Overview
AI is transforming M&A, including the “knowledge” buyers and sellers can have of the target. This may require revisiting M&A risk allocation mechanisms based on knowledge. It raises the question, does an issue identified by AI become the “knowledge” of the party using the AI?
For sellers, a key issue is how the seller’s “actual” or “constructive” knowledge is defined for the purpose of the knowledge qualifiers in their representations and warranties (R&Ws). For buyers, a key issue is how AI in due diligence (DD) could impact sandbagging, e.g., by increasing the value of a “pro-sandbagging” clause.
The follow-on question in both cases is whether the conventional drafting of these definitions and clauses should evolve. Questions around how AI “knowledge” could impact representation and warranty insurance (RWI) are also raised.
Context will matter. As our Exit Insights: A Market Apart deal study shows, market trends around knowledge definitions and “sandbagging” differ from VC-backed and high-growth tech deals versus M&A generally. Other factors such as governing law could also come into play.
For our thoughts on AI’s impact on DD questions and answers in M&A, see here. For our thoughts on AI’s impact on RWI in M&A, see here. For more Fasken M&A thought leadership, visit our Capital Markets and M&A Knowledge Centre and subscribe.
The Sell-Side: AI and Knowledge Definitions
AI tools in M&A aren’t limited to buy-side DD. More and more sellers are using AI in, among other things, answering DD questions and drafting disclosure schedules. Along the way, the seller’s AI may uncover potential issues of which the seller is not yet aware. These issues may not come to the attention of a human reviewer. Even if they do, they may not be escalated internally. In other words, they may not come to the attention of the seller representative(s) whose “knowledge” informs the seller’s “knowledge” definition and qualifiers.
Knowledge definitions in M&A come in two main forms, “actual” knowledge and “constructive” knowledge. The former means the actual knowledge of the specified person(s) without any further investigation by them. The latter includes both the actual knowledge of those persons as well as what those individuals would know after further inquiry. The key question that follows is whether matters flagged (or otherwise identified) by the seller’s AI become the “knowledge” of the seller under these definitions.
Under a simple “actual” knowledge definition, the answer appears to be no. If the named person(s) didn’t actually become aware of the issue, it was outside their knowledge.
The situation is less clear under a constructive knowledge definition. This formulation aims to capture knowledge that was accessible to the identified person(s) even if it was outside their actual knowledge. It’s therefore arguable that, if the issue was discovered by the seller’s AI, then that issue was discoverable by the person.
The critical question from a drafting perspective becomes the exact degree of inquiry (e.g., “reasonable inquiry”) the constructive knowledge definition applies. Several related practical considerations follow. One is AI’s tendency to misidentify, overexaggerate or overproduce diligence results. Another is the nature of the target. For example, as our Exit Insights: A Market Apart deal study shows, constructive knowledge definitions are markedly less common in VC-backed and high-growth tech deals (37.5%) than in M&A deals generally (80%). The considerations which inform this divergence could in turn inform the different possible approaches to a constructive knowledge definition tailored to AI “knowledge”.
The Buy-Side: AI and Sandbagging Clauses
AI is fundamentally changing buy-side DD in M&A. AI is taking over first level review. Human assessment is shifting to second level verification and analysis. Deals teams are operating under ever-more compressed timelines and heightened expectations.
“Sandbagging” refers to a buyer alleging a breach of a seller R&W where the buyer arguably had prior knowledge of the issue. The possibility of sandbagging presents M&A parties with three options. They can include a “pro-sandbagging” clause that expressly permits it. They can include an “anti-sandbagging” clause that expressly prohibits it. Alternatively, they can forego any sandbagging clause and remain silent on the issue.
Historically, a solid majority of M&A agreements in the U.S. and Canada have taken the third approach and gone silent on sandbagging. A chief explanation is that sandbagging is controversial. Buyers argue that they’re entitled to act on the R&Ws they’ve bargained for. Sellers argue that sandbagging can be abused to effectively allow the buyer to re-trade on the deal. Neither party is eager to add such a sensitive issue to their negotiations.
But the interaction of AI and sandbagging raises complications similar to those raised by knowledge definitions. It begs the question: should sandbagging clauses evolve to account for DD by AI? For “pro-sandbagging” clauses, this would presumably involve clarifying that the buyer’s right to claim on the seller’s R&Ws won’t be impaired by any buy-side knowledge, whether human or AI. An anti-sandbagging clause could do the opposite and say any type of buy-side knowledge, human or AI, would trip the clause.
Overall, DD by AI doesn’t fundamentally change the sandbagging debate. It may, however, drive more M&A parties to address sandbagging head-on so they can address the ambiguities that DD by AI can raise.
Once again, context will matter.
Take governing law. The Delaware courts have recently made clear that, in the face of contractual silence, the default is a pro-sandbagging approach. In common law Canada, by contrast, the situation is far cloudier. Under the civil law of Quebec and its obligation of good faith in the negotiation of contracts, a buyer that is arguably attempting to sandbag could face a steep climb.
Regarding the nature of the target, we again see different market trends between VC-backed and high-growth tech deals versus M&A deals generally. As we discuss in our Exit Insights: A Market Apart deal study, while pro-sandbagging clauses only appear in 10% of M&A deals generally, this figure doubles to 22.5% of VC-backed and high-growth tech deals.
AI and RWI: Added Complexity in AI vs Human Knowledge
RWI is a fixture of the M&A landscape that many buyers and sellers rely on to grease the wheels of their transactions and get deals done. AI’s integration into dealmaking adds complexity to this picture, including as relates to the parties’ knowledge.
The seller relies on AI in preparing its answers to the buyer’s DD questions. The buyer relies on DD by AI and the seller’s DD responses in making its no claims declaration. The buyer’s RWI policy excludes breaches that the buyer had knowledge of. The definitions of knowledge across the purchase agreement and the RWI policy may not perfectly align, including as relates to AI-identified issues. A problematic issue may have been identified by AI, ranging from a “smoking gun” to merely the first in a long line of breadcrumbs. Any number of different levels of issue escalation within the buyer may or may not have occurred.
Simply put, should a seller representation in an RWI deal turn out to be wrong, it’s not hard to see how things could get complicated.
Closing Thoughts: Negotiating and Drafting Around AI Knowledge
Knowledge, in its conventional, human sense, has historically driven dealmaking. It has also driven the traditional approaches to knowledge definitions and sandbagging clauses.
AI changes this equation. On the one hand, AI enables accelerated and more detailed review of the target’s business, documents and information by both buyers and sellers. On the other hand, it inserts a non-human intermediary between that information and the M&A parties. It can also generate an excessive amount of information that becomes counterproductive to efficient negotiations and dealmaking. This highlights the value that seasoned dealmakers bring to the table, e.g., in their ability to separate signal from noise. It also highlights that, depending on context, traditional M&A risk allocation mechanisms based on human knowledge, as well as the negotiation and drafting strategies behind them, may be due for a rethink.
