Conversational Context Is the Missing Layer of the Agentic Enterprise
Your best rep closed a deal last quarter by remembering a comment a customer made two months earlier about a renewal deadline nobody else had caught. Where did that line live? Not in a field. Not in a report. It lived in an old email thread, a disconnected call transcript, or a Slack message everyone else overlooked.
Ask yourself how much of what actually happens with a customer makes it into your CRM. Not the stage change or the close date, but the actual conversation: what was said on the call, what your team said about the account in Slack, the offhand comment in an email that explains why a deal really slipped. For most organizations, the honest answer is not much. The system of record has never been where the conversation happened.
That gap was tolerable when a CRM record existed mainly for pipeline reporting. It is not tolerable now. Salesforce has been explicit about what it takes to ground an AI agent that reasons on your behalf: conversation data (calls, emails, meetings), customer data (unified sales, service, marketing, and commerce), and external, unstructured data pulled in through Data 360. Two of those three inputs are conversational. An agent that can see your opportunity stages but not the conversation that produced them is reasoning with one eye closed.
What is the conversational layer of the customer data model?
Call it the conversational layer of the customer data model: the words exchanged with a customer and about a customer, external and internal, spoken and typed. Every enterprise already has a structured layer (accounts, opportunities, cases), and many have a connected behavioral layer (web engagement, product usage). The conversational layer is the one most organizations still capture by accident, if they capture it at all.
Salesforce, with Momentum and a connected Slack workspace, now captures and connects the conversational context AI needs. As a Sales Cloud Partner Advisory Board member, I’ve watched Salesforce build this out from the inside. Three capabilities do the work, and they are meant to function as one system, not three separate features.
Einstein Activity Capture: the record you didn’t have to ask for
Einstein Activity Capture connects a rep’s email and calendar to Salesforce and automatically logs emails and meetings to the right record. Every external exchange (the negotiation over email, the reschedule, the “one more thing” follow-up) becomes part of the account’s history without anyone stopping to log it by hand.
Conversation Intelligence, powered by Momentum: the words themselves
What used to be called Einstein Conversation Insights is now marketed as Conversation Intelligence, running on technology from Momentum, the company Salesforce acquired in March 2026. It’s included in every Sales Cloud edition today, not sold as an add-on. It captures calls, meetings, and emails and turns them into structured signals: call summaries, sentiment and competitor mentions, coaching clips, and Call Explorer, which lets a manager search across every recorded call in plain language. This is the step that matters most for agentic AI: it turns something an agent cannot reason over (a recording) into something it can (a field, a tag, a searchable transcript).
Salesforce channels and Slack CRM: the conversation about the customer, not just with them
This is the piece most organizations miss entirely. The conversation your team has about a customer, in a Slack channel off to the side, is often more honest and more useful than the conversation with the customer. Salesforce channels bind a Slack channel directly to a CRM record, so the team can update the record from the conversation while existing Salesforce permissions still apply. That internal thread and the account it’s about are no longer two disconnected systems.
Conversation is data, and it belongs inside the CRM
The organizations that get ahead here will not be the ones with the most Salesforce licenses. They’ll be the ones that stop treating conversation as something that happens near the CRM and start treating it as data that belongs inside it: governed, connected, and ready for an agent to reason over. Everyone else will keep asking their AI agents to make judgment calls with half the conversation missing.
