Harvey and the cost of an AI outcome

Harvey spent part of this year running a business that got worse the more customers loved it. Bloomberg reports that its gross margin fell from about 50% to negative 50% by June, after a March update to its agents sent usage up twentyfold while revenue merely doubled. One of the most valuable legal AI companies in the world was spending roughly a dollar and a half to deliver each dollar of software it sold.

The easy reading is that the frontier models were too expensive, so Harvey built its own. The more useful reading is that the pricing model failed before the technology did. Harvey sold a fixed annual seat while its suppliers charged by the token. That arrangement holds until the product stops being something a person clicks and starts being something that works through a data room overnight. No flat seat survives a cost curve that steepens twentyfold. The margin did not collapse because the models were dear. It collapsed because the revenue model assumed usage that agents had already outgrown.

This is the part of the AI transition I think boards are underestimating. The technology question gets most of the attention, yet Harvey's rescue was a business model decision as much as a technical one: move the high-volume work onto a post-trained open-weight model at roughly a tenth of the cost on Harvey's own numbers, keep the expensive frontier models for the hard reasoning, and treat the routing between them as a margin decision made thousands of times an hour. That is not a model strategy. It is unit economics, redesigned.

Every vendor selling AI into enterprises should be doing this arithmetic now. A revenue model that does not cover both the fixed cost of building the product and the variable cost of how each customer actually uses it is not a revenue model, it is a bet that usage stays polite. Buyers should ask the same question from the other side of the table: if my vendor's margin depends on me using the product less, whose interests is the contract serving?

There is a governance question underneath as well. The cheap model now doing the bulk of Harvey's work carries weights released out of Beijing. Cost, capability and provenance have become one decision, and someone senior needs to own all three.

Source: https://lnkd.in/gia5rc3R

#AIEconomics #BusinessModels #Strategy