Subscription companies made money on forgotten renewals. Now an AI assistant cancels them for customers

In late September, YouTuber Nate B Jones asked Muse, the personal AI assistant Meta launched in the US on September 8, to track down every subscription Nate was paying for. Muse found subscriptions costing $5,350 a year and canceled $1,285 a year's worth of them. Nate added that the $5,350 is money that would go out in the future, not cash returned, and that not every subscription found was a waste. It's a scene subscription companies don't want to see. Stanford researchers estimate that a large share of those companies' revenue comes from customers like this, who put off canceling. youtube +1

Stanford researchers used card data from 10 big subscription services and estimated that delayed cancellations lift seller revenue by 87% on average

In a September 27 story, CNBC wrote that the subscription business has long leaned on a habit: people are good at signing up and much worse at canceling. The Stanford researchers cited in the story estimated that forgetting or putting off cancellation, plus the hassle of canceling, roughly doubles seller revenue. The estimate comes from “Selling Subscriptions,” a paper by Einav, Klopack, and Mahoney published in the American Economic Review in 2025. Using payment card data, the researchers examined renewals at 10 large subscription services and wrote that inertia in canceling raises seller revenue by 87% on average. The figure ranged from 14% to 230% depending on the service. An 87% increase means 1.87 times, the same result CNBC described as roughly double. cnbc +1

Taking the average at face value: if revenue is 100 when customers cancel as soon as they want to, actual revenue is 187. Of that, 87, or about 46% of total revenue, comes from customers who put off canceling. This division is our own calculation using the 10-service average, and the value varies widely by service.

In months when a new card forced customers to decide whether to keep a subscription, they were about four times as likely to cancel

When a payment card changes, the customer has to decide whether to keep the subscription going on the new card. Comparing renewals in those months with other months, the researchers found that renewal rates dropped sharply in months when the card changed. Mahoney explained the finding to CNBC this way. stanford.edu

“We found that when people are forced to decide, they are about four times more likely to cancel”

Neale Mahoney, CNBC 2026-09-27 cnbc

Mahoney is describing the months when a card change made customers choose whether to keep a subscription.

The researchers also estimated that if customers had to actively choose whether to continue every six months, the extra revenue from inertia would shrink by about half. Subscription revenue piles up as long as the moment of decision doesn't come, and it shrinks when that moment arrives. stanford.edu

Two reporters who tried Muse saw it find subscriptions they had forgotten and either ask whether to cancel or cancel outright. When Kim In-yeop, Silicon Valley correspondent for The Korea Economic Daily, tried Muse, it offered on its own to cancel auto-renewing subscriptions. When Won Ho-seop, Silicon Valley correspondent for Maeil Business Newspaper, connected a Bank of America account, Muse found a forgotten $99 annual Otter subscription and canceled it. The paper, however, came out before Muse and does not cover cases where an AI cancels on someone's behalf. The idea that an AI assistant asking customers about forgotten subscriptions would increase cancellations is our own inference from the paper's numbers. hankyung +1

In one person's collection of public Muse use cases, clawing back money by cutting forgotten subscriptions and lowering bills was the most common single task

Meta said Muse can open a browser, fill out forms, and negotiate on a user's behalf, and its launch post used lowering bills as an example. What people actually handed to Muse can be seen on a site called shipwithmuse. One person with no ties to Meta hand-picks use cases there and lists claimed savings exactly as the poster stated them. As of October 1, its Errands & personal agent category held 206 entries. For this article, we removed 23 posts by Meta employees or official accounts, 26 introductions of tools, guides, or reviews and pre-launch demos, and 2 compilations of secondhand reports where it was unclear who did the task. We then sorted the remaining 155 by task. about.fb +1

Recovering money was the largest single category. Looking only at the X posts among the 155, the 36 money posts had a median of 11 likes, higher than the other categories (1 to 6, excluding Work, which had only one post). This ranking can change depending on how the categories are drawn. Combine booking and travel with scheduling, family, and household admin into a single everyday-errands category, and it reaches 61, more than recovering money. People also tend to post results worth showing off, so the table shows the share of tasks people highlighted in public posts, not everything they gave Muse.

