The scenario
At Falcon Software, sales ops is the human search engine. All day long reps drop into a Slack channel asking "what's the ARR on the Acme deal?", "is the Globex renewal at risk?", "who owns the Initech account now?" — and someone opens Salesforce, looks it up, and pastes the answer back. We'll replace that with an employee-facing Agentforce agent the team can @mention in Slack, grounded on live CRM data so the answers are real, not guessed.
For beginners: an employee agent runs with the user's own context — it's effectively logged in as the person asking, so it respects that user's permissions and record access. Slack is a natural home for it: conversations, apps, and data already live there, and anyone can summon the agent with a simple mention.
Step 1 — Build the employee agent
In Agent Builder, create an employee agent (not a customer-facing one) scoped to sales questions. Its topics should cover the things reps actually ask: deal status, account ownership, open opportunities, recent activity.
- Write instructions that keep it in lane — it answers about accounts, opportunities, and contacts, and declines unrelated requests.
- Because it inherits the user's permissions, a rep only ever sees data they're already allowed to see. Security is not an afterthought here — it's the default.
Employee vs customer agent: a customer service agent runs with a fixed, guest-like identity and grounds on Knowledge. An employee agent runs as the logged-in user and can reason over the CRM records that user can access. Picking the wrong type is the most common early mistake — for internal sales ops, you want employee.
Step 2 — Ground it on CRM data with Data 360
Answers about "the Acme renewal" have to come from your actual records, not the model's imagination. Configure a Data 360 data library so the agent retrieves from live CRM and any unstructured sources you connect.
- Ground on the objects reps ask about — Accounts, Opportunities, Contacts — plus notes or call summaries if you have them.
- Keep retrieval tight and relevant; over-broad grounding makes answers slower and noisier.
- Test with real questions phrased the way reps actually type them in Slack, abbreviations and all.
Step 3 — Deploy to Slack
- Install the Slack platform connector from your Agentforce org (admin installs for the workspace).
- Confirm your My Domain URL is set — the connection relies on it.
- A Slack admin with the right workspace privileges approves and installs the agent.
- Reps
@mention the agent from any channel or DM and ask in plain language.
Common mistake: deploying to the whole company on day one. Start with one sales-ops channel and a handful of power users. Watch what they ask, fix the gaps, then widen access.
Make it stick
- Seed the real questions: pull a week of that Slack channel's history and make sure the agent nails the top twenty.
- Respect security visibly: reassure the team that the agent only sees what they can — because it literally runs as them.
- Measure deflection of human effort: how many "can someone look this up?" messages disappear. That's the time you gave back to sales ops.
- Grow into actions: once answers are trusted, add write actions (log an activity, update a next step) behind confirmation — see the governed-actions build in this lab.
Sources:
Connect an Agent to Slack (Salesforce Help) ·
Trailhead: Deploy an Agent in Slack for Your Team ·
Trailhead: Get Started with Agentforce for Employees