Deep|META: Could Muse be the Inflection Point Market Looking for?

Our view on Meta has shifted several times since we initiated coverage. We were bullish throughout 2024 and for most of 2025.

1. Our initial optimism centered on Meta’s Conversions API (CAPI) and the Advantage+ suite built around it. Together, they materially improved conversion performance and campaign automation.
2. The second catalyst was Meta’s generative recommendation system. We saw the value of scaling laws in recommendations and discussed both the mechanics and the ROI framework early on in our earnings previews. We later published a dedicated report on Meta’s progress
3. The developments that followed were more incremental. For example:
   1. We remained firmly bullish in 1Q25, arguing that FX tailwinds and a much smaller-than-feared dutiable value for containerized imports could put Meta’s 2Q25 guidance well ahead of consensus.
   2. We also saw incremental product improvements. In 2Q25 and 3Q25, overlay ads contributed meaningfully by rapidly expanding Reels ad inventory.
   3. WhatsApp Updates and Business AI, however, failed to deliver the progress we had expected.
4. We turned more cautious in 3Q25 with our widely circulated report, Preview | META 3Q25: What Happens When META Doubles Down on AI Compute. From that point, we repeatedly highlighted the pressure that CapEx would put on margins and looked for a clearer framework to assess the returns on Meta’s investment.
5. By 2Q26, we saw some potential catalysts for a recovery, but none was likely to materialize quickly. Our 2Q26 Preview examined why Meta’s deal with Anthropic might take time to generate revenue and why investors would need to be patient.

Muse is the first development that has made us think a real inflection may be possible. We had a similar reaction a year and a half ago when we wrote our GOOG Narrative Shift report. AI Overviews subsequently contributed incremental growth, while GCP accelerated.

This report examines what Muse could mean for Meta.

FUNDA will also publish weekly Muse and Instinct use-case updates on Substack and provide institutional clients with detailed agency channel checks to track whether Muse is beginning to affect ad budgets.

This report covers

1. Muse use cases by category, with a focus on those that offer the greatest monetization potential for Meta.
2. How these use cases, informed by our expert interview, could affect Meta’s ad revenue across industries and narrow the gap with Google.
3. What Muse could mean for the monetization of WhatsApp and Business Agents.
4. An early update on 3Q26 performance based on our latest agency checks.

1. How People Are Using Muse

Twelve days after launch, the task users most want Muse to handle has not changed: arguing with companies on their behalf. Saving money, securing refunds and managing personal finances remain the largest category in our database, with 128 of 654 cases, or 19.6%. Phone calls appeared in just 26 cases in our first cut four days ago; that number has more than doubled to 58. The most-viewed post is now a phone-call story: a user had Muse negotiate with Comcast and streamline a mobile plan, saving more than $800 a year. The post drew 230,000 views.

Muse Use Cases by Category

The six largest primary categories are saving money, refunds and personal finance, with 128 cases (19.6%); work and small business, 73 (11.2%); administrative tasks, 69 (10.6%); shopping, 62 (9.5%); technology and agent setup, 54 (8.3%); and travel, 53 (8.1%). The remaining cases cover personal organization (34), health (29), food and dining (29), family (28), entertainment (22), content creation (22), research (15), social activities and home tasks (13 each), and other uses (10).

More savings cases involve negotiating lower bills by phone.

1. One user had Muse call AT&T and replace an $85, 500 Mbps plan with a $40, 1 Gbps plan.
2. One user cut a Verizon Fios bill from $134 to $64. Another had spent 2.5 hours waiting for AT&T support without success; Muse called on two lines and reached a person in 10 minutes.
3. A Giants fan had Muse activate and cancel streaming subscriptions each month around the NFL schedule, cutting the season’s cost from $310 to $181.
4. Of the cases in our database, 49 provide enough detail to calculate savings: $35,426 annualized in total, or roughly $723 per case, with a median of $490. Another 14 cases recovered $17,652 in unclaimed property, unpaid wages and old escrow balances; the largest recovery was $4,700. Five refund cases totaled $1,264. Combined, these three groups amount to roughly $54,000. These are all self-reported figures that we cannot independently verify, but their scale helps explain why users are so eager to share these experiences.

Administrative tasks still focus on clearing backlogs, now on a larger scale.

1. One user had Muse clear 165,000 unread messages from two old Yahoo accounts and prepare a migration checklist, using just 5% of the weekly allowance.
2. Another cleared 36,939 unread Gmail messages while also having Muse call to book a restaurant.
3. Government paperwork is a newer use case: one user sent a photo of weeds, and Muse filed a 311 complaint and obtained a confirmation number. Another had Muse retrieve receipts from Gmail and submit a dependent-care FSA reimbursement claim from start to finish. An attempt to update passport information for an Overseas Citizen of India (OCI) record on an Indian government website, however, ended after seven minutes of errors. That post drew 27,000 views.

Work and small business became the second-largest category.

