Skip to content
Tech Interview Prep
Technical interview guide

Data Storytelling

Presenting analysis so the insight and recommended action are unmistakable, not just the numbers.

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

Scope: WCAG 2.2 and current W3C WAI, USWDS, UK Government Analysis Function and Statistics Authority, CDC, Tableau, Microsoft Power BI, and NIST Privacy Framework guidance reviewed 2026-09-04.

Curated: · Written: · Reviewed:

25 questions

MCQ-2Your incident tracker imports the first 500 characters of each incident review as the follow-up ticket description; nothing after that is imported. The review for a 40-minute checkout outage is laid out as: summary (characters 0–620), timeline (621–2,400), root-cause analysis (2,401–3,100), action items (3,101–3,400). The summary's opening sentence (characters 0–180) states the 40-minute outage and that the connection pool maxed at 20; the fix, raise the pool limit to 80 and add a queue-depth alert, is written in the action items section. What will the follow-up ticket description contain?
MCQ-1You have 45 minutes with the VP of Product, who must decide by Friday whether to ship a feature that triples write volume. Your analysis: current p99 write latency is 40 ms; a load test at 3× volume showed p99 rising to 180 ms once the connection pool saturated at 20 connections; raising the pool to 60 held p99 at 55 ms at 3× volume and costs one day of work. You also have 40 exploratory charts and two unconfirmed hypotheses. Which opening claim and evidence should the story lead with?
MCQ-18Your design doc for a sharding migration goes to a 45-minute review where the platform and payments leads will approve or reject it. The main path carries the decision and the numbers that drove it. Which of the following belongs in the appendix?
MCQ-13A growth team is preparing a weekly review slide. The chart shows daily active users for a web app over 10 days: 8,000; 8,100; 8,050; 8,000; 12,400; 12,600; 12,500; 8,050; 8,100; 8,200. A marketing email went out on day 5 and its traffic ended on day 7. Which title should go on the chart?
MCQ-10An exec has 30 seconds per chart in a review. The chart plots p99 latency for four services over 12 weeks. It carries a dashed SLO line, a shaded band over the week-9 deploy, a footnote that the metric definition changed in week 7, heavy gridlines, and a legend, though each line is already labelled at its right end. The exec must say which service is drifting toward the SLO. What should be cut first?
MCQ-14A monitoring dashboard renders a 24-hour error-rate chart for five services. A screen-reader user is on call and must decide within seconds whether to page the payments team. The chart shows payments at 12% error rate at 14:05 while the other four services stay below 0.5%. The current alt text is 'Error rate chart.' Which replacement lets the on-call engineer make that paging decision without sighted help?
MCQ-8A quarterly review covers 12 divisions. Between Q1 and Q2 the four largest divisions each fell 8–12% and the four smallest each rose 15–20%, but no division overtook another. The analyst wants the audience to see that the spread between the top and bottom groups narrowed. Which condition makes a slopegraph the better encoding than a grouped bar chart?
MCQ-21A payments team has to answer one question at Thursday's ops review: rank the 14 sales regions by chargeback rate so the bottom three can be put on a remediation plan. They have a choropleth of chargeback rate by region. The regions differ 10× in area (West 1.8M km², Northeast 181k km²) and 3× in rep count (West 210, Northeast 640). In a dry run, 9 of 11 attendees read the map as 'the West is the worst region'; the West actually ranks 8th of 14 on chargeback rate. What should the team put in front of the review?
MCQ-22A growth team spent two weeks generating dozens of exploratory views (scatterplots, histograms, small multiples) and traced a 12% Q3 activation drop to a checkout error on mobile Safari. The VP of Product has a 10-minute slot, has not seen any of the analysis, and must decide whether to hold the Q4 mobile release. Which set of slides correctly applies exploratory and explanatory visualization?
MCQ-15A capacity review cites a chart: 'p99 latency fell from 180 ms to 40 ms after the connection-pool change.' The citation reads 'Source: Grafana latency dashboard, March.' A reviewer opens that dashboard and sets the range to March. Its default all-services panel shows p99 = 40 ms for the current week and 95 ms for the week before the change; the checkout-api panel shows 180 ms for the week before the change. What drop can the reviewer reproduce from the citation as written?
MCQ-9A daily-active-users dashboard shows DAU flat at 100k through March 3, then a step to 118k on March 4 that holds flat for the rest of the month. The March 4 pipeline release notes say logged-out sessions, measured at 18% of sessions in the prior week's audit, are now counted in DAU. A marketing campaign launched March 1 with spend ramping over two weeks. The team wants one annotation on the March 4 step. Which annotation should they place?
MCQ-3A payments service's dashboard shows failed charges rising from 4,000 to 12,000 per day after a retry-logic change, while total charges rose from 200 million to 400 million per day. The headline reads 'Failed charges up 200%'. Which statement correctly describes the change?
MCQ-5A subscription business reports revenue in $M by quarter. 2024: Q1 20.0, Q2 22.0, Q3 21.0, Q4 28.0. 2025: Q1 24.0, Q2 26.4, Q3 25.2, Q4 33.6. The CFO wants one growth figure on the board slide, one a shareholder can hold the company to over the next four quarters. Which figure goes on the slide?
