AI's Best Customer May Still Be Hiring
Rising payrolls can coexist with value adding AI. The commercial question remains: what can customers do with the capacity they buy?
The best customer for an AI productivity tool may very well be a business that keeps hiring. On one hand, a growing team has somewhere to put the time it saves. On the other hand, a team with too little work may struggle to justify the same subscription, however much its employees like the software. Therefore, I would be careful about judging an AI product by whether its customer's payroll falls.
Zooming in, there is evidence that the extra capacity can be substantial. In Generative AI at Work, Brynjolfsson, Li and Raymond studied an assistant used by more than 5,000 customer-support agents at one company. On average, access to this tool increased issues resolved per hour by 15%. That turned out to be a valuable finding, with a precise meaning: the agents could get through more work. What it does not tell us is what another employer should pay for a similar gain.
Consider a hypothetical asset manager winning new institutional mandates. Each mandate brings new accounts to set up, new records to reconcile and new reports to prepare. Internal operations teams are stretched. Software that takes useful work off that team could let the firm accept other mandates without first making every planned hire. To be clear, the firm might still recruit, because business is growing faster than the software expands its capacity. From the vendor's perspective, that can be an excellent account.
Now put the same tool in an otherwise comparable firm that has stopped winning business. The operations team has spare capacity, and management intends to keep it intact. Faster reporting might make the working day better. There may be a sound case for buying on that basis. But multiplying hours saved by salaries produces a cash benefit this customer likely has no plan to collect. The vendor can improve the calculation without getting any closer to a sale.
These prospects could look identical in a sales database: same industry, similar assets, same reporting systems. The difference that matters is the decision in front of each manager. One manager is trying to accommodate new business. The other manager is being asked to buy more capacity than it needs. A founder who understands that difference can qualify an account before spending months on a technically successful pilot.
For the growing firm, I would want to know how much extra work the existing team actually absorbs, with checking and implementation effort included. Does staffing have to rise as quickly with the next mandate? Does service hold up? More clients alone prove very little, since the firm might have served them anyway. The useful evidence is a more credible comparison with how it would have staffed and delivered the work without the product.
This can affect pricing as well. If the product helps a manager take on business that would otherwise have to wait, a price anchored to clerical wages may miss much of its value. That still gives the vendor no automatic claim on the client's revenue. Other products, additional staff and the option to wait constrain what the customer will pay. But it is a more useful negotiation than arguing about the monetary value of an hour nobody intends to redeploy.
Cost reduction remains a legitimate business case. So do better controls, fewer mistakes and relief from miserable work. The point is to establish which benefit is funding the purchase. I would also separate a customer clearing an inherited backlog from one facing a continuing flow of new business. Both customers may sign today. Only the second has an obvious recurring need for the added capacity; the first needs a reason to renew once the backlog is gone.
Before underwriting a larger salesforce, I would go back through the vendor's recent wins. Establish what each purchase allowed the customer to do, then look for the same decision among the next prospects. That is a much stronger basis for expansion than similar company sizes and job titles. A growing payroll need not trouble an investor if the product helps the customer grow without staffing every new piece of work as it did before.
Source: Generative AI at Work, The Quarterly Journal of Economics, May 2025. Its findings concern one company's customer-support agents. The asset-manager comparison is hypothetical; the commercial conclusions are the author's analysis.