1. A user with no prior experience rebuilt a church website in a day.
2. After an ad was rejected, Muse identified the reason, revised it to comply with the rules, then entered the ad account, cloned the ad and resubmitted it.
3. Another user researched 50 prospective customers and drafted outreach emails in three minutes.
4. A pastor had Muse transcribe 50 sermons and compare communication techniques.
5. Others used bank statements to draft federal and state tax returns for a C corporation.

Shopping slipped to fourth place.

1. One user had Muse research and buy a $10,000 airline ticket and a suitcase costing several hundred dollars, without dealing directly with the merchants. Another searched nearby Facebook Marketplace listings for baby winter clothes and only needed to approve messages to sellers; that post drew 20,000 views.
2. One user sent Muse to call Best Buy about Pixel availability, only to find that the other end was also staffed by AI. Among the 53 travel cases, two stand out as tests of trust:
3. One user had Muse book a flight for a wedding while “watching the browser the entire time” (26,000 views).
4. One user watched Muse book a New York hotel in real time and compared the experience to “using the internet for the first time in 1998” (36,000 views).

Technology and setup cases rose to 54.

1. One user connected Robinhood through Plaid and automatically logged new transactions in a budget spreadsheet (32,000 views).
2. Others used Muse on Mac for dictation in place of Wispr Flow.

The most widely viewed of the 29 health cases involved Muse finding a free coronary artery calcium (CAC) scan offered as part of a research study, completing the forms and booking appointments for a couple (73,000 views).

Taking Everyday Tasks Off Users’ Hands

Grouping savings, administrative tasks, travel, shopping, dining, family, health and home tasks as “life admin” gives 411 cases, or 62.8% of the sample. Work and business account for 11.2%; technology and setup, 8.3%; personal organization and social activities, 7.2%; content and research, 5.7%; and entertainment, 3.4%.

Life admin’s share fell roughly two percentage points from 64.9% a week earlier, with work and business gaining 1.1 points and technology and setup gaining 0.8 points. This reflects a broader user base rather than fewer chores. More freelancers, small-business owners and developers arrived in the second week, bringing more productivity-oriented tasks. Still, more than 60% of cases involve things people need to do but would rather avoid: spending an hour with customer support, filling out a government form, comparing three insurance policies or calling five stores. These tasks require little intellectual effort but involve considerable friction, and putting them off has a real cost. When someone says, “I put this off for three months and Muse did it in ten minutes,” the appeal is the relief of finally getting it done.

Mostly Positive Feedback, with Some Early Setbacks

Of the 654 cases, 27 failed, 17 were partially successful and one claimed success that appeared questionable. Together, these 45 cases represent 6.9% of the sample, versus 7.3% in the first cut. Another 48 were still in progress or had no clear outcome. Several recurring problems deserve attention.

1. Merchants refusing to deal with AI: AT&T declined to negotiate with an AI agent after Muse identified itself. A cake shop refused an order placed by an AI caller, although rival agent Mae completed the task. A real estate agent also declined to speak with AI. On the same day, one user saved money through a Muse call to AT&T while another was turned away. The outcome depended on who answered the phone.
2. Phone calls remain challenging: A request to call a user’s brother was rejected because calls to private individuals were not allowed. Five attempts to leave a voicemail also failed.
3. Fabrication: One user asked Muse for auto insurance quotes and found that every quote was made up. Another user said Muse listened to a call with Geico without being asked and fabricated a detail about the situation.
4. Blocked actions: Nike and Dick’s rejected Link payments, forcing users to enter card details manually. A registration process on an overseas T-shirt site stalled at a CAPTCHA. Muse on Mac could not write to Calendar or Reminders despite having permission.
5. It still cannot publish a Facebook post, even on Meta’s own platform.
6. Our failure-mode breakdown includes five cases of website blocking, four incorrect actions, four instances of fabricated data, two slow responses, one false claim of completion, one stalled task and 24 other cases. Fabrication and false completion claims together rose to five cases from three in the first cut.

Trust and data access remain the real hurdles, with privacy becoming more prominent in week two.

1. In 58.7% of the 654 cases, users explicitly granted access to some form of personal data; 23.7% explicitly required none.
2. By data type, 94 cases involved payment cards (14.4%), 92 email (14.1%), 70 social accounts (10.7%), 51 calendars, 47 files or photos, 33 identity documents (5.0%), 25 health information, 23 bank or investment accounts (3.5%), and 23 business accounts. Payment cards or bank accounts featured in 114 cases, or 17.4%.
3. On approval requirements, 407 cases (62.2%) did not identify any user-set checkpoints; only 51 (7.8%) explicitly mentioned steps such as confirming before payment or sending a message. Thirteen required a verification code or manual intervention. Posts about watching the browser throughout a task attract tens of thousands of views because many users are still deciding how much control to give up.
4. Two new privacy complaints also merit attention. One user reported targeted insurance ads immediately after renewing a policy, followed by wealth-management ads tailored to net worth after connecting financial accounts. A Threads post with 1,100 likes alleged that Muse on Mac read system notifications, including iMessage, without notice and used them in conversation. That post did not enter our candidate pool because it lacked search terms such as “asked Muse” or “used Muse,” but it is worth tracking separately.

2. Muse’s Commercial Potential in Practice