How much incremental value do subscription-based generative artificial intelligence models have to provide to be viewed as reasonable by business users? In other words, if an Office 365 subscription costs X, is an Office 365 Copilot subscription worth 2X, and if so, for what percentage of users at a firm?
In many cases, the value assessment will come in the form of estimated “time saved” metrics, which will vary based on job roles. One study conducted for the Federal Reserve Bank of St. Lous suggests that “among workers who used generative AI in the previous week (21.8 percent of all workers), between six percent and 24.9 percent of all work hours were assisted by generative AI, for example.
But usage varies by role. “Among all workers, including those who used it only in the previous month and non-generative AI users, we found that between 1.3 percent and 5.4 percent of total work hours were assisted by generative AI,” the study authors note.
Keep in mind those are end user estimates, with the imprecision that likely includes. But it might be reasonable to note that, at this time, perhaps only 20 percent of a firm’s entire workforce might actually be routinely using generative AI, for example. And those use cases might represent less than five percent of total work hours.
There are some use cases where value is easier to grasp. Customer support agents might save 19.7 hours monthly with a 14 percent productivity boost, while programmers could save 44.8 hours with AI coding tools cutting time by 56 percent for half their tasks. The value added is calculated as time saved multiplied by the user's hourly rate (e.g., $20–$100/hour), according one McKinsey estimate.
Much hinges on the assumed hourly labor rates. For example, we might assume $20 for customer support, $50 for general professionals, $100 for high-skilled roles.
Perhaps the business case is easiest for roles including customer support and coding. It might not be so clear for many other roles. If “time saved” is usefully captured, customer service and coding use cases might justify significant per-user monthly subscription fees.
Application/Use Case | Estimated Time Savings per Month (hours) | Assumed Hourly Rate ($/hour) | Value Added per Month ($) | Per-Seat Cost Range ($ per Month) |
Customer Support | 19.68 | 20 | 393.6 | Up to 394 |
Programming | 44.8 | 50 | 2,240 | Up to 2,240 |
General Professional | 9 | 50 | 450 | Up to 450 |
Of course, you know the drill. As much as proponents and suppliers use such metrics, few customers actually believe the claims.
if a feature costs $30 per month and saves nine hours monthly for a user earning $50 per hour, the value added is $450, making the cost reasonable, with the unstated assumption that the saved time is put to some other productive use. If not, the “savings” might be questionable.
It’s sort of the same exercise we might make when looking at work-from-home productivity. Assume WFH leads to a given worker’s ability to complete the standard “in office” work load in half the time. The firm gains if that time, or some of it, is redeployed for other outcome-producing activities. There actually is no firm gain if the employee simply uses the free time for non-work activities.
A Thomson Reuters report suggests AI could save a professional four hours a week now, and perhaps up to to 12 hours per week within five years. But it matters where those time savings are used.
Consumer users might have a harder time justifying a subscription fee for AI-enabled apps. Few of us would claim language model features increasingly available to work with any existing major platform provide some value, some of the time, whether that is search, customer relationship management, e-commerce, communications, social media or productivity suites.
For products based on advertising, transaction or pay-per-use models, perhaps the incremental value can be relatively low, so long as the incremental cost (time, attention, clutter or out-of-pocket fees) are low enough.
That probably is not true for subscription revenue models, though. And that might be a growing issue for subscription-based products where the AI features are offered as an incremental “premium” price to existing subscription products.
That might be a key issue for some products including Office 365 or other subscription-based products whenever the incremental value of the AI add-on effectively doubles the price “per seat” or per user, since many of us would not see the incremental value of the integrated AI as 2X.
There is value, to be sure. It is often helpful to have the AI summarize and “take notes” of a videoconference; summarize key points of a document; draft email responses or generate graphics from a spreadsheet. Other functions, such as creating presentations, might yet leave much to be desired.
But the point is the value-cost evaluation. How valuable are the capabilities; how often are they used and and how do those outcomes compare with the cost of having them, at this point in time? Which workers actually benefit most, and which benefit rarely?
At least so far, reasonable people might agree that, generally speaking, the value of embedded AI features often is not 2X. But is the reasonable business case 0.2X or 0.1X or some other percentage in some cases, but 0.5X in some cases?
And whatever value estimation we might make at this point, will perceptions change in the future if more-compelling capabilities are added?
