The technical due diligence questions that matter are the ones whose answers change the price, the deal structure or the 100-day plan. Ask the CTO and engineers about what breaks first under growth, how a change reaches production, who holds the knowledge and the passwords, what the security record shows, who owns the code, and what the AI really does. Interview engineers separately from the CTO, and ask to be shown things rather than told about them.
This article is about the conversations. The documents and data to request, area by area, are in the technology due diligence checklist and our numbered technical due diligence checklist, and how the answers end up in the deliverable is covered in what a technical due diligence report should contain.
The best interview question in technical due diligence is a request to be shown something.
I've sat on both sides of these interviews: as a founder and CEO answering questions in a sale and a merger, and as an investor asking them. Two things are true from both chairs. People prepare for the questions they expect, so the expected questions tell you little. And the most useful information comes out sideways, when someone explains how they'd do a thing and you notice who they say they'd ask.
Technical due diligence questions to ask the CTO
Each question below has a reason and a rough guide to what a good answer and a worrying answer sound like. They're starting points. Follow the answer, not the list.
Architecture and scale
"If customer volume doubled next year, what breaks first?" This tests whether the CTO knows the system's limits. A good answer names a specific component, explains why and gives a rough plan and cost ("the reporting queries on the main Postgres instance; we'd move them to a read replica, about a month's work"). A worrying answer is "it scales" or "it's all in the cloud", or a long pause.
"Walk me through what happens when a customer does the one thing they do most." This gets the CTO to describe the real architecture rather than the slide. Good is a clear, specific path through services and data stores. Worrying is an explanation that doesn't match the diagram in the data room.
"What would you rebuild if you had six months and no feature requests?" Every CTO has an answer. A good one is honest and bounded. A worrying one is "most of it", which is a re-platform hiding in plain sight, or "nothing", which is not credible.
Delivery and engineering practice
"Show me the last change that went to production, from the ticket to the deploy." The best interview question in technical due diligence is a request to be shown something. A good answer opens the pipeline on screen: a pull request, a review, automated tests, an automated deploy, all within days. A worrying one is a manual release run by one named person, or a long explanation of why it's different this week.
"How often do you deploy, and what share of deploys need fixing afterwards?" These map to DORA's delivery metrics, deployment frequency and change fail rate. Good teams know their numbers roughly. Teams that have never measured them usually deploy less often than they think.
"What did you ship from last year's roadmap?" Ask for the roadmap from twelve months ago and go through it. A team that delivered most of it on time is a team whose current roadmap you can believe. If they delivered a third, discount the new-product revenue in the plan accordingly.
Code quality and AI-assisted development
"How much of your code is now written with AI tools, and how is it reviewed?" This is a newer question and an important one. Google's 2025 DORA report, based on nearly 5,000 technology professionals, found AI use among developers had reached 90%, and described AI as an amplifier: strong teams get faster, fragmented teams see their weaknesses magnified. The 2025 Stack Overflow Developer Survey found more developers distrusted the accuracy of AI tools (46%) than trusted it (33%). A good answer describes which tools are used, under what rules, and the same review and test gates as human code. A worrying one is either "we don't use it" (unlikely to be true) or heavy use with no change to review or testing.
"Which part of the codebase is everyone afraid to touch?" Every codebase has one. Good is a team that can name it, explain why and has a plan. Worrying is when the feared part is the billing or permissions code, and nobody has a plan.
Team and key-person risk
"If you were hit by a bus tomorrow, what would stop working, and who would fix it?" Blunt, but it works. A good answer names people and systems and mentions documentation someone else has used. A worrying answer is a pause followed by "well, I'd be reachable". I go into this in detail in key person risk in due diligence.
"Who has admin access to the production cloud account, the domain registrar, the app store accounts and the code repository?" The answer should be at least two named people, using company-owned accounts with multi-factor authentication. One person's personal email on the domain registrar is a finding.
"Who have you lost in the last two years, and why?" Attrition among engineers is a leading indicator. Good is low churn with explainable departures. Worrying is several senior departures, especially if they left for the same reason.
Security and data
"Tell me about your last security incident." Every business has had one, even a minor one. A good answer describes it, the response and what changed. "We've never had one" usually means they weren't looking. Check the answer against the incident log and anything reported to the ICO.
"Where does personal data live, including the places it shouldn't?" A good answer covers production databases, backups, logs, analytics tools, support systems and any copies on laptops or in spreadsheets. A worrying answer covers only the main database.
"What's still open from the last pen test?" Good is a short list with dates. Worrying is critical findings more than a few months old.
IP, open source and suppliers
"Who wrote the first version of the product, and were they employees?" Under section 11 of the Copyright, Designs and Patents Act 1988, copyright in work made by an employee in the course of employment belongs to the employer; work by contractors and agencies does not, unless it is assigned in writing. Founders who wrote code before the company existed are in the same position. A good answer comes with signed assignments. A worrying one is an early agency or freelancer nobody has a contract with.
"Which outside service, if it disappeared or doubled its price, would hurt most?" Good is a CTO who knows, has read the contract and has a fallback. Worrying is a core feature built on an API with a short termination clause.
Cost to run
"Why did the cloud bill go up last year?" A good CTO can explain the main lines. If hosting cost grew faster than revenue and nobody can say why, that is money in the 100-day plan and possibly a question about run-rate EBITDA.
