed by re-entering the analyst's password, and the dataset otherwise refreshes daily. Both metrics use the same Monday 00:00 UTC week boundary. The personal report reads 8.2% below the certified metric, and has done so for six months. Which cause accounts for the gap?
MCQ-16A self-service template for a revenue dashboard runs `SELECT SUM(amount) AS revenue FROM analytics.orders WHERE status = 'completed'` and exposes a `region` filter that appends `AND region = :region`. An analyst adds a custom filter `status = 'refunded'` and the dashboard shows revenue = 0. When the analyst changes the custom filter to `status = 'completed'`, the dashboard shows the correct completed revenue. What is the most likely cause?
MCQ-17A BI tool enforces row-level security in its query layer: before sending SQL to the warehouse it appends `AND region = :user_region`. The warehouse has no native RLS, and the BI service connects with a single shared read-only account. The new "Download CSV" worker takes the saved query's SQL, runs it on that shared connection, and streams the rows to the browser. A week after launch, an EU analyst's download contains APAC rows. Which change fixes the leak?
MCQ-12A retail analytics team runs one semantic layer over a shared warehouse. The base tables belong to a platform team that will not attach per-consumer policies to them. Business users in four regions have explore rights, which include writing custom SQL against the semantic layer's SQL endpoint as well as opening dashboards. Compliance requires that an EU user never sees APAC transaction rows and that customer email and phone reach only the fraud team. Users transfer between regions, and the IdP already carries a region claim. Which configuration satisfies all of these?
MCQ-18A 400-person analytics org runs a quarterly access review in its self-service BI tool. Reviewers work from a flat export of 1,200 direct role grants; another 900 grants reach users only through 14 nested groups and never appear in the export. The 14 groups are owned by the platform team, which does not track who transfers between teams. Last quarter reviewers approved 97% of lines at a median 3 seconds each, and the audit afterward found 11 users still able to query the finance dataset after moving to sales: all 11 held that access through a nested group, and the same gap has produced findings in each of the previous three quarters. Which change to the review process addresses the finding?
MCQ-5A 25-person analytics team needs a sandbox to explore a 10 TB event table that contains user emails and IP addresses. The security team requires that raw PII never leaves the governed zone. Finance caps sandbox spend at $2,000/month and requires that no single query scans more than 2 TB. A full clone with masking takes 6 hours and costs $500/day; a zero-copy clone with masking takes 2 minutes and costs $0.10/day. Certified revenue models must remain read-only to sandbox users. Which sandbox design should you implement?
MCQ-20A retail team wires its natural-language analytics assistant to the warehouse with a service account that has SELECT on all 60 schemas, and lets it write SQL against base tables. In the first month: 'Q3 revenue' comes back 12% above the finance dashboard, because the generated SQL sums `orders.amount` without the refund and test-account filters the certified revenue metric applies; a regional manager's 'top accounts' answer includes accounts from a region they are not entitled to; and 9 of 40 sampled answers reference a column that does not exist. The assistant has to keep answering in under 3 seconds at 400 questions a day. Which change removes the cause of all three failures?
MCQ-6A platform team adds one blocking gate to its promotion pipeline: an asset cannot leave the sandbox until every column is registered in the governed semantic layer with a stable definition. The gate runs as a step in the promotion job. Four months later, a producer renames `order_total` to `order_amount` in the `orders` table. The asset was promoted 14 months ago and is registered in the semantic layer. The next morning, 12 dashboards that query the semantic layer fail. Which explanation accounts for the failure?
MCQ-25An upstream migration forces a certified orders dataset to rename `customer_id` to `account_id` and change its type from string to integer. Forty published dashboards and twelve scheduled extracts read the dataset, and their owners are not all known. What rollout keeps that user-built content working while the certified model evolves?
MCQ-14Two certified metrics are both published as `orders` in the semantic layer, and both contracts say the metric counts orders excluding cancellations. For 2024-06-01 the finance metric returns 18,402 and the product metric returns 21,977. The `orders` table has 3,575 rows that day with `status = 'cancelled'`. The product metric feeds 40 dashboards, the finance metric 6. What should the analytics engineering team do?
