Solutions Architect Interview: Rounds, Questions and Prep

Engineers who move toward a solutions architect role tend to prepare for the solutions architect interview the way they prepared for their last software engineering loop: system design drills, a few stories, some cloud revision. Then the interviewer says "I'm the CTO of a retail chain, we have a problem with our stores" and waits. No requirements are offered, and the question is whether you will ask for them or start drawing.
That moment is the difference between the two loops. A software engineer's design round tests whether you can design a system. A solutions architect's rounds test whether you can find out what a customer needs, design something they can afford and run, and explain it to a room where half the people are not engineers. This guide goes through the rounds, the questions and a three-week plan.
The job the loop is testing
The title covers several jobs. At a cloud or software vendor, a solutions architect (or sales engineer, or customer engineer) works with customers before the sale: discovery, architecture, proofs of concept, objections. At a consultancy the work continues into delivery. Inside a large company, a solutions architect designs across internal teams and vendors. Ask the recruiter which of these the role is, who you would report to (sales or engineering) and whether there is a quota attached, because the weighting of the rounds follows from the answers.
Across all three, interviewers look for the same four things: breadth across infrastructure, data, networking and security; depth in at least one area; the ability to run a conversation with a customer; and the ability to present. Coding is usually light or absent, though some teams add a scripting or troubleshooting exercise.
Published detail on any one company's loop is thin and mostly second-hand. As one example, a prep guide to the AWS solutions architect process from Interview Kickstart (last updated December 2024) describes a phone screen split between technical questions and questions on Amazon's leadership principles, then four to six interviews and a 30-minute presentation on a technical problem the candidate has solved, with the interviewers acting as customers. Treat that as a report, not a specification; formats vary by team and change over time. The reliable source is your recruiter, who will normally tell you the list of rounds and whether a presentation is required if you ask.
The architecture design round
The format looks like a system design interview: a prompt, a whiteboard, usually 45 minutes to an hour. The grading is different in three ways.
Requirements come from a person, not a prompt. The interviewer often plays a customer and holds back information until asked. Starting to draw in the first two minutes is the most common way to fail.
Constraints are commercial as well as technical. Budget, the skills of the customer's team, existing contracts, compliance, a migration deadline tied to a data centre lease. A design that ignores them is wrong even if it would scale.
You are expected to recommend. A software engineering design round rewards exploring trade-offs. Here you explore them and then commit: "Given a team of four and the nine-month deadline, I would move the database as it is first and modernise it later, and here is what that costs you."
A workable order for the session: ten minutes of discovery, five on restating the problem and agreeing priorities, twenty on the design from the outside in (users, entry points, compute, data, integration with what exists), ten on resilience, security and cost, and the remainder on a phased migration and risks. Use managed services by name where you know them and say plainly when you are describing a category instead of a product you have run. Typical prompts are migrations (a monolith and its database out of a data centre), hybrid connectivity, disaster recovery to a stated recovery time, a data platform for analytics, and multi-region for a regulated customer.
For the mechanics of scale estimation and component choice, the backend architectural patterns guide covers the building blocks, and the Amazon system design interview guide shows the software engineering version of the round for comparison.
The customer scenario and discovery questions
Some loops run this as a separate role-play; others fold it into the design round. Either way it is the part engineers under-prepare.
The scenario is a meeting. The interviewer is a customer with a stated problem ("our reporting is too slow") and an unstated one (the finance director no longer trusts the numbers). You are assessed on whether you find the second.
Questions worth having ready, grouped by what they uncover:
- Outcome. "What would be different for the business if this were solved?" "How will you measure that?"
- Current state. "Walk me through what happens today when a store closes its till." "What have you already tried?"
- Constraints. "Is there a date this is tied to?" "What can't change?" "Which regulations apply to this data?"
- People. "Who runs this system day to day, and how big is that team?" "Who else needs to agree to a change?"
- Money. "Is there a budget range I should design within?" "What does the current setup cost you to run?"
- Risk. "What is the worst outcome you are trying to avoid?" "What happened the last time there was an outage?"
Listen for the answer that does not fit and follow it. Then play the problem back in two sentences before proposing anything, and get a yes.
The scenario usually includes pressure. The customer says a competitor is cheaper, or that their own architect thinks your approach is wrong, or asks for a feature the product does not have. The graded behaviour is the same each time: acknowledge, ask one question to understand, answer honestly and offer a next step. Saying "it doesn't do that; here is how customers usually handle it, and I can set up a session with a specialist" scores well. Bluffing does not, and interviewers who play the customer are practised at spotting it.
The presentation round
Where it exists, you are asked to present, commonly for 20 to 30 minutes, either on a topic of your choice (often a project you delivered) or on a case sent in advance, followed by questions. The panel may play a mixed customer audience.
What works:
- Lead with the customer's problem and the outcome. The first slide is the situation and the result, not an agenda and not an architecture.
- One diagram, built up. Show the whole design once, simply, then go into two areas. Six dense diagrams lose a mixed room.
- Speak to both audiences. For each major choice, one sentence for the executive (cost, risk, time) and one for the engineer (how it works).
- Leave time. For a 30-minute slot prepare 18 to 20 minutes. The questions are where most of the grading happens.
- Prepare for the interruption. Someone will stop you on slide two with a detailed question. Answer briefly, offer to go deeper at the end, and continue.
