The Legal AI 'Basket Calculation' Blind Spot: Why AI Contract Review Tools Flag the Indemnity Basket But Miss Whether It's a Deductible or a Threshold — and Why That Distinction Is Worth Millions

Andy Armstrong | The Legal Stack

Every M&A practitioner has seen the demo. The sales engineer pastes in a purchase agreement, clicks a button, and the AI tool triumphantly highlights the indemnification basket. "Basket identified: $500,000," it announces. The room nods. Someone asks about the cap. The tool finds that too.

What almost no one in the room asks is the question that actually matters: Is that basket a tipping basket or a deductible?

That question — which takes an experienced lawyer approximately three seconds to answer after reading the relevant clause — is one that the current generation of AI contract review tools handles with stunning unreliability. And the economic consequences of getting it wrong are not marginal. They are, in many middle-market deals, the difference between a recoverable indemnification claim and a total loss.

The Mechanic, Explained

The distinction is structurally simple, which makes the AI failure all the more inexcusable.

A deductible basket (sometimes called a "true deductible") works exactly like an insurance deductible: the buyer absorbs losses up to the basket amount, and the seller is only liable for losses exceeding that threshold. If the basket is $500,000 and the buyer suffers $600,000 in qualifying losses, the seller owes $100,000.

A tipping basket (or "first-dollar" basket) works differently: once cumulative losses cross the threshold, the seller's indemnification obligation springs back to dollar one. Same $500,000 basket, same $600,000 in losses — the seller owes the full $600,000.

The operative language is usually a single clause, often a subordinate phrase buried within the indemnification procedures section. A deductible formulation reads something like: "the Indemnifying Party shall only be obligated to indemnify the Indemnified Party for Losses in excess of the Basket Amount." A tipping basket uses language along the lines of: "the Indemnifying Party shall be obligated to indemnify the Indemnified Party for all Losses from the first dollar once the Basket Amount has been exceeded," or simply refers to the basket as a "threshold" rather than a "deductible."

The difference is grammatical. The economic consequence is structural.

Why AI Tools Consistently Miss It

Current AI contract review tools — including the category leaders — are trained and tuned to identify the existence of a basket and extract its dollar amount. These are classification and extraction tasks, and the tools perform them reasonably well. What they fail to do is reliably interpret the operative mechanic governing how that basket functions.

The failure has two root causes. First, the drafting conventions are inconsistent. The same economic concept appears across dozens of formulations — "threshold," "deductible," "first-dollar recovery," "basket amount" — and the tools lack the contextual reasoning to distinguish a threshold that tips from a threshold that merely deducts. Second, the tools are optimized for the questions their users are asking in the demo, not the questions that matter in a deal. "Does a basket exist?" is a yes/no classification. "Is this basket a deductible or a tipping mechanism?" requires semantic interpretation of clause mechanics combined with economic reasoning — a materially harder task.

The result is that most tools return something like: "Indemnification basket: $500,000." Full stop. That is near-worthless information to a practitioner.

The Scenario That Makes This Concrete

Consider a hypothetical mid-market acquisition: purchase price of $40 million, basket of $500,000, cap of $4 million. The buyer discovers post-closing that the target's key enterprise software license — undisclosed in the data room — was non-transferable and has been terminated, triggering a customer contract cascade. Total qualifying losses: $510,000.

Under a deductible basket, the seller's exposure is $10,000. The claim is economically irrational to pursue once you account for the cost of litigation and management time. The buyer eats $500,000 of loss.

Under a tipping basket, the seller owes $510,000 from dollar one. The claim is commercially viable, the indemnification provision does its intended job, and the buyer is made substantially whole.

The AI tool identified the basket. The AI tool told you nothing about which of these outcomes applies. A lawyer who relied on the tool's output without reading the clause has potentially counseled their client into a $500,000 error of omission.

What Buyers' Counsel Must Check

Buyers' counsel reviewing a draft purchase agreement — or conducting post-signing diligence on a target's existing contracts — need to look at four specific things that current tools are not reliably surfacing:

First, the precise operative phrase governing basket mechanics. "In excess of" versus "from the first dollar" or "threshold" language is determinative. Second, whether the agreement defines "Basket Amount" in a definitions section that modifies the apparent clause-level language — a not uncommon drafting trap. Third, whether the basket applies differently to fraud or fundamental representations, which often carry separate first-dollar mechanics regardless of the basket type. Fourth, whether materiality qualifiers in the representations have been "scrubbed" for indemnification purposes — a materiality scrape changes the effective basket exposure in ways that interact directly with the deductible versus tipping distinction.

What Sellers' Counsel Must Check

Sellers' counsel should be equally attentive, for the opposite reasons. A tipping basket in a seller-favorable deal is a negotiating concession that may have been granted on the assumption that it functions as a deductible. If the drafting ambiguity is in the buyer's favor — common in buyer-form agreements — sellers' counsel reviewing at the markup stage must affirmatively confirm the mechanic, not rely on a tool's basket extraction to tell them whether a tipping obligation has crept in.

Sellers should also watch for definitions of "Losses" that include consequential damages or multiplied damages in specific contexts. A tipping basket with an expansive Losses definition compounds the deductible-versus-tipping stakes significantly.

The Tool Is Not the Lawyer

ABA Formal Opinion 512, issued in July 2024, addresses the competence obligations of lawyers who use generative AI, covering both the duty to understand the tool's limitations and the duty of independent verification. The basket mechanic problem is precisely the kind of issue Opinion 512 contemplates: a task where AI output creates a false confidence that the relevant issue has been addressed.

The tools are getting better. Some of the more sophisticated platforms are beginning to surface the operative mechanic language, not just the dollar figure. But better is not the same as reliable, and in M&A indemnification drafting, the gap between "basket identified" and "basket mechanic confirmed" is worth millions.

Read the clause.

Andy Armstrong writes The Legal Stack. Views expressed are his own.