Somewhere between the term sheet and the money in the bank, an investor sends a technical reviewer to look at your product. Sometimes it's a partner who used to be a CTO, sometimes a specialist firm, sometimes an engineer from a portfolio company. They have a few days, a questionnaire and a call with your team.
Their job is not to judge whether your code is beautiful. It is to answer three questions for the investment committee: can this product scale with the plan in the deck, is there a hidden liability (security, IP, data protection, one person who knows everything), and how much will it cost to fix what's broken. This checklist is organised around those questions, and it's the one we use when we prepare companies for a round.
Technical due diligence questions, area by area
Expect questions in seven areas. Answer each with a short document or a screen share, not a promise.
1. Architecture and scalability
- Draw the system: clients, services, databases, queues, third-party APIs. Where does state live?
- What breaks first at 10x the current load, and what's the plan when it does?
- Single points of failure: one database instance, one region, one server someone set up by hand.
- Why did you choose this stack, and would you choose it again? "We knew it" is a fine answer at seed. "Nobody decided" is not.
2. Code quality and engineering process
- Is every change reviewed and tested before it reaches production? Reviewers will often ask to see a few recent pull requests.
- How do you deploy, how often, and how do you roll back? A CI/CD pipeline with a one-click rollback answers most of this. See our list of CI/CD anti-patterns for what reviewers frown at.
- Test coverage matters less than whether the critical paths (signup, payments, the core workflow) have tests.
- How much known technical debt is there, and is it written down anywhere? A tracked list scores better than a clean-looking codebase with no list.
3. Security and compliance
- MFA and single sign-on for every human, no shared accounts, no long-lived access keys on laptops.
- Secrets in a secrets manager, not in the repository or in environment files passed around on Slack.
- Backups that have actually been restored, with the date and how long it took.
- Any past security incidents, how you found them and what you changed.
- If you sell to enterprises: SOC 2 or ISO 27001 status, or a realistic plan. Our SOC 2 checklist for AWS covers what that plan looks like. If your customers are in scope for NIS2, expect supply-chain questions too.
4. Infrastructure and cost
- Is infrastructure defined in code (Terraform, OpenTofu, CloudFormation), or clicked together in a console?
- Are production and development separated, ideally in different cloud accounts?
- What does the cloud bill look like per month, per environment and per customer, and how does it grow with usage? Reviewers compare it with gross margin assumptions in your model.
5. Team and key-person risk
- Who knows how each part of the system works? If one engineer leaves, what stops?
- Who has admin access to production, the domain registrar, the app stores and the payment provider? Is any of it in a personal account?
- What does the hiring plan in the deck assume, and is it realistic in your market?
6. IP ownership and open-source licences
- Does the company own all the code? This is the question that surprises founders most. Under Greek copyright law, the economic rights in software written by an employee in the course of their job pass to the employer automatically (Article 40 of Law 2121/1993). Code written by a freelancer, an agency or a co-founder before incorporation does not. It needs a written assignment, and Article 14 of the same law makes an unwritten transfer void.
- Which open-source licences are in your dependencies, and do you meet their obligations? Permissive licences (MIT, Apache-2.0, BSD) need attribution. Copyleft licences can require you to publish source code. The AGPL's section 13 extends this to software users reach over a network, so a modified AGPL component inside a SaaS product is a real issue, not a theoretical one.
- Any code copied from a previous employer, or written by founders while employed elsewhere?
7. Data protection and GDPR
- What personal data you hold, where it's stored and who can see it. The GDPR requires a record of processing activities (Article 30), and most reviewers ask for it or something like it.
- Data processing agreements with every processor (cloud, email, analytics, support tools) under Article 28.
- Transfers outside the EU and the legal basis for them. US SaaS tools are the usual gap. Our post on GDPR-friendly logging in EU regions covers a common place personal data leaks into.
- A breach process that can meet the 72-hour notification deadline in Article 33.
The five documents to prepare
Five documents answer most of a reviewer's questions before they ask them. Put them in a data-room folder, keep each to a few pages, and date them.
| Document | What it should contain | Effort |
|---|---|---|
| Architecture diagram | One page: components, data stores, external services, regions, how traffic flows. Plus a paragraph on what changes at 10x. | 1 to 2 days |
| Data-flow map | Which personal data enters where, which systems and vendors it touches, where it's stored and for how long. Doubles as the base of your Article 30 record. | 2 to 3 days |
| Incident history | The last 12 months of outages and security events: date, impact, duration, root cause, what changed. Our incident response runbook template has the postmortem format. | 1 day if you kept notes |
| Cost breakdown | Monthly cloud and SaaS spend by service and environment for the last 6 to 12 months, cost per customer or per transaction, and the main drivers. | 1 day |
| Access list | Every system with admin access (cloud, code hosting, DNS, registrar, app stores, payments, email), who holds it, MFA status, and which accounts are company-owned. | 1 day |
Add a short technical roadmap and a list of known technical debt with rough effort estimates. Showing that you already know the weak spots is the single most reassuring thing you can do in due diligence.
Three checks you can run this afternoon
You don't need a tool vendor to find most of what a reviewer will find. These commands surface the access, key-person and licence issues that come up most often.
Who has access. The IAM credential report lists every IAM user with password, MFA and access-key status. GitHub's API lists organisation members and outside collaborators.
