The debate about AI's return on investment swings between two extremes: it will replace everyone, or it is a bubble that delivers nothing. The evidence from the first wave of enterprise deployments is more useful than either. In specific, repetitive, high-volume tasks the productivity gains are real, measurable, and large—while the cost of delivering them keeps falling.
The gains are real—where the work is repetitive
The most credible evidence comes from controlled settings. A large field study of customer-support agents found that access to an AI assistant raised productivity—issues resolved per hour—by about 14% on average, and by roughly a third for novice and lower-skilled workers, while barely moving the most experienced. AI acts as a leveler, compressing the gap between new and expert staff.
In software engineering, developers given an AI pair-programmer completed a standardized task around 55% faster than those without. These are not speculative figures; they are measured differences on defined work.
What it looks like at company scale
The clearest public example comes from consumer finance. One company's AI customer-service assistant, within its first month, was handling about two-thirds of all service chats—work the company equated to several hundred full-time agents—while cutting average resolution time from roughly eleven minutes to two and reducing repeat inquiries.
The nuance matters as much as the headline: the same company later rebalanced, rehiring for premium human support. The lesson is not "replace people" but "reallocate them"—let AI absorb the repetitive volume and move human effort to the judgment- and relationship-heavy work where it earns its cost.
Falling cost is the quiet story
The second force is price. The cost of a given level of AI capability has been falling rapidly as models get cheaper to run and smaller models catch up to last year's frontier. Falling unit cost widens the set of tasks where automation clears the return threshold—work that was uneconomic to automate at last year's prices becomes obvious this year.
That is why efficiency programs should be designed as portfolios, not one-off bets. As inference costs drop, the boundary of "worth automating" keeps moving, and the organizations with governance and data plumbing already in place capture each new tranche first.
Why most programs still disappoint
If the gains are real, why do so many AI initiatives underwhelm? Usually because value is spread thin: dozens of disconnected pilots, none production-grade, governed, or measured. Analysis of where generative AI creates value keeps landing on the same few functions—customer operations, marketing and sales, software engineering, and R&D.
The companies that win pick a small number of high-volume use cases, wire in accuracy and oversight, and industrialize them. The efficiency dividend is real; it accrues to operators who treat AI as an operating-model change, not a science project.
Sources
Disclaimer. This report is produced by Dendra Capital for informational purposes only. It reflects views as of the date of publication, draws on sources believed to be reliable but not guaranteed, and is subject to change without notice. It does not constitute investment research, or financial, legal, or tax advice, nor an offer or solicitation to buy or sell any security. Past performance is not indicative of future results.