Meta put money up front too. In an interview on launch day, Zuckerberg said Muse would pay for itself by making and saving people money. Zuckerberg said Muse would be free up to generous usage limits and that Meta would eventually make money by taking small fees on transactions. The more payments run through Muse, the more Meta earns in fees. On September 12, Meta employee Matt Schlicht launched #MuseMoneyChallenge, inviting people to make or save money with Muse. Business Insider reported that chief AI officer Alexandr Wang kept the challenge going by asking people to post what they had saved with Muse, and that many of the replies posting savings came from Meta employees. Splitting the 155 cases at the time Schlicht posted the challenge on September 12, money tasks made up 7 of the 39 earlier posts (18%) and 33 of the 116 later ones (28%). The difference is only a matter of timing. This data can't show whether the challenge caused it. youtube +2

Most savings figures are self-reported, and most attached screenshots are reports written by Muse. We found no case confirmed by a bill. The most widely shared figure is Peter Yang's “$800 a year.” In a September 18 video on Peter Yang's own YouTube channel, Peter Yang said Muse had saved more than $800 a year on cable and phone bills. On the AT&T phone bill, Muse used a browser to enroll in an autopay discount, cutting the bill from $200 a month to about $160 and saving about $480 a year. On the Xfinity cable bill, Muse called and got only a discount offer. The representative offered to move the $84 monthly rate to a $60 promotional rate, with another $10 off for autopay, but said Peter Yang had to call in personally to get the discount. The video doesn't say the discount was applied. The only discount that appears to have been applied in the video is AT&T's, worth about $480 a year. Getting to $800 requires adding the Xfinity discount, which was only offered. By our calculation from the rates in the video, the Xfinity discount comes to $288 to $408 a year, and the video never adds the two together. The Next Web wrote that none of the savings claims posted to the challenge came with independently verified before-and-after amounts. youtube +1

Wall Street began to see companies that rely on delayed cancellations, like gym chain Planet Fitness, as companies AI assistants could hurt

On September 29, CNBC reported on a new investment theme called “consumer inertia.” The idea is that agents like Muse could hurt companies that make money because customers don't cancel subscriptions or compare prices. The same story noted that since Muse launched on September 8, shares had fallen 20% at Planet Fitness, 16% at Booking Holdings, 13% at Airbnb, about 11% at SiriusXM, and 9% at Charles Schwab. cnbc

Jim Cramer, who wrote the piece, thought the market may have overreacted. Cramer pointed out that shares of Life Time, an upscale gym chain whose members go more regularly, fell only 7%, compared with 20% at Planet Fitness. We read this gap as investors judging that a gym people actually use would keep many of its members even if an AI asked whether to keep going. But the story includes neither company's cancellation rates nor any data showing that cancellations actually rose. cnbc

Once AI assistants start asking customers whether to keep paying, subscription companies have to protect revenue with reasons to stay, not with the time customers spend putting off canceling

When the moment of decision came, the first thing companies offered was a discount. X user chanduiiit handed Muse an AT&T bill of $80 a month. Muse reported that the retention department offered three months free followed by $40 a month for 24 months, and noted that the next bill should be checked to confirm it was applied. X user InvestKaye was paying Verizon $89.99 a month, a rate that already included a $38 monthly discount. Muse reported that it secured a loyalty discount of another $20 off each month for 12 months, bringing the bill to $69.99 a month. Discounts that never appeared while customers put off deciding showed up at the moment of decision. X +1

The discounts have end dates. chanduiiit's $40 rate lasts 24 months, and InvestKaye's discount lasts 12. When they expire and the price goes back up, an AI assistant can find the same subscription again and ask whether to keep it. A discount stops holding customers once it ends, but a service customers actually use holds them with no end date. If the reading in the previous section is right, a gym like Life Time, where members go more often, is the kind that keeps members even when an AI asks whether to keep going. Nate B Jones also said not every subscription Muse found was a waste. If customers really use a subscription, they may well say yes when the AI asks. natesnewsletter.substack

Until Muse arrived, subscription companies relied on the time customers spent forgetting or putting off cancellation. Now Meta's AI assistant finds forgotten subscriptions for customers, asks whether to keep them, and even cancels them. So what subscription companies can rely on is no longer the time customers put things off, but a service used enough that customers say they'll keep it when the AI asks.

Go deeper

Selling Subscriptions The Stanford researchers' paper, which uses card payment data from 10 large subscription services to estimate how much cancellation inertia boosts revenue. stanford.edu

Muse use case collection A list, hand-picked by one person, of errands people have given Muse. shipwithmuse.live