Rehearse it aloud, timed, at least three times, and at least once in front of a person. Check the practicalities with the recruiter: slides or whiteboard, remote or in a room, who will be on the panel.
Technical depth areas
Breadth is tested by quick questions across areas; depth by staying on one until you run out. You do not need to be deep everywhere. You need no blank areas and one or two where you can go five levels down.
- Networking. Private networks and subnets, routing, DNS, load balancing at different layers, connecting a data centre to a cloud (VPN versus dedicated links), what happens when a user types a URL.
- Compute. Virtual machines, containers and functions, and when each is the wrong choice; autoscaling and its limits.
- Storage and databases. Object, block and file storage; relational versus key-value versus document stores; replication, backups, read replicas, recovery time and recovery point.
- Security and identity. Least privilege, roles versus users, encryption at rest and in transit, key management, network segmentation, audit logging, the shared responsibility idea.
- Resilience. Availability zones and regions, failover patterns, how to state a disaster recovery plan in terms of cost against recovery time.
- Migration. Rehost, replatform and refactor, how to sequence a migration, how to cut over a database with little downtime.
- Cost. What drives the bill (compute hours, storage, data transfer), reserved versus on-demand capacity, how to estimate a monthly figure to an order of magnitude.
- Data and AI workloads. Batch versus streaming, warehouses and lakes, and increasingly where a model sits in an architecture. If the role touches AI products, read the LLM system design interview guide.
If the role is tied to one cloud, learn that provider's names for each of these and its published architecture guidance; a certification helps you revise but will not carry the scenario rounds.
Behavioural questions for an architect
The behavioural round uses the same question forms as any other, with a customer in most of them. Prepare six to eight stories and make sure the set includes: a customer you told no, a deal or project you lost and what you learned, a time you were wrong in front of a customer, influencing a team that did not report to you, simplifying something for a non-technical audience, and working under a sales deadline without overpromising. If the employer has published principles or values, map your stories to them; the AWS prep guide mentioned above, for instance, reports leadership principle questions in both the phone screen and the loop. The STAR method keeps each story to two minutes.
Twenty questions to practise
These are examples of the kinds of questions candidates should expect, written for practice; they are not taken from any company's list.
Design and migration
- A retailer with 400 stores wants to move its point-of-sale back end out of two data centres. Where do you start?
- Design disaster recovery for an order system with a recovery time of one hour and a budget that rules out a full second copy.
- A media company wants to serve video in three continents. Sketch the architecture and the monthly cost drivers.
- A bank wants analytics on transaction data without that data leaving its country. What are the options?
- How would you migrate a 2 TB relational database with less than ten minutes of downtime?
- A startup's bill tripled in a quarter. How do you find out why, and what do you look at first?
Technical depth
- Explain what happens between a user entering a URL and the page loading, and where it can fail.
- When would you advise a customer against containers?
- Compare a VPN with a dedicated link for connecting a data centre to a cloud provider.
- How do you design access so that no engineer has standing production credentials?
- What is the difference between recovery time and recovery point, and how does each change the design?
- A queue is backing up. Walk me through diagnosing it.
Customer and commercial
- The customer's lead architect says your proposed design is over-engineered. Respond.
- The customer asks for a capability the product does not have. What do you say in the meeting?
- Explain eventual consistency to a finance director.
- A customer wants the cheapest option and it will not meet their stated availability target. How do you handle it?
- Halfway through a proof of concept the requirements change. What do you do?
Behavioural
- Tell me about a time you told a customer or stakeholder no.
- Describe a technical recommendation of yours that turned out to be wrong.
- Tell me about a time you had to learn an unfamiliar domain quickly to be credible in a room.
A three-week plan
Week 1: breadth and stories. Go through the depth list and mark each area as solid, rusty or blank. Spend the week on the blanks, an hour a day, explaining each topic aloud as if to a customer. Write your six to eight stories. Choose the project you would present.
Week 2: design and discovery. Do four design sessions of 45 minutes, each from a one-line customer problem, with the first ten minutes spent only on questions. Two of them should be with another person playing the customer and withholding information; discovery cannot be practised alone, because you already know what you would have hidden. If nobody in your network has done pre-sales work, a session with a solutions architecture coach can fill that seat. Draft the presentation.
Week 3: presentation and full runs. Rehearse the presentation three times, timed, once to a non-engineer. Run one full mock of the scenario round with objections. Answer the twenty questions above aloud. Prepare your questions for the panel, which in this role are also a sample of your discovery skills: ask about their customers, their hardest deals and how success is measured in the first year.
Where coaching fits
Solutions architecture is a small specialty on techinterview.coach: three of the 40 coaches list it as of November 2026, and the solutions architecture coaches page shows who they are and what they have worked on. A session with a coach of that background can cover a design round with a customer role-play, a dry run of your presentation with questions, or a review of your stories. For speaking practice in between, the practice interview runs a voice round with an AI interviewer on system design or behavioural questions and sends a written recap. It is free; sign in to start.
Related guides
- Amazon system design interview: the software engineering design round, useful as a contrast.
- Backend architectural patterns: the components most architecture answers are assembled from.
- System design interview questions by level: how depth expectations rise with seniority.
- LLM system design interview: for roles where the customer's question is about AI workloads.
- STAR method: structure for the customer stories in the behavioural round.
- Mock interviews: how to set up a role-play that gives usable feedback.