AI claims
"Show me the AI feature running on real data, and tell me how many customers used it last month." This separates AI that exists from AI in the CIM. Good is a live demonstration, usage numbers, the model or vendor behind it and a cost per call. Worrying is a recorded demo and no usage data. The fuller method is in AI washing in due diligence, and the full question bank is in our AI due diligence questions.
How to run technical due diligence interviews with the tech team
The questions matter less than the way the interviews are set up. Here's how I'd run them for a software business with somewhere between five and thirty engineers.
Who to interview
The CTO or technical co-founder, twice. The most senior engineer who isn't the CTO. One engineer who joined in the last year, because they remember what was confusing. Whoever is on call or handles production incidents. The product lead, if there is one. If a contractor or agency writes a meaningful share of the code, one of them too.
Leave the CEO and CFO out of the technical sessions. Their presence changes the answers.
How long, and in what order
Start with a 90-minute session with the CTO to get the overview and the CTO's own account of strengths and problems. Then the engineers, 45 to 60 minutes each, on separate calls. Then a second CTO session of 60 minutes, near the end, to reconcile what you heard with what you were told at the start, and to raise the findings you are minded to put in the report.
That second session is where much of the value is. A CTO who, faced with a gap between their first account and the evidence, says "you're right, that's a problem, here's what I'd do" is someone you can work with. One who argues every point is telling you something about the next five years.
Interview people separately
Engineers should be interviewed without the CTO or founders on the call. Say at the start that you're trying to understand the system, not to assess them. Ask each engineer some of the same questions: how a change reaches production, what they'd fix first, who they'd ask if something broke at 2am. Then compare.
When three engineers describe deployment the same way, you can believe it. When the newest engineer describes a process with two manual steps that the CTO didn't mention, the newest engineer is usually right.
What to listen for that people don't say
Most of the useful signal in tech team interviews comes from what isn't said directly.
Listen to who gets named. If every answer to "who would you ask?" is the same person, you have found your key-person risk, whatever the org chart says. Listen to "we" and "I" too. A CTO who says "I" about every system may be describing reality, and that's a finding.
Notice where the answers get vague. People are precise about the things they are proud of and general about the things that worry them. A detailed account of the test suite followed by a hand-wave about the data migration is a pointer to the migration.
Watch for rehearsal. A polished, identical answer from three people about security suggests a briefing. Ask a follow-up nobody would have prepared for: "when did you last restore from a backup, and how long did it take?"
And notice tone about the business. Engineers who talk about customers and revenue tend to work in teams that ship what matters. Engineers who only talk about technology, or who are visibly tired of the founders, are telling you about retention.
Practicalities
Agree the interview list and the confidentiality position with the seller up front. Engineers often don't know about a sale early in a process, so the seller may want to limit interviews to a small circle. That's reasonable, but get it written into the diligence plan, and record in the report's scope section who you couldn't speak to.
Do it on video with screen sharing. You can ask someone to open the cloud console, the incident log or the deployment pipeline while you watch, which is better evidence than any document.
Take notes during the call and write them up the same day, tagged to the area they relate to. By the second CTO session you want a short list of discrepancies to raise, not a pile of transcripts.
Questions not to bother with
Some questions appear on every list and rarely change a decision. "What's your tech stack?" is in the data room. "Do you follow agile?" gets a yes from everyone. "Is your code well documented?" gets a yes from everyone too; ask to see the document a new joiner used in their first week instead. "What is your test coverage?" tells you less than "show me the tests for the billing code". Spend the time on questions whose answers you can check.
Interviews are the part of technical due diligence that can't be done from a data room, and the part most often rushed. They're built into every technical due diligence engagement we run, and the fees are set out in how much technical due diligence costs in the UK. If you have a deal coming up, the technical due diligence page explains how we scope them.
Questions
Should the CTO be in the room when engineers are interviewed?
No. Interview engineers without the CTO or founders present, and say at the start that the aim is to understand how the system works, not to assess individuals. Engineers speak more freely about deployment pain, fragile code and who really knows what when their manager isn't listening. Most sellers accept this if it is agreed in the diligence plan rather than requested on the day.
What if the engineering team doesn't know about the sale?
This is common, especially early in a process. Agree with the seller who may be told, under what story and when. Often the CTO and one senior engineer are brought inside the confidentiality circle first, and the wider team later or not at all. If you can only interview the CTO, say so in the report's scope section and weight the code and commit history more heavily.
How many people should you interview in technical due diligence?
For a software business with 5 to 30 engineers, four to seven interviews is usually enough: the CTO twice, the most senior engineer, one more recently hired engineer, whoever handles production incidents, and the product lead. Larger teams need more, sampled across teams. Beyond that, extra interviews tend to repeat what you already know, and the time is better spent in the code.
Can technical due diligence interviews be done remotely?
Yes, and most are now. Video calls with screen sharing work well, and in some ways better, because you can ask someone to open the deployment pipeline, the cloud console or the incident log while you watch. In-person time is worth it for the first CTO session on a larger deal, where the relationship with incoming investors starts.
What documents should you read before interviewing the CTO?
Read the architecture diagram, the org chart with tenure, the last pen test report, the last year's roadmap and the cloud bills before the first interview, and skim the commit history. You want to spend interview time on the gaps between what the documents say and what the evidence shows, rather than having the CTO describe things you could have read.