← All posts

How to prepare for Airbnb's system design interview

Airbnb's careers FAQ describes the company's entire interview process in a few sentences and never mentions system design. Airbnb's engineering blog, by contrast, published a long account in April 2026 built around a hypothetical server crashing halfway through a multi-step process, after a host's insurance claim has been validated and before the payout goes out. If you're preparing for the Airbnb system design interview, the second source is the one worth your evening.

That gap is the useful part. Airbnb's own engineers have published what they actually worry about, in detail, with the trade-offs named. A candidate who has read it walks in knowing which failure cases matter at this company.

What does Airbnb publish about its system design round?

Very little. The careers FAQ says the process "may involve taking a test, making a video, or joining us for an event," depending on the position, and that Airbnb makes it "a priority for candidates to meet with multiple individuals on the current team." It sets one expectation on timing: "We aim to respond to everyone within 3-4 weeks." It also explains why rejection usually comes without notes. Airbnb says it reviews about 15,000 applications each month, so personalized feedback is unlikely.

Everything more specific comes from outside the company. interviewing.io's Airbnb guide, which says it spoke to current and former Airbnb interviewers and cross-referenced their accounts with Glassdoor data and its own mock interviews, describes an onsite with coding, system design, behavioral and culture-fit rounds, with more senior engineers getting more design and less coding. That is a third party's reconstruction, and Airbnb has not confirmed it. Our Airbnb interview process guide traces where most of the other circulating claims about Airbnb's loop came from, and the short version is that few of them trace back to Airbnb.

Why is Airbnb's engineering blog the best prep source?

Because a system design interview is a compressed version of the job, and the engineering blog is the job written down. Two 2026 posts are especially useful. "Skipper: Building Airbnb's embedded workflow engine" explains how Airbnb keeps multi-step processes correct through crashes. "Pay as a Local" explains how its payments platform handles payment methods that complete outside Airbnb's own app.

Both posts show what Airbnb's own engineers describe as a hard problem and how they argue about solutions. The Skipper post says the engine powers 15+ use cases across insurance, payments, media, infrastructure, incentives and wallet teams. That's a reasonable signal about how much of Airbnb's backend revolves around long-running processes that must not half-finish.

Read each post once for the problem, then close it and try to design the system yourself before reading how Airbnb did it. The comparison is the practice.

What happens when an Airbnb workflow crashes halfway through?

This is the scenario Airbnb's Skipper engineering post opens with, and it's the strongest practice prompt Airbnb has published. The post describes a host's insurance claim that has to be validated, checked, assessed and paid out, then asks what happens when the server crashes "after the validation passes, but before the payout." Its answer for a traditional architecture is "often 'it depends.'"

Airbnb's design choices give you the vocabulary a strong answer needs. Skipper checkpoints the result of each step, called an action, to the database the service already uses. After a crash, it replays the workflow from the top, and completed actions return their saved results instead of running again. The post is candid about the limit. Actions "may execute more than once in edge cases (crash after execution but before checkpoint)," so "Actions should be idempotent." When a later step fails permanently, paired compensation methods run "in reverse order (releasing held inventory, refunding charges, reverting state changes)."

The post's own code sample shows the same pattern on a booking, a charge followed by marking the reservation accepted. Practice it as "design the flow that charges a guest and confirms a reservation." Then answer three follow-ups out loud. What happens if the process dies after the charge and before the confirmation? How do you stop a retry from charging twice? What do you undo, and in what order, if the confirmation can never succeed?

How does Airbnb handle payments that finish outside its app?

Airbnb's "Pay as a Local" post describes a payments problem specific to a global marketplace. Airbnb operates in more than 220 global markets, and its research identified over 300 unique payment options worldwide. Many of the methods it added require the guest to finish paying in a third-party wallet app or by scanning a QR code.

The post sorts them into three flows. In a redirect flow the guest leaves for a third-party site and returns with a result token. In an async flow the guest pays externally and Airbnb learns about it later "via webhooks." In a direct flow the guest enters credentials inside Airbnb, much like a card payment. Airbnb also built a config-driven layer, a central YAML file it calls the single source of truth for each method's flows, eligibility and refund rules.

The async flow is a compact design prompt. A booking is waiting on a payment Airbnb cannot see happen. Decide what state the reservation sits in while it waits, what happens if the webhook arrives twice, and what happens if it never arrives at all. Those three questions cover idempotency, state machines and timeouts in one short answer.

How should you talk about monitoring in an Airbnb design answer?

Build it into the design from the first diagram. Airbnb's "Pay as a Local" post makes a specific point that's easy to borrow. A single "payment success rate" may describe card health well "but say little about an asynchronous" payment method, so Airbnb built metrics across four layers: client, payment backend, payment provider and webhooks.

