AI technical due diligence.
An independent look at an AI product's models, code, data flows and security, for investors who are buying and founders who are about to be reviewed.
What a reviewer asks first
Who it is for, and what each side gets.
The review is the same. The report is written for whoever has to act on it. It sits alongside our wider tech audit.
Investors and acquirers
You want to know what you are buying. We review the AI product independently and report what holds up, what does not, and what it would take to fix.
Founders before a raise
You want no surprises in the data room. We run the same review on your own product first, so you find the gaps and fix the ones that matter.
Buyers of an AI product
An enterprise customer may ask for a technical review of a vendor's AI before signing. We can run it, or help the vendor prepare the answers.
What we review, and what it tells you.
Reviewers rarely mind that AI wrote part of the code. They mind that nobody can show how it works, who can reach what, and where the data goes. Our guide to technical due diligence for vibe-coded startups covers that in detail.
Models and vendors
What we checkWhich models run where, how they are called, what each vendor's terms say about retention and training, and how hard it would be to switch.
What it showsHow exposed the product is to one vendor's price, terms or outage.
Data flows and personal data
What we checkWhat enters prompts, logs and vector stores, where each copy goes, and which vendors see it. In India, we check this against the DPDP Act.
What it showsWhether the product can meet privacy duties and enterprise questionnaires.
Quality and evaluation
What we checkWhether there are tests for model output, how changes to prompts and models are checked before release, and how failures are recorded.
What it showsWhether the product is measured, or only demonstrated.
Security and access
What we checkSecrets, who can change production, tenant isolation, and the AI-specific risks, mapped to the OWASP Top 10 for LLM Applications 2025.
What it showsThe risks a buyer's security team will find first.
Architecture and code
What we checkWhether the system builds from a clean checkout, how the AI parts are separated from the rest, and how much code nobody can explain. This includes code written by AI tools.
What it showsHow costly it is to keep building on.
Cost and operations
What we checkWhat each AI feature costs to run, whether spend has limits, and whether there are logs, alerts and an incident process.
What it showsWhether growth makes the unit economics better or worse.
Before a Series A or B, get these ready.
There is no official investor checklist for AI products. This is our summary of what investors and buyers commonly ask for. For the data and privacy side, see our guide to DPDP readiness for a fundraise or IPO.
Access and ownership
Repositories held by the company, a list of who can change production, and signed IP terms from staff and contractors. Your counsel confirms the legal side.
A one-page data-flow map
Every AI feature, the data it uses, the vendor it calls, and the plan or contract in place.
Proof the AI works
A test set, the results over time, and a short note on known failures and how you handle them.
Keys, spend and limits
A secret scan with a rotation log, per-user and per-day spend limits, and loop limits on agents.
A vendor and model register
Each model and AI vendor, its terms, and the fallback if it fails or changes price.
Someone besides the founder
A second person who can explain and change each core module without an AI assistant open.
What you get, and what it costs.
We do not publish a price, because scope decides the effort: the number of AI features, repositories and vendors, and whether you need a written report for a data room. We quote after a short call.
This is a technical review. It is not a legal opinion, a certificate or a guarantee of how a deal will go.
A ranked findings register
Each finding with severity, evidence, an owner and a suggested fix.
An AI data-flow map
Where data goes, including every model provider and tool.
An architecture and vendor-risk note
How the system fits together and where it depends on one party.
A summary for the deal team or board
Short enough to read in one sitting, with the open questions marked.
A fix list for founders
What to change before a data room opens, in priority order.
How it runs.
Questions an investor’s team may raise on architecture are in our list of AI architecture review questions.
- Step 1
Scope
We agree who the report is for, which products and repositories are in, and what evidence we can see.
- Step 2
Collect
We take read-only access and ask for the records above, with the company’s team.
- Step 3
Review
We read the code, trace the data, test access and check how the AI behaves under bad input.
- Step 4
Report
We write up what we found with evidence, and walk the readers through it.
- Step 5
Fix, if you want it
Founders can ask Mono to fix what the review found. For an investor's review, we keep the two separate.
Related services.
Tech Audit
The full audit of infrastructure, code, security, spend, AI tools and data flows.
See Tech AuditSecurity questionnaires
Help with the AI questions in enterprise security reviews.
See questionnaire helpShadow AI audit
Find the AI tools your team uses without approval, and what data goes into them.
See shadow AI auditAI due diligence FAQs.
Is this a legal opinion or a certification?
No. Mono is not a law firm, an auditor or a certification body. We review the technology and report what we find. Questions on IP ownership, licences and contracts go to your counsel.
Does it matter that AI tools wrote the code?
Not by itself. Reviewers care about what that changes: whether anyone can explain the code, which tools saw your source and data, and whether the output is safe to build on. A record of which tools were used helps.
What does an AI technical due diligence cost?
We do not publish a price. The effort depends on how many AI features, repositories and vendors are in scope, and on whether you need a written report for a data room. We scope it on a short call and quote after that.
Is there a standard checklist that investors use?
No official or standard investor checklist exists for AI products. Investors and buyers commonly ask for evidence on access, data flows, testing, vendor terms and spend, which is what this review covers.
Can founders use this before they raise?
Yes. We run the same review on your own product, so you see the gaps before an investor does. Reviewers expect findings. What builds confidence is a team that already knows its weak points and has an owner against each one.
Does the DPDP Act come into this?
For products that handle personal data of people in India, buyers ask whether you can meet the DPDP Act. The core duties start on 13 May 2027. There is no government DPDP certificate, so buyers look at evidence such as data maps, contracts and logs.