Glossary
AI due diligence: a full check before you invest in AI
In short
AI due diligence is a full review of a company before decisions about artificial intelligence: its processes, data, systems, people and legal limits.
Business and adoption · AI due diligence
MMOOZZGG.com editorsChecked: 1 min read
For a child
Before building a big tree house at the summer cottage, dad first looks at the tree itself. Are the branches strong, is there any rot, will it hold the boards and the kids? Only then do they buy materials. Adults look at their company the same way before they bring smart programs into it: the branches might turn out to be weak.
In plain words
Due diligence, a full review of a company, comes from deals: before buying a business, the buyer studies its documents and risks. In the AI version, you find out whether the company is ready for neural networks. You look at how processes work and where time is lost, how information is stored, which programs are installed and whether they can be connected, and who will use the new tool. It is like a mechanic checking a car before a long trip: it is cheaper to find a fault in the garage than on the highway.
For business
The review answers the question of whether to invest in AI now and where exactly. It reveals scattered data, processes without an owner, systems without integrations, and the limits of 152-FZ, Russia’s personal data law, on sending personal data to cloud models. The price depends on the company’s size and the number of departments and systems. The result: a list of risks, the readiness of each process and priorities. Without it, money often goes to pilots that cannot be scaled later. Investors also order such a review before a deal to assess a startup’s AI product. The findings become the basis of the AI solutions map and the rollout budget.
For a specialist
AI due diligence combines two scenarios. Operational: an AI readiness assessment of the organization across processes, data, IT architecture, skills and regulatory limits. Investment: a technical audit of an AI product in a deal, which checks the origin of and rights to training data, the reproducibility of training, quality metrics, dependence on external APIs and model licenses. In both cases it helps to rely on the NIST AI Risk Management Framework (January 2023): its Govern, Map, Measure and Manage functions set how to describe the system’s context, measure risks and assign measures. For Russian companies, a check of personal data processing under 152-FZ and of data storage localization is added. The result is set out as a risk register and a readiness matrix by process.
Example
A retail chain wanted to launch an AI assistant for support. The review showed that the knowledge base was out of date and spread across several systems, so the documents were put in order first, and the launch moved by a quarter.
Related terms
Questions and answers
Sources
Where companies use it
Not sure whether your company needs this?
AI solutions map