Skip to content
Tech Interview Prep
Technical interview guide

Stakeholder Communication & Alignment

Keeping engineering, leadership, and customers aligned on priorities and tradeoffs without every conversation becoming a negotiation.

Read
31 min
Practice MCQs
25
Interview Q&A
25
Edition
v3
Editorial status
Reviewed

Scope: GOV.UK Service Manual and Consultation Principles, IETF RFC 7282, U.S. federal plain-language guidance, and NIST RMF current 2026-08-31.

Curated: · Written: · Reviewed:

25 questions

MCQ-10A payments team must freeze the ledger-table partition key before Friday's ship. The call needs three people: Ana (UTC-8), Bruno (UTC-3) and Priya (UTC-1). Ana and Bruno work 09:00–17:00 local; Priya works 09:00–18:00 local. Ana has a standing on-call handover every day from 10:00 to 11:00 local. The call needs 60 minutes. Which UTC window is the only one that works for all three?
MCQ-9A search team keeps one ADR per decision, and the template marks every record with a status of proposed, accepted, rejected, or superseded. The folder holds four records for the search backend: - ADR-009 (2023-03): use Elasticsearch. Status: accepted. - ADR-022 (2024-01): use OpenSearch. Status: accepted. Supersedes ADR-009. - ADR-035 (2024-11): use Elasticsearch. Status: rejected. - ADR-041 (2025-04): use Typesense. Status: proposed. A new engineer is about to add a query-parsing feature that uses engine-specific syntax, and asks which engine the service runs today.
MCQ-13A staff engineer keeps two live roadmap decks. The engineering version ranks the payments migration as the top Q3 priority; the version shared with the sales org marks it "deferred, no date." Both are current, each audience is satisfied, and nobody has compared them. Which of these is the communication anti-pattern the situation illustrates?
MCQ-24A staff engineer DMs the tech lead: the proposed single sharding key will hot-partition at 3× peak, so use a composite key. The tech lead agrees and switches the plan. The design doc is the only artifact the on-call rotation and next quarter's team read; DMs are not searchable, and the single key is the obvious first choice, so it will be proposed again. Which doc edit closes the loop for a reader who arrives in six months with no access to the DM?
MCQ-19A payments team closed its 12 March decision meeting with the line "Decided: idempotency keys in Redis, 24-hour TTL. Platform team." In June the mobile client's offline retry queue was extended from 24 to 72 hours. In September a 26-hour Redis outage triggered a retry storm that produced roughly 4,000 duplicate charges, and the postmortem found the March decision had never been revisited. Which element of the March close would have surfaced the problem before September?
MCQ-1You run a 10-minute written check before a migration window. Each of the 12 people in the review writes one line: the decision, their team's first action, and what would make them stop. The tally: 12 of 12 wrote the same decision; 12 of 12 wrote a first action; 4 of the 12 first actions contradict another team's; 12 of 12 wrote a stop condition, and all 12 are different. You have 30 minutes left in the meeting. What should they be spent on?
MCQ-2You are the staff engineer on a migration of the internal auth service. The stakeholder map was built from the org chart: it names the platform VP, the two service team leads, and the security director. Three weeks in, the cutover slips twice: the payments team's on-call lead refuses the date because their client library pins the old token format, and legal only hears about the retention change after the second slip. Which change to the map would have surfaced both blockers before the first slip?
MCQ-20A payments team is rewriting the refunds flow. The kickoff workshop drew 11 attendees, all from the merchant-onboarding squad; the invite went only to the #payments Slack channel (12 members). Support, 3 agents handling about 200 refund tickets a week, was never invited. The workshop produced 38 requirements: 31 concern refund configuration during onboarding, and none mention partial captures. Support's ticket tags show partial-capture failures are 60% of refund complaints. Which reading of the output is best supported?
MCQ-15Nine engineers review a schema change: six in the room, three on video. The facilitator closes with 'Any objections?', waits four seconds, hears nothing, and logs 'approved, no objections'. The meeting ended twelve minutes early, all nine had read and commented on the design doc 48 hours before, and the two engineers who spoke most were the two most junior in the room. Two days later five attendees say they had doubts about the backfill window: three assumed the room's silence meant the other teams had signed off, one expected to file comments after the meeting, and one, the staff engineer who owns the service, had not realised the window was still open. At the next review, what should the facilitator do differently?
MCQ-5An IETF working group of 40 people meets to adopt a draft that adds a length field to a signalling header. The chair hums: 37 in favour, 3 against. One of the three objectors then says the field collides with the encoding used by a middlebox series already deployed in firmware, and those devices cannot be upgraded in the field. Nobody responds to the point, and the chair has nothing else to go on. What should the chair do?
MCQ-23A 12-engineer platform team added a second daily standup and an #eng-announce channel at the start of the quarter. Over that quarter: Slack messages per engineer per day rose from 40 to 95; meeting hours per engineer per week rose from 3 to 6; median RFC time-to-decision rose from 6 days to 11 days; 3 of the 8 RFCs decided that quarter were reopened within two weeks; and in the postmortem of a 40-minute payments outage, the on-call engineer reported not knowing that the payments team had posted a mitigation in #eng-announce 12 minutes before the outage ended. Which reading of the quarter is correct?
MCQ-11A staff engineer's 14-page dual-write migration proposal stalled twice: the review board spent most of a 60-minute slot decoding terms like 'idempotent backfill' and deferred both times. The rewrite is 4 pages in plain language, one decision per section, and it is approved in 20 minutes, but the approval attaches a condition that reverses the chosen cutover order. The finance reviewer read 'about $40k/month' as a steady-state cost; the original had said $38,400/month at 1.2M events/day, rising to $61k at 2M. Which change to the rewrite would have prevented the wrong approval?