# AWS: IAM users, MFA and access keys (run per account)
aws iam generate-credential-report
aws iam get-credential-report --query Content --output text | base64 --decode | cut -d, -f1,4,5,8,9,10,11 > iam-users.csv
# GitHub: org members and outside collaborators
gh api orgs/YOUR_ORG/members --paginate --jq '.[].login' > gh-members.txt
gh api orgs/YOUR_ORG/outside_collaborators --paginate --jq '.[].login' > gh-outside.txt
If you've moved people to IAM Identity Center, the credential report should show almost no human users. That is the answer reviewers want.
Key-person risk. Who has written the code in each part of the repository over the last year?
# Top contributors over the last 12 months
git shortlog -sn --no-merges --since="12 months ago" HEAD | head -5
# Main author per top-level directory
for d in */; do
echo "$d $(git log --no-merges --since='12 months ago' --format='%an' -- "$d" | sort | uniq -c | sort -rn | head -1)"
done
If one name dominates every directory, that's your key-person risk, in a form you can show a reviewer together with your plan to spread the knowledge.
Open-source licences. Syft builds a software bill of materials from a directory or container image. Filter it for licences that carry source-disclosure or commercial-use conditions, then review each hit by hand.
syft dir:. -o syft-json=sbom.json
jq -r '.artifacts[]
| [.name, .version,
([.licenses[] | if .spdxExpression != "" then .spdxExpression else .value end]
| join("; "))]
| @tsv' sbom.json | grep -Ei 'AGPL|GPL|SSPL|BUSL|Commons-Clause'
A GPL or LGPL hit is not automatically a problem. It depends on how the component is used and distributed, which is a question for a lawyer once you have the list. Not knowing the list is the problem.
Red flags and how to fix them in 30 to 60 days
Most red flags are fixable in weeks, and a reviewer who sees a dated fix plan treats them very differently from one who discovers them. This is the order we'd work in.
| Red flag | Why it worries an investor | Fix | Time |
|---|---|---|---|
| Freelancer or agency code without a written IP assignment | The company may not own its product | Signed assignment agreements from every past contributor, reviewed by a lawyer | 1 to 4 weeks, depends on reaching people |
| Production in a founder's or contractor's personal account | Ownership and continuity risk | Move to company-owned accounts under an AWS Organization; transfer domains and app store listings | 1 to 2 weeks |
| No MFA, shared logins, long-lived keys | Breach likelihood | SSO with MFA, delete IAM users and keys, OIDC for CI | 1 to 2 weeks |
| Backups never restored | One bad day ends the company | Automated backups, a documented restore test, the time recorded | Days |
| Infrastructure clicked together by hand | Can't rebuild, can't review changes | Import the core (network, database, compute) into Terraform or OpenTofu | 2 to 6 weeks |
| One engineer knows everything | Key-person risk | Architecture docs, runbooks, pairing, a second person deploying | 4 to 8 weeks, ongoing |
| Cloud cost growing faster than revenue | Gross margin story breaks | Tagging, rightsizing, commitments; see our AWS cost checklist | 2 to 4 weeks |
| No GDPR records, missing DPAs | Regulatory and customer-contract risk | Data-flow map, Article 30 record, signed DPAs, privacy notice updated | 2 to 3 weeks |
| Copyleft component in a distributed or network-served product without compliance | Forced disclosure or rewrite | Replace the component or meet the licence terms, with legal sign-off | Varies |
How long before the round to start
Start two to three months before you expect a term sheet. That leaves time for the 30 to 60 day fixes without doing them in the same weeks as the round itself, when the team is answering questions and still shipping.
- Weeks 1 to 2: run the checks above, write the five documents, list every red flag honestly.
- Weeks 3 to 6: fix ownership, access and backups first, because they are fast and they are what reviewers check first. Chase IP assignments in parallel.
- Weeks 6 to 10: infrastructure as code, cost work, documentation to reduce key-person risk.
- Before the data room opens: update the documents with what you fixed and a dated plan for what's left.
An independent review before investors run theirs is the cheapest way to find out what they'll find. Our five-dimension cloud audit covers the infrastructure, security and cost side.
If you're heading into a round without a senior technical person on the team, this is what a fractional CTO is for: preparing the data room, sitting in the reviewer call and owning the fix plan. If you mainly need the infrastructure side checked and documented, start with a cloud audit.
Frequently asked questions
What is included in a technical due diligence checklist?
A technical due diligence checklist covers architecture and scalability, code quality and engineering process, security and compliance, infrastructure and cloud cost, team and key-person risk, IP ownership and open-source licences, and data protection. Prepare an architecture diagram, data-flow map, incident history, cost breakdown and access list before investors ask.
How long does technical due diligence take for a startup?
The reviewer's work usually takes days to a couple of weeks, depending on the round and how ready you are. Preparation is what takes time: allow two to three months before the round to write the documents and fix red flags such as missing IP assignments, personal-account ownership or untested backups.
Do investors check open-source licences during due diligence?
Many do, especially at Series A and later. They want to know which licences are in your dependencies and whether you meet their obligations. Copyleft licences such as GPL and AGPL can require you to publish source code, and AGPL applies to software users reach over a network. Generate an SBOM and review the copyleft hits with a lawyer.
Does a Greek startup own code written by freelancers?
Not automatically. Under Greek Law 2121/1993, economic rights in software written by employees pass to the employer, but code written by freelancers, agencies or founders before incorporation needs a written assignment. Article 14 makes unwritten transfers void, so collect signed assignments from every past contributor before due diligence.