MCQ-1An analytics team of six supports 400 analysts in finance and marketing. After the warehouse was opened for self-service, analysts published 1,200 dashboards in one quarter. In the last exec review, finance reported 2.1M monthly active customers and marketing reported 1.7M, both from dashboards labeled "active customer"; a postmortem found five different SQL definitions of that metric. The central team's backlog for new data requests is six weeks. Which action best resolves the reporting conflict without giving up self-service?
MCQ-3An e-commerce company's warehouse holds 1,400 raw tables in `raw_ops` and 90 certified models in its semantic layer. Analysts query `raw_ops` directly because the semantic layer has no drill-through. Last quarter, gross margin appeared in 7 executive dashboards with values spread across a 12-point range, and finance spent 4 days each month reconciling which number was right. Leadership wants one gross margin number without losing the ability to inspect individual orders. Which change satisfies both requirements?
MCQ-24A 600-seat self-service BI rollout reports 78% monthly active users and 2,400 dashboards created this quarter, and the team calls it a success. An audit finds only 310 of those dashboards were ever opened by someone other than their author, the semantic layer holds four competing definitions of 'net revenue', and 11 decisions this quarter were traced to a stale or wrong number. What does this pattern indicate?
MCQ-8A 600-analyst org ran a mandatory data-literacy course: 97% completed it, 88% passed the end-of-course quiz, and satisfaction averaged 4.6/5. Six weeks later, a review of 50 dashboards found 31 that used a raw count where a rate was intended and 4 that shared row-level customer email with an external vendor. The governance team has 40 reviewer-hours per quarter and one dashboard review takes 45 minutes. The VP wants a quarterly number that would move if the course changed how people build dashboards. Which measurement should the team run?
MCQ-10A data platform team runs two 60-minute office hours a week and has named champions in 3 of its 11 product teams. Last quarter the sessions logged 480 questions: 310 came from the 8 teams without a champion, and 240 were one of five recurring questions about metric definitions and dataset freshness. The team can take on one new initiative this quarter. Which initiative should it take on?
MCQ-19Your team owns an analytics catalog. A "Checkout Funnel" dashboard last refreshed 14 months ago: the pipeline that fed it was deleted in a warehouse migration, so its tiles are frozen at the old numbers. It still draws 900 views a month, the growth team opens it in its Monday review, and its 3.1% conversion figure has appeared in the last two board decks. The live checkout conversion rate is 2.4%. What should you do with the dashboard?
MCQ-21A 400-engineer analytics org chartered its Center of Excellence to set standards, review designs, and coach embedded champions, not to build. Eight months in, the CoE's 12 engineers spend about 70% of their time implementing dashboards and analyses that domain teams request; the queue is 300 deep and growing by roughly 40 a month, and three domains have started building ungoverned pipelines to avoid the wait. The steering committee asks which single CoE practice to stop. Which one?
MCQ-22A 40-person analytics group has a central governance board that reviews every new dashboard before release. After two quarters: median review time is 11 days, 38% of new dashboards ship as ungoverned personal extracts, and two incidents traced to stale vetted datasets. The board meets weekly and has no queue metrics. Leadership asks what to change first. Which action best evolves governance for this situation?
MCQ-11A shared Trino cluster has 20 workers, each running 4 concurrent query tasks, so the cluster admits 80 tasks at once. Production dashboards need 30 concurrent tasks at peak to hold their p95 under 2 s. Analysts run ad-hoc exploration on the same cluster, and one query can currently take all 80 tasks. Which resource-group configuration keeps dashboards inside their p95 budget while leaving the most capacity for exploration?
MCQ-23A self-service BI portal renders its 12-month revenue trend as a single canvas element. A screen-reader user tabs to the chart region and hears only 'graphic'; the 12 monthly values are unreachable. The team cannot replace the charting library this quarter. Which change makes those 12 values available to that user?