It also published an example alert rule: "Naver Pay resume failures > 5 and failure rate > 20% in 30 minutes." Pairing a count threshold with a rate threshold is there, per the post, to minimize false positives during low-traffic periods. That's a concrete, defensible detail you can bring into any design answer. A rate alone fires on two failures out of three attempts at 4 a.m., and a count alone misses a slow leak on a busy day.

When you describe your design, name the one metric that would tell you it is broken and say where you measure it. "We'd add monitoring" gives an interviewer nothing to push on. "Alert when both the failure count and the failure rate cross a threshold in a 30-minute window, per payment method" gives them a decision to probe, which is the conversation you want.

What trade-offs does Airbnb say out loud?

Airbnb's Skipper engineering post includes a section titled "Key tradeoffs" that lists what Airbnb gave up, and it's the best model in this post for how to close a design answer. Airbnb chose an engine embedded in each service over a central orchestration cluster because its "Tier 0" services, the ones that directly affect user-facing transactions, could not take on a new critical dependency. In exchange it accepted three costs: workflow code must be deterministic, actions run at least once, and changing a workflow's structure "can break in-flight workflows."

The same post gives scale. At peak, it says, Skipper has scaled to 10,000 workflows per second on Amazon DynamoDB. Airbnb doesn't present the embedded model as the right answer everywhere. It says teams needing cross-language support or cross-service orchestration "may find a dedicated orchestration system more appropriate." That sequence gives you a template for any answer. Make a choice, give the reason, name what it costs, and say when you'd choose differently. Our guide to system design interview questions covers how panels score that kind of reasoning across companies.

What is overrated

Memorizing a "design Airbnb" walkthrough is overrated. A rehearsed search-and-booking answer only helps if you get that prompt, and the one interviewer-derived list cited here is not Airbnb-shaped. interviewing.io's Airbnb guide lists designing a cache, a voting system, a load balancer and an auth platform among its examples.

You may get a generic prompt, so the Airbnb-specific value comes from how you handle partial failure, duplicate events, compensation and the trade-off you chose on any prompt. Over-indexing on one prep site's question list, including that one, is a bet you don't need to make.

A five-step playbook for Airbnb's system design round

  1. Read the two Airbnb engineering posts before any prep-site guide. Skipper for multi-step correctness, "Pay as a Local" for payments that complete elsewhere.
  2. Design the charge-and-confirm flow cold, then compare. Write your version first and check it against Airbnb's checkpoint, replay and compensation model afterward.
  3. Answer every prompt's three failure questions. What if it crashes midway, what if an event arrives twice, and what if it never arrives?
  4. Close with a named trade-off. Say what you gave up and when you would choose the other option, the way the Skipper post does.
  5. Rehearse the follow-up out loud. Four-Leaf's AI mock interviews include a system design round you can answer by voice, and in question-by-question practice the feedback checks whether you called out trade-offs explicitly and penalizes vague answers like "use a cache" without specifics. For latency follow-ups in particular, see how to discuss latency in a system design interview.

Where this is heading

Airbnb keeps its interview process close and its engineering problems in public. Its own posts, published in February, April and November 2026, describe systems where the hard part is staying correct when a crash, a retry or a missing webhook gets in the way.

Candidates who prepare from that material will sound like they've already sat in one of those design reviews.

Frequently asked questions

Does Airbnb ask system design questions?+

Airbnb's careers FAQ does not name system design or any other stage. interviewing.io's Airbnb guide, based on conversations with current and former Airbnb interviewers plus Glassdoor and mock-interview data, describes a system design round in the software engineering onsite alongside coding, behavioral and culture-fit interviews. Treat the round as likely for mid-level and senior engineering roles, but plan around what you can verify rather than a fixed round count.

What system design questions does Airbnb ask?+

Airbnb does not publish its prompts. interviewing.io's Airbnb guide lists generic examples such as designing a cache, a voting system, a load balancer and an auth platform. Airbnb's own engineering blog is a better guide to what the company cares about, because it describes how Airbnb designs payments and multi-step workflows to survive failure.

How long does Airbnb take to respond to an application?+

Airbnb's careers FAQ says it aims to respond to everyone within 3-4 weeks. The same page says Airbnb reviews about 15,000 applications each month and that a candidate who is turned down will likely not receive personalized feedback.

What is Skipper, and why does it matter for the interview?+

Skipper is a workflow engine Airbnb built to keep multi-step processes, such as a charge followed by a reservation update, correct when a server crashes partway through. Airbnb described it on its engineering blog in April 2026 and, per its November 2026 open-source announcement on the same blog, released it under the Apache 2.0 license. Its design choices around checkpoints, idempotent actions and compensation are a ready-made list of the failure cases a strong system design answer covers.

Try it free

Ready to ace your next interview?

Practice with AI-powered mock interviews, tailor your resume, and negotiate your salary, all in one platform.

3-day trial. No credit card required.