Last year, our team asked the finance lead at a North American growth-equity firm a simple question: Do you keep a master checklist of your side letter obligations?
“No. We are a little more ad hoc and not in a good way.”
And while nobody at that firm thought the arrangement was fine, it had simply never risen high enough to become a priority to resolve. But the ability to take an ad hoc approach—and stay on top of everything—is rapidly diminishing.
Side letters are on the move
Bain estimates there are roughly 32,000 unsold portfolio companies worth about $3.8 trillion, with holding periods now averaging around seven years, up from five to six for most of the previous decade. Longer holds mean the same manager carries the same investor obligations for longer, across more funds, and likely with more staff turnover along the way.
They are also changing hands. Global secondary transaction volume reached $240 billion in 2025, up 48% year over year and the largest year on record. Every transfer is a side letter moving to a new holder and an obligation that has to be remapped.
Meanwhile, operations are no longer just a formality. 87% of institutional investors have declined or reconsidered a fund commitment due to AML/KYC concerns, and 41% cite reliance on manual or paper-based compliance as a top operational due diligence risk. Regulators are taking up a similar closer look. As the SEC’s Division of Examinations has observed, private fund advisers routinely “did not follow practices described in their limited partnership agreements (‘LPAs’), operating agreements, private placement memoranda, due-diligence questionnaires, side letters or other disclosures,” examinations that have resulted in “a range of actions, including deficiency letters and, where appropriate, referrals to the Division of Enforcement.”
The problem isn’t discipline
The assumption is often that the firms that get caught were sloppy, but they usually weren’t. In an Ontra and Wakefield Research survey, 43% of asset managers said they spent four to six months building a compendium that was outdated the moment it was finished. That means half a year of careful work resulted in no meaningful efficiency gains.
It’s also often a process whose only failsafe is one person’s memory of what is in the documents. One legal counsel at a global insurance investor walked us through her “process”: she opens the side letter file every month to confirm nothing has been missed, and reviews it again whenever a new project starts. As another legal counsel at an EMEA asset manager explained to us, reminders mattered more than reports: “No one is going to go check the side letters if they’re not aware that there is something like that in there.”
That manual approach is impossible to scale as obligations grow. Our data shows that the median number of obligations per side letter has risen by roughly a third since 2021, and about a million provisions were added in 2025 alone.
Mapping is not monitoring
It’s tempting (and understandable) to assume that a powerful model can solve this: point any AI tool at a pile of documents, ask it questions, and get the answers. After all, general-purpose tools are great at retrieval, and one will find most of what someone asks for. But as Matt Crowley, former VP of Product at Ontra, said on a recent webinar: “It’s not that valuable to avoid four out of five landmines if you still step on the fifth.”
A model can only surface an obligation that has already been identified, structured, and verified. Ask a general-purpose tool what you owe, and it will answer based on whatever it can find, with no record of who checked it or when, which is the opposite of what an examiner wants. Proof is a documented chain: this is the obligation, this is the document it came from, this is who completed it, and this is the date. Or, as our CTO Eric Hawkins put it: a prompt is not a repository.
How to deliver scalable outcomes with AI
Delivering on side letter obligations at scale, while staying compliant, requires firms to look beyond building static compendia. Relying on costly external counsel for ongoing compliance questions doesn’t scale either: it adds expense without investing in a system that improves over time.
AI can solve for both, but only if it’s built on the right foundation. A general-purpose tool will retrieve an obligation; it won’t tell you who verified it or when, which is the entire question an examiner is asking. Purpose-built solutions:
Structure fund documents into a trusted source of truth: Ontra’s AI identifies and categorizes legal clauses across LPAs, side letters, and credit agreements into a searchable compendium, verified by Ontra’s expert team for accuracy. Obligation summaries, contract language, and precedent then surface through AI Search to support better negotiation outcomes.
Run dedicated, purpose-built workflows for side letter management: With Ontra, your team can track, assign, and report on obligations through workflows built for collaboration across legal and compliance. A digital MFN election process streamlines investor elections and keeps every resulting obligation on record.
Answer compliance questions accurately and quickly: Export execution data to demonstrate compliance to investors or during an SEC exam. Get trusted, cited answers on provisions, obligation data, or tasks from wherever you already work, whether that’s Ontra’s AI Search or MCP in Claude or ChatGPT.
AI works best when the architecture underneath is designed for the specific outcome. Any model can surface a clause, but turning that clause into a trusted record requires continuous verification against a human standard, workflows built for legal and compliance rather than retrofitted from elsewhere, and an answer that traces back to who confirmed it and when.