The five dimensions, the artefacts to request, and the ten red flags that change deal terms. The framework we use on every TDD engagement.
You are about to invest in or acquire a software company. The pitch deck is persuasive, the founders are impressive, the revenue graph goes up and to the right. Someone has to look under the hood before the term sheet gets signed. This is a field note on what technical due diligence actually looks like in 2026 — the questions that matter, the artefacts to collect, and the red flags that deserve a renegotiated multiple.
What technical due diligence is for
A technical due diligence does not exist to find reasons not to do the deal. It exists to price risk, to identify integration costs, and to surface the specific commitments the seller will need to make in the SPA so the buyer is not left with surprises on day thirty-one.
A good report ends with three outputs: a verdict (proceed, proceed with conditions, walk), a prioritised list of material findings each with an estimated remediation cost, and a short list of specific SPA warranties the buyer should require. Vague verdicts like "codebase is in reasonable condition" are useless. An investment committee needs numbers.
The five dimensions
Our standard framework covers five dimensions, in roughly this order of time spent.
Architecture and scalability. What is the system, how does it fit together, what are the bottlenecks that will constrain growth? Does the architecture match the TAM the deck implies, or does it collapse at 10x current users?
Security and compliance. Where does customer data live, who can access it, is encryption in place where it needs to be, how is authentication handled, what is the posture against the obvious categories (OWASP Top 10, UK GDPR, sectoral regulation)?
Operational maturity. How does the team deploy, monitor, back up, respond to incidents? What does an outage look like? How often do deploys break things? This is often the single most predictive dimension: operationally mature teams can fix anything; operationally weak teams accumulate unseen risk.
Team, process, and key-person risk. How much of the codebase is understood by only one person? What is the hiring pipeline? What happens if the technical founder leaves after close? What are the vendor dependencies (hosting, SaaS tools, open-source projects) that constitute a concentration risk?
Technical debt, dependencies, and legal clarity. What libraries does the codebase depend on? What are the open-source licences? Is any code under restrictive copyleft that would complicate commercialisation? How much technical debt is there, and how much of it is strategic (deferred on purpose) versus accidental?
Artefacts to request, in order
Day 1-2, before you even open the codebase:
An architecture diagram, however informal. If one does not exist, that is itself a finding.
A production infrastructure diagram or a list of cloud resources. "What is in production right now?" is the single most important question.
A data model overview or ERD. Does not need to be perfect; needs to exist.
A copy of the last six months of production incidents. Everyone has them; the question is how they are handled.
A copy of the engineering OKRs or roadmap for the next six months.
The CVs of every engineer still employed, with tenure and primary responsibility noted.
A list of every third-party vendor the system depends on, with contract terms and SLAs.
Access to the source repository (read-only), the CI logs, the error monitoring dashboard, and the production observability surface.
Day 3-5, once you have access:
Run an automated dependency scan (we use several). Note any packages that are unmaintained, have known CVEs, or are on old major versions.
Clone and build the project. Follow the README. Does it work? How long does onboarding take? A new engineer will have to do this; the time it takes you is a lower bound on their experience.
Read the top-five files by commit frequency. These are the parts of the system that change most and reveal the team's conventions.
Read the authentication, authorisation, and payment flows end-to-end. These are where real liability lives.
Pull the schema, compute total row counts, and identify the largest tables. Ask the team about query patterns, indexing, and growth rate.
Review the CI configuration. Does it run tests? Do the tests pass? What is the coverage? What is the merge policy?
Day 6-7, synthesis:
Write up findings against the five dimensions. Each finding gets a severity (material, moderate, minor), an evidence pointer, and an estimated remediation cost in weeks or thousands of pounds.
Draft a verdict with conditions.
Draft a list of specific warranties for the SPA.
The ten red flags that matter
Not all findings are created equal. These are the ones that genuinely change deal terms.
- 1No one in the current team wrote the authentication code. Deep unknowns in security-critical code are an outsized risk.
- 2Backups exist but have never been restored. A backup that has never been restored is not a backup. Insist on seeing a documented restore drill from the last twelve months.
- 3The database has no migration history, or migrations are applied manually. This signals a team that cannot evolve the system under pressure.
- 4Customer PII is stored in cleartext or with weak hashing. Not a dealbreaker in some sectors; renegotiable in all of them.
- 5A single engineer accounts for more than 60 percent of recent commits. Concentration risk; price it.
- 6"We use AI" turns out to mean "we proxy ChatGPT." Fine if it is disclosed; worth a long conversation about moats if not.
- 7No written incident playbook. Teams that invent incident response during the incident perform badly.
- 8The deployment process requires manual steps. Every manual step is a risk and a hiring handicap.
- 9The .env file has been committed to git at some point in history (check git log, not just HEAD). If yes, assume secrets rotation is needed; sometimes the deal gets priced accordingly.
- 10Technical founder is also the only person who can deploy, has the only AWS root credentials, and is the primary system admin. Walk away, or accept that you are buying a key-person risk you cannot insure against.
What a strong report actually looks like
An executive summary of one page, in plain English, written for a managing partner who will not read any other page. A verdict in a single sentence. A table of findings with severity, evidence, and cost. A section per dimension with specific observations. An appendix of evidence — screenshots, log excerpts, repository paths, conversation notes with the target's engineers. Signed and dated by the practitioner who did the work.
The report takes three to seven working days to produce and costs between eight and twenty thousand pounds depending on codebase size and deal complexity. That is a rounding error on a seven or eight figure transaction. The deals where diligence is skipped or handed to a generalist consultant are the deals that produce the most expensive surprises.
Why we wrote this
AANDA Labs provides pre-investment technical due diligence as a standard offering. The questions and framework above are the ones we use, in that order. If you are about to sign a term sheet, you should either be able to answer each of these questions about the target with confidence — or you should have an independent third party answer them for you. Not later. Before you sign.