Salesforce agrees to buy Braze, its third marketing deal in three weeks

1. Salesforce agrees to buy Braze, its third marketing deal in three weeks

What happened. On Oct 20, Salesforce signed a definitive agreement to acquire Braze for about $5.66 billion in enterprise value, or $47.50 a share in cash, with closing expected in Salesforce's fiscal 2028 pending Braze shareholder and regulatory approval. Braze runs real-time customer engagement: its event-streaming architecture updates customer profiles and adjusts journeys as people browse, buy and use a product, across in-app, web, email, SMS and WhatsApp, for roughly 2,770 customers. Salesforce says Braze will strengthen Marketing Cloud's cross-channel personalization and connect to Data 360 and Agentforce; Braze CEO Bill Magnuson says the company will keep building as Braze.

It's the third deal in about three weeks. The Oct 8 issue covered Listen Labs (Sept 29), whose agents run customer research and simulate how customers might respond to a message before launch, along with Braze's four AI releases from its Sept 29 event. On Oct 12, Salesforce agreed to buy Loops, which explains why a business metric moved by ranking the likely causes, and is headed for Tableau. None of the three has closed.

What it means for a GTM team. Read together, the deals cover a message before launch (Listen Labs), at the moment it's sent (Braze) and after the numbers come in (Loops). Salesforce hasn't described them as one workflow, and Braze's release frames its strength as high-volume B2C engagement, so B2B teams should expect the fit to vary; Loops itself says it usually needs around 3,000 monthly users. Teams on Marketing Cloud or Tableau may want to ask their account team what changes before and after these acquisitions close.

How it connects. The rest of this week is about the inputs any feedback loop depends on: what got said, who is actually reading, and what a model keeps.

2. ZoomInfo folds Workbounce into GTM.AI

What happened. On Oct 14, ZoomInfo unveiled Workbounce inside GTM.AI, its MCP layer for agents. Workbounce, a London-founded sales knowledge platform whose controlling stake ZoomInfo's UK subsidiary took in November 2024, feeds call transcripts, decks, account notes and Slack threads into ZoomInfo's GTM Context Graph, with lineage, confidence and permission controls on each input.

What it means for a GTM team. Agents working from ZoomInfo can now draw on what sellers said in calls and Slack as well as on the positioning a team wrote down. It will be interesting to see how what reps say in calls compares over time with the approved messaging and positioning.

3. Agent traffic gets a number

What happened. DataDome's Q3 AI Traffic Report, published around Oct 15, puts AI agent traffic up 41.5% from Q2 to nearly 25 billion requests, and agentic browser requests up 4.7x in September. Meta's Muse browser sent 29.1 million requests in its first 12 days without identifying itself as an agent. These figures are DataDome's own. DataDome recommends validating agent identity rather than relying on traffic source, and deciding deliberately which agents to allow.

What it means for a GTM team. Anyone judging page or message performance on web traffic is now measuring some software along with people. Whoever runs analytics or bot management can say what share of visits to pricing and product pages comes from agents, and whether anything identifies them.

4. Meta releases Muse Wonder, with an emphasis on memory

What happened. On Oct 12, Meta released Muse Wonder through the Meta Model API at $4 per million input tokens and $20 per million output tokens, which Meta says is roughly on par with GPT-6 Sol or Opus 5.5. Meta emphasizes the model's ability to retain what users provide and says it beats named rivals on MemoryBench, its own benchmark.

What it means for a GTM team. With pricing roughly level across the top models (Meta's comparison), persistent memory becomes a bigger part of the choice, and whatever a model is told about your company early may shape its answers later.

How it connects. Testing, field capture, traffic data and memory all feed the same question: what actually happened after a team said something.

My take

In a LinkedIn post this week, I asked whether we're heading toward writing hypotheses instead of campaigns, where by Friday the system tells us which ones survived contact with real customers. After a week of reading about these deals, I'm still asking it, with one condition attached: a loop like that can only grade a bet somebody actually made.

Most of us were trained to launch a message, make a sale or push a campaign with conviction. A closed loop rewards something different: launching with a prediction, being fine when it loses, and pivoting fast. IMO this might be part of the answer to moving faster when the pressure on GTM to speed up keeps growing. Does the scarce skill then become the sharp angle, the hypothesis? Something like "support leaders at mid-size companies will respond to the cost story faster than the speed story, and we'll see it in demo requests within two weeks."

The DataDome numbers throw in an honest wrench. "Real customers" now includes agents that don't announce themselves as agents. So part of every bet now has to cover how it might perform with both agentic and human audiences. If we really are moving from campaigns to hypotheses, then knowing who (or what) is on the other end of the test becomes as much a part of the bet as the angle we're testing.