Salesforce Just Bought Your Support Roadmap. Here's What to Do About It.

Two things happened in the customer-service AI world in the span of a few weeks, and together they change how you should think about your support stack. Salesforce agreed to buy Fin, the company formerly known as Intercom, for about $3.6 billion. Around the same time, Fin started selling itself as a service agent that runs on top of HubSpot and Freshworks, two products it competes with. Let me catch you up on what this means for your team.

The big players are buying their way in

The headline number is real. Salesforce signed a deal to acquire Fin for ~$3.6B, with close expected in the fourth quarter of Salesforce's fiscal 2027. Fin's CEO Eoghan McCabe framed it as four young founders pulling off a late-stage pivot to AI and inventing a category. Strip the sentiment and you get a clear signal: a giant decided it was cheaper to buy the agent than build it.

For you, the takeaway is timing. The companies that own customer-service software are now competing on AI agents, and they will pay billions to skip the build. That means the tool you pick today may belong to a different company in two years. Plan for it.

The walls are coming down on purpose

The second move is the more interesting one. Fin announced you can now run it as a service agent on HubSpot and Freshworks without leaving your helpdesk. Live in under an hour, self-serve, a claimed 76% average resolution rate across chat, email, voice, and social. They put it plainly: closed and protectionist vendors "will accelerate their demise."

That is a bet you should pay attention to. The pitch is that swapping agents will get easy, so the best product wins on merit instead of lock-in. If that holds, you are not making a forever decision when you pick an agent. You are renting one, and you can leave. Build your contracts and your data plumbing so that staying is a choice, not a trap.

Buying the agent is the easy part

Here is the catch nobody puts on the billing slide. Most teams that adopt an agent never get the value out of it. Fin's own numbers say 82% of senior leaders invested in AI over the past year, but only 10% reached mature deployment. The gap is integration. An agent that can only answer questions still hands the actual work to your reps.

The proof is in the workflows. When Fin's own support team rebuilt scripted tasks into procedures with real system access, a bounce-list workflow jumped from 9.3% to 79.9% resolution. A bug-report flow went from 9.2% to 66.5%. But a simple messenger install barely moved, from 67% to 69.2%. The lesson: deep integration pays off where work needs judgment and live data, and does almost nothing where it doesn't. Pick your battles.

The part you can't buy off the shelf

An agent ships sounding like a language model, because it is one. Somebody on your team has to teach it to talk like your brand. Fred Walton calls this conversation design, and it covers tone, response length, handoffs, and follow-ups. It is not optional polish. When Fin A/B tested a warm opening message against its old default, CSAT rose from 72.8% to 78.4%. One line of copy.

You do not need to hire a conversation designer on day one. You do need to name an owner. Walton's rule is worth stealing: if it's information, put it in the knowledge base; if it's about tone or handling a situation, put it in the agent's instructions. Teams keep piling on rules for every edge case until the agent has paragraphs to read before it answers. The discipline is knowing when to stop.

The deep cut

The acquisition and the open-platform play look like opposite news, but they point the same direction. The agent itself is becoming a commodity you can swap and even rent on a rival's helpdesk. What you cannot buy off the shelf is the two things that decide whether it works: system access wired into your CRM and billing tools, and conversation design that makes it sound like your team. Those are yours to own. So spend your platform energy on staying portable, and spend your people on the integration and the voice. Even Fin admits voice is the part that has been faked in demos for years, with Fin Voice 2 only now claiming it's ready. Treat any agent's hardest channels as unproven until you test them on your own traffic.

Three questions for your team

1. If our agent vendor got acquired or changed terms tomorrow, how fast could we move our data, our integrations, and our conversation design to another platform? If the answer is "slowly," fix that before you sign a bigger deal.
2. Which single high-volume workflow are we still handing to reps that the agent could resolve end-to-end with read or write access? Scope that one integration now and measure the lift.
3. Who owns how our agent sounds today, by name? If nobody does, pick the person this week and have them fix our worst handoff first.