Ten million is a setup count

Start with the largest number, because it is also the loosest. Zuckerberg's own words on the call were that, in about a month and a half, more than 10 million people in the US and Canada "have set up Muse personal agents". That is a setup count for the two countries where Muse exists. It is not a monthly-active figure, and Meta did not present it as one. The number that does the engagement work is the second one: about 25% of people who set up Muse use it "on any given day".

Read that 25% for what it is. It is the engagement measure Meta chose to present, not a published retention cohort, and it carries no information about whether the rest have abandoned the product — nothing in the call separates daily users from weekly ones, or either from people who tried Muse once. Zuckerberg noted the daily level is more than twice where Threads started, which is context, not a cohort. The setup figure is also consistent with the estimate Sensor Tower published earlier that week; two reported numbers agreeing is not the same as either counting method being validated. The disclosure that would settle the retention question is whether the people who set Muse up keep returning over months, and Meta has not published it.

The $200 average is narrower than it sounds

The $200 figure will travel furthest, so it deserves the closest reading. Zuckerberg called shopping the most common use case for Muse: more than a quarter of "people who are using Muse" are engaging in it, "with an average purchase size of over $200". A few minutes later, CFO Susan Li described the same behaviour against a different denominator — shopping is used by over 25% of Muse's weekly actives. TechCrunch scoped the number to "engaged" shoppers, a word nobody used on the call. On any of these denominators, the average belongs to the minority who shop through the agent, not to the 10 million who set it up. And it is not a retail figure: asked in the Q&A whether the commerce numbers include travel, Zuckerberg confirmed that they do, calling travel one of Muse's other top use cases. The $200 average is a purchase size across a mixed basket that Meta does not break down.

What Meta did not provide matters as much. There is no transaction count, no total value of purchases made through Muse, and no Muse revenue line anywhere in the call. When an analyst asked for per-user cost metrics, Li declined, saying Meta does not have specific numbers to share around Muse's cost structure today. Without a count or a volume, the $200 average cannot be turned into a sales figure, and any version of this story that multiplies the headline numbers together is inventing a business Meta has not disclosed. What the call establishes is narrower, and still notable: Meta reports that more than a quarter of people using Muse are shopping through the agent — which is not proof that each of those interactions ended in a completed purchase — and that the average ticket, travel included, is above $200.

The trust colour around shopping comes with its attribution chain attached. Zuckerberg relayed that Plaid told Meta it had not seen activity like the Muse connector's on any other service before. That is a Plaid statement presented by Meta's CEO on Meta's earnings call, not an independently obtained Plaid assessment, and neither the comparison's underlying metrics nor a transaction count came with it. He added that first-week volumes were high enough that Meta worked with Stripe to raise credit limits — a demand signal, again without figures. Users are also connecting email and calendars, the two data sources people usually guard most, and Zuckerberg presented the Secure VM as the reason they are willing. Meta describes that VM as providing state-of-the-art privacy and security; that is the company's characterisation, not an independent audit. The stronger protection, a Confidential VM that would make it so even Meta cannot see user content, is forthcoming, not shipped — a promise attached to a product already handling bank-linked purchases.

The forecast does the heavy lifting

The boldest claim of the call was not a result. Zuckerberg expects Muse to reach an even higher revenue per user over time than the family of apps produces, and he clarified on the call that the baseline — several hundred dollars a year — is specifically the family of apps revenue of Meta's existing Muse users, not a company-wide average. No date was attached: "over time", and "this will take some time". No current Muse revenue was given against which to measure the distance.

The mechanism he described is a fee on commerce: Meta will build commerce tools and be paid based on the transactions it helps facilitate, with the fee expected to be paid by businesses — in his Q&A phrase, a model where Meta gets a percent of the value it helps create. On top of that, people who want the most advanced models or extra tokens will subscribe directly, and businesses will be able to subscribe as well. The tiering is already visible: the free tier runs on Muse Spark by default, while Muse Wonder sits in the paid tier.

The cost side explains why the timing matters. Muse is, in Zuckerberg's words, very compute-intensive to run, and the profitability case rests on improving cost efficiency by more than tenfold within a year. Li named the intended drivers: better utilisation of off-peak capacity, more efficient models and inference optimisation. The same efficiency work is the condition attached to the scale claim. Meta says it is building capacity to serve more than a billion people over the next few years, but in the Q&A Zuckerberg tied actually serving a billion-plus people to achieving certain compute efficiency targets. These are plans stated on an earnings call, not results in a release. Meta's Q3 revenue of $64.6 billion, up 26%, gives the company ample room to fund the attempt. It does not make the attempt a business yet.

Read the call that way and the shape is clear. Meta has distribution, a reported fast start, and early signs that a minority of users will spend real money through an agent. What it does not yet have, on its own disclosure, is the unit economics: no transaction count, no repeat-purchase rate, no Muse revenue figure, and no per-user cost it is willing to state even when asked directly. The next useful numbers are not another setup milestone. They are evidence that people keep using Muse, complete transactions through it, and generate revenue at a sustainable cost.

Sources

  1. [1] Meta Platforms, Q3 2026 earnings call, 21 October 2026 — full call transcript via MarketBeat, read directly for this article: Zuckerberg's and CFO Susan Li's remarks on Muse setups, daily use, shopping share and average purchase size, the travel inclusion in Q&A, monetisation, cost efficiency, capacity, and Li declining to state per-user costs. Transcripts of the same call published by AlphaStreet and TradingKey corroborate the wording.Read source
  2. [2] Meta Platforms, “Meta Reports Third Quarter 2026 Results”, official results release via PR Newswire, 21 October 2026 — Q3 revenue of $64.601 billion, up 26% year over yearRead source
  3. [3] TechCrunch, “Meta says Muse has over 10M users, with shopping a top activity”, 21 October 2026 — contemporaneous report of the same call; its scoping of the shopping figure to “engaged” users is its own wording, not the call'sRead source
  4. [4] Engadget, “Mark Zuckerberg says Meta is 'entering a new phase' with Muse”, 21 October 2026 — contemporaneous report of the same call; reported the $200 average without the shopping-subset scope, and Muse on smart glasses next month (November 2026), which the call transcript also statesRead source