MCQ-6A team evaluates a new onboarding checklist with two studies run in the same quarter. Study A randomizes 20,000 new users: the checklist group's 7-day retention is 4.1 points higher than control (95% CI 3.2 to 5.0). Study B compares 20,000 users who adopted the checklist on their own with 20,000 who did not: adopters retain at 61% versus 42% at day 7. Which statement is a causal overclaim?
MCQ-4A platform team is deciding whether to replace Postgres 15 with a new store. Both were benchmarked with fsync on, the same dataset and client: the new store sustained 12,000 writes/s on a 32-core machine, Postgres 4,000 writes/s on an 8-core machine. The production workload is 5,000 writes/s on 8-core machines. Which conclusion does the benchmark support?
MCQ-11A payments team is deciding whether to fund a refund-flow rewrite. Over 90 days, telemetry shows 240,000 refund attempts, 1,800 failures (0.75%), and 62% of those failures were bank-side timeouts. A support engineer forwards one customer's 40-minute screen recording of a failed refund that ended in a manual bank transfer. What should the team do with the recording in the proposal?
MCQ-12A checkout redesign ran for three weeks on 41,200 users: conversion was 6.4% in control and 6.9% in treatment, a +0.5 pp lift with a 95% CI of [+0.1, +0.9]. The team also sliced by platform. On Android 9 (190 users per arm, 22 conversions in total) the estimated lift is +3.2 pp with a 95% CI of [-1.5, +7.9]. The PM wants to ship to Android 9 first because that slice shows the biggest lift. What should the team do?
MCQ-16A pricing team pre-registered revenue per user as the primary metric. The experiment returns +2.1% overall, 95% CI [+0.4%, +3.8%], p=0.01. The analyst then runs 12 pre-specified subgroup cuts, roughly independent of each other. One cut, EU Android, returns −3.4%, 95% CI [−6.5%, −0.3%], p=0.03. The launch decision is contractually fixed for two days from now. What belongs in the launch memo?
MCQ-7A capacity team must present next quarter's request-rate forecast to the VP who approves server spend. Model v3, trained through June, predicts 12,000 req/s. Its 80% backtest interval is 9,000–16,000 req/s, and in one of the last four quarters actual demand hit 17,500 req/s. The current deck shows a single 12,000 line and a 12,500 capacity plan. What should the team put in the deck?
MCQ-24A fraud model's holdout evaluation reports AUC 0.81 with a 95% confidence interval of 0.74 to 0.86. The decision-maker who must approve the pilot has no statistics background and asked for a one-slide readout. Which statement should the readout make?
MCQ-20A 12-store pilot measured a +1.8% basket-size lift with a 95% CI of [-0.4%, +4.0%]. The executive committee meets once, on 14 March, and locks the FY allocation that day; the next measurement read lands in August. Rollout is phased, and halting after any phase costs about 3% of that phase's committed spend. Full rollout would consume the entire discretionary budget, and its lower bound would put the division's margin at 17.6%, below the CFO's 18% floor. The pilot is the only initiative with the required capability. How should the presenter communicate the interval to the committee?
MCQ-23A data team's monthly spend report, a five-section data story, goes to 50 engineering directors. Telemetry: 49 of 50 opened it; 46 of 50 expanded every section; 43 of 50 correctly identified the largest cost driver on the end-of-report check. Only 4 of 50 could state the recommended reserved-instance purchase unprompted, and 3 of 50 approved that purchase within two weeks: all three were among the four who could state the recommendation. The team rewrites the report so the recommendation is as memorable as the cost driver, and expects 43 of 50 directors to be able to state it unprompted. Holding the approval rate among directors who can state the recommendation constant, how many of the 50 are expected to approve?
MCQ-17A staff engineer's recommendation to move the payments service from a 20-connection pool to a 60-connection pool was approved with this operational block: - owner: Dana - expected effect: p99 checkout latency falls from 180 ms to below 100 ms - measure: p99 checkout latency, 1-minute buckets - risk: database CPU saturation - trigger: revert to 20 connections if database CPU exceeds 80% for three consecutive 1-minute buckets After the change, database CPU reads 84%, 86%, 83%, 79%, 88% over five consecutive minutes, and p99 checkout latency is 65 ms. What does the recommendation require the team to do?
MCQ-19A payments team publishes a weekly 'settled transactions' series through a versioned public API. Consumers pin a version and learn about changes only from the changelog. A join bug dropped card-present transactions, undercounting the series by 11% for the 7 weeks from 2024-03-04 to 2024-04-15. A finance team used the 2024-03-04 figure in an external investor update, and an ML team trained a fraud model on the full series. The pipeline fix is ready. What should the team do?
MCQ-25A churn deck goes to the exec team. The reviewer traced every figure to the source table, recomputed the aggregates, and checked the chart labels. The deck's headline reads "Churn fell 3 points." The exec team cancels a $2M retention program. The 3-point drop was measured on the self-serve cohort; enterprise churn, 60% of revenue, rose 2 points over the same period. Which review step would have caught the failure before the deck shipped?