MCQ-22You lead a payments migration and send a Friday status update to a six-person steering committee. Last Friday's update ran 900 words, listed all 34 tickets closed that week, and ended "continuing to make good progress." Within an hour, three members emailed asking the same question: does the March 14 cutover still hold now that the acquirer's sandbox certification has slipped? The committee makes the go/no-go call on Monday and reads updates on their phones. What should the next update do?
MCQ-25You own a 200-item service migration. The weekly stakeholder update reports percent complete, and it has read 80% for the last three weeks. Forty items are still open, the team has closed about 6 items a week for the last month, the best week so far closed 13, and the migration's 20-week average is 8 a week. Product has booked the launch for next week. What should the next update report?
MCQ-3A staff engineer is preparing the quarterly reliability review after a checkout incident. The canonical incident record says: a connection pool max was lowered from 50 to 20, p99 latency rose from 180 ms to 1.2 s for 47 minutes, 3.1% of checkout attempts failed, the pool size was reverted, and a saturation alert at 80% pool use is owned by the payments team with a fix date of May 15. The engineer must brief executives, customers, and the payments engineering team. Which briefing plan best preserves one source of truth while fitting each audience's decisions?
MCQ-16A staff engineer reviews this risk register entry the day before a launch review: | field | value | | --- | --- | | Risk | Recommendation service adds 120 ms to checkout p99 | | Likelihood | 4/5 | | Impact | 5/5 | | Mitigation | Add a 500 ms timeout and a fallback to cached recommendations | | Owner | — | | Residual risk | — | | Decision | — | The launch board's rule: any entry with impact 5 must name an owner, a mitigation that changes likelihood or impact, and a residual rating, or the board returns it. What should the reviewer do?
MCQ-14Your team of five has a two-week sprint with 34 points committed: 23 points are a compliance fix due at sprint end, 8 points are an audit-log export with no external date, and 3 points are carryover. A sales VP asks you to add SSO domain restriction for a $1.2M renewal; the customer signs in 10 days or walks. Your estimate is 8 points. Historical velocity is 34 points. What should you do?
MCQ-6You lead a five-engineer payments team. The quarter is committed to a PCI-DSS remediation the card network requires by Friday 14 November; missing it suspends card processing. A sales director asks for a custom CSV export for a prospect worth $180k/year, needed 'in two weeks.' The export is about three engineer-weeks of work and touches cardholder data, so it has to ship inside the audited environment. Which reply should you send?
MCQ-18A platform team has 4 engineers for a 6-week quarter (24 engineer-weeks). Ops and on-call consume 8, leaving 16. The published roadmap says: Now, audit log export (6 engineer-weeks); Next, SSO (8); Not now, custom roles (10); Won't do, per-tenant schema changes. Two weeks in, a customer offers to pay for custom roles if the team drops audit log export. What does the roadmap's explicit non-commitment list let the team conclude about that offer?
MCQ-21A contract clause bars collecting user data until compliance signs off on the study design. Compliance runs one first-in-first-out queue, currently three weeks deep, and takes no priority requests. The funder's milestone needs external participants inside this two-week sprint, and the VP of Product owns both the compliance gate and the milestone commitment. What should the team do?
MCQ-4A platform team is retiring a shared internal API used by 40 service teams. The architecture board has fixed the replacement and the cut-off date, but has not decided whether to ship a compatibility shim for the two heaviest consumers. The team has three weeks before the deprecation notice goes out and wants the consultation to be more than a formality. Which plan makes it so?
MCQ-7At 09:12 your payments API's error rate jumps from a 0.2% baseline to 6% of authorization requests. The deploy 8 minutes earlier is the prime suspect but unconfirmed. Rollback plus verification is estimated at 25 minutes; root-cause analysis at 45 minutes or more. Enterprise contracts require the status page to carry notice of any incident above a 1% error rate within 15 minutes of detection, including the measured impact and the time of the next update. What should you do at 09:15?
MCQ-12Your team owns `ledger` and exactly these copies of its data: the append-only `ledger_entries` table in Postgres (about 480M rows) with one synchronous streaming replica; a nightly logical dump into Finance's `s3://fin-recon/ledger/` bucket, one object per run keyed by date, where the lifecycle rule expires objects 400 days after creation; and a Kafka topic `ledger.entries.v3` with retention.ms = 604800000 (7 days), which Risk's nightly fraud job consumes to write feature rows into a store with a 3-year TTL that only ever adds rows. There is no PITR archive and no warehouse export. A GDPR erasure request for user 8817 arrives. Your whole plan is one statement, run tonight at 01:00: DELETE FROM ledger_entries WHERE user_id = 8817. Assume autovacuum has already cleared the dead tuples. Thirty days later, how many stores still hold a row that identifies user 8817, and which are they?
MCQ-17You are the tech lead moving the orders service from a single Postgres 14 primary to 8 shards keyed by customer_id. The 40-million-row customer table is replicated in full on every shard, and order lookups are always per-customer. In design review, a risk engineer objects: 'Our nightly risk scan joins orders to the customer table across all customers. Sharded by customer_id that becomes 8 queries, and each one still scans the full replicated customer table, so each runs 3.2 s against the 5 s statement timeout.' The scan runs 40 times a night at 02:00, when write traffic is at its lowest. What do you do?
MCQ-8A team of five engineers must migrate 240 services off a runtime that is decommissioned in 16 weeks. Their last 12 weeks of throughput average 20 services/week with a standard deviation of 5. One engineer is on leave for four weeks mid-migration and no backfill is approved. The director asks the team to commit to 12 weeks: the mean of their throughput forecast. Which response can the team defend?