What Do Investors Check About Your Corporate Structure at Series A?

Ideally, six to twelve months before you expect to raise. In practice, most founders begin when they receive a term sheet or when investor conversations become serious. At that point, you typically have four to six weeks before a data room request arrives. That is enough time to fix clean documentary issues — missing signatures, absent assignment agreements, basic cap table errors — but not enough time to remediate a fundamentally compromised training dataset or renegotiate a contractual structure that embeds IP in the wrong entity. The earlier you start, the more options you have.

A founder IP assignment is a written agreement in which each founder transfers ownership of any IP they created — code, models, datasets, inventions, designs — to the company. It matters at Series A because investors are buying equity in the company, and the value of that equity depends on the company owning the technology. If a founder retains personal ownership of the core IP — even unintentionally, through an unsigned or inadequate agreement — the investment is buying equity in a shell. Courts and investor counsel look at the date the IP was created, the date of incorporation, and the date of assignment. Pre-incorporation work is a specific risk area that requires explicit treatment in the assignment agreement. See our detailed guide on Founder IP Assignment for AI Startups.

It depends on the specific licence version and your use case. LLaMA 2 required Meta’s written approval for commercial use above a threshold number of monthly active users. LLaMA 3 and later versions have different licence terms but include an Acceptable Use Policy that prohibits specific applications. Mistral models are generally permissively licensed but some variants carry restrictions. The general principle is: never assume an open-source model permits commercial deployment without reading the full licence, Acceptable Use Policy, and any supplementary terms. Investor counsel will pull these licences during DD and map them against your actual product use case. A mismatch after term sheet is one of the hardest problems to fix quickly.

The IP section of a Series A data room for an AI company should contain: (1) all founder IP assignment agreements, signed and dated; (2) a schedule of all employees and contractors who contributed to the product, with copies of their relevant contracts and assignment clauses; (3) a training data register listing each dataset used, its source, and the licence or rights basis for commercial use in AI training; (4) copies of all model licences and API provider agreements with the relevant terms highlighted; (5) a change-of-control analysis flagging any agreements with CoC provisions; and (6) if EU exposure is present, the EU AI Act risk classification memo and compliance roadmap. Presenting a well-organised IP section is itself a signal to investors — it demonstrates operational maturity and reduces the perceived risk of investing.

Directly, EU AI Act compliance status does not have a standard valuation formula at Series A. Indirectly, it affects valuation in two ways. First, EU investors and investors with EU LPs increasingly treat regulatory readiness as a de-risking factor — a company with a documented compliance roadmap for its EU market is priced as lower-risk than one with no awareness of its obligations. Second, for companies building high-risk AI systems (as defined by EU AI Act Annex III — HR tools, credit scoring, healthcare AI), a credible compliance programme is becoming a prerequisite for enterprise sales in the EU. Enterprise customers increasingly require EU AI Act compliance documentation as part of procurement. A company that cannot demonstrate a path to compliance cannot win the enterprise contracts that justify its Series A valuation. See the August 2026 high-risk deadline for current timing context.