AI Pulse

Salesforce Just Spent $3.6B on Support AI. Here's What to Do While the Deal Closes.

By Felix Maru · July 20, 2026 · 8 min read

The biggest single check written in support software this year closed when Salesforce agreed to acquire Fin (formerly Intercom) for roughly $3.6 billion. That deal is still pending regulatory approval and is not expected to close until around the end of 2026 at earliest. But between now and that close date, support teams running on Fin have real decisions to make. And Salesforce is not waiting for the ink to dry before reshaping what the support desk looks like. Here is what happened, what it means, and what to actually do about it.

What Salesforce Was Actually Buying

This was not an inbox acquisition. Salesforce was not buying Fin for its chat widget or its email routing. It was buying resolution rate and the model behind it.

Fin's AI Agent, powered by the company's proprietary Apex model, handles complex customer queries end-to-end across live chat, email, WhatsApp, SMS, phone, and Slack. According to Salesforce Ben's coverage, Fin brought roughly 30,000 AI customers to the deal, with reported end-to-end resolution rates averaging around 76% of support volume across those customers. That is the number Salesforce paid for: proven resolution at scale, pre-trained, deployable quickly, without months of configuration.

That context matters because it explains the acquisition logic from a support perspective. Resolution rate is the metric that now determines value in support AI. Not seats, not API calls. How much of the ticket queue does the AI actually close without a human touching it? Salesforce looked at Fin's answer to that question and decided $3.6 billion was the right price to acquire it rather than build toward it. Every major vendor is now competing on that single metric.

The Three Questions Fin Customers Should Answer Now

If your support team runs on Fin today, the platform is not going anywhere before the deal closes. Fin is still operating independently and continues to ship updates. The practical risk is longer-horizon: post-acquisition, Fin customers will eventually be migrated into or alongside Salesforce's broader ecosystem. For organizations outside the Salesforce CRM world, that migration may be significant.

Three questions worth answering now, before the close:

1. What does your CRM stack look like? If your organization runs on Salesforce, this acquisition is a reason to slow down the migration conversation, not accelerate it. You will likely inherit native Agentforce integration once the deal closes. If you are on HubSpot, Pipedrive, or Close, start your due diligence on what a Fin-inside-Salesforce world means for your stack. The integration story will change.

2. How much of your support runs through Fin's AI Agent specifically? Teams that have leaned heavily into Fin's Apex-powered automation are the most exposed to model and feature changes post-acquisition. Document which workflows depend on Apex behavior today so you are not reconstructing that knowledge under time pressure in six months.

3. When does your Fin contract renew? If renewal falls before the deal closes, negotiate appropriate exit provisions. This is standard practice during any M&A window. It is not pessimism about the outcome; it is protecting your optionality.

A $3.6B acquisition does not mean a platform collapses. It often means the opposite. But it does change your dependency calculus, and the time to understand that is now, not after the close.

Agentforce Help Agent: What Pay-Per-Resolution Pricing Actually Means

Separately from the Fin acquisition, Salesforce shipped Agentforce Help Agent to general availability in July 2026, and the pricing model deserves its own read.

The structure: you pay when the AI resolves an issue autonomously. If the customer requests a human agent, or if they disengage without resolution, no charge. Salesforce is tying cost directly to outcome rather than to activity or seat count.

This is a meaningful change from traditional per-seat or per-conversation billing. It shifts the ROI question from "what does this tool cost per month" to "what is my cost per resolved ticket." If you have been struggling to justify AI support spend to finance or ops leadership, this framing is considerably cleaner: the tool's cost scales in proportion to the autonomous work it actually completes.

The practical implication is worth spelling out. With per-resolution pricing, you need accurate, consistent tracking of what "resolved" means in your context. A customer who stops replying is not the same as a customer whose issue was closed. A ticket marked resolved by the AI but reopened within 48 hours is not the same as a genuine close. Getting that definition right in your workflows and tagging logic matters more under this model than it did under flat-rate billing. If your resolution tagging is loose, your costs will be too.

Worth noting: Agentforce charging nothing when a customer escalates to a human is not a flaw. It is a design acknowledgment that human escalation has real value and is a normal, expected part of good support. The human in the loop is not a fallback; it is part of the intended flow. Combined with Zendesk's earlier move to outcome-based billing, per-resolution pricing is now becoming the industry standard for support AI, not a niche experiment.

What Zendesk Shipped in July (the Pieces Worth Knowing)

Zendesk's July 2026 release notes dropped several updates worth knowing about for teams on the platform.

Forethought AI agents are now a purchasable add-on. Zendesk acquired Forethought earlier this year and has been integrating its capabilities into the product. As of July 2026, teams can purchase Forethought AI as a standalone add-on: autonomous resolution, intent identification, agent assistance, and quality review baked into the Zendesk workflow. This gives Zendesk an AI agent layer that can complete work end-to-end, not just surface suggestions. The competitive response to Fin's resolution rate is implicit in the rollout timing.

Voice AI Agents got multilingual voice configuration. Teams can now choose between male and female voices with locale-level overrides, using higher-fidelity models. More practically, Voice AI Agents can now add tags to a call ticket. That sounds minor but is not: tags are how Zendesk drives routing, reporting, and automations. Giving the voice AI the ability to set tags means a voice conversation can now trigger the same downstream workflows that web tickets already trigger. Before this, voice calls were frequently a dead end in the automation chain.

Predictive omnichannel routing is out. AI assigns incoming messaging tickets to the agents predicted to resolve them fastest, tracking a new metric called Agent Engagement Time. If your team handles high volume across chat and email simultaneously, this is worth evaluating before you write it off as incremental. Routing to the predicted-fastest resolver rather than simply the next available agent is a meaningful change in how work gets distributed.

Native contact center telephony is now generally available, fully integrated into Agent Workspace with unified status management. If you have been running a third-party telephony integration stitched to Zendesk, you now have a native path that removes one integration layer from the stack.

If you are on Zendesk Suite Professional or above and have not reviewed the July release notes, set aside 20 minutes this week to do it. There is enough in there to either save budget or improve queue flow, often both.

The Pattern Across All of This

The support software industry has stopped debating whether AI can do the work and started competing on how much of it AI can close, at what cost, and with what human backstop. That is broadly good for support teams. But there is a risk in the race to maximize resolution rates: the tickets that genuinely require human judgment, empathy, and relationship work can start to look like a problem to optimize away rather than the most valuable part of the job.

The teams positioned well are the ones that have already defined clearly where AI closes the loop and where a person must stay in it. Not because AI is unreliable for everything, but because customers know the difference between a fast resolution and a felt one. Keep that handoff clean and both sides of the operation get better.

Sources

Want to talk through what the Salesforce/Fin deal means for your specific support stack? Reach me here and I'm happy to think it through with you.