Use case

Sales operations agents

Sales ops teams lose hours every week to manual Salesforce upkeep, lead research and pipeline reviews. A Dezifi sales ops agent turns each of those into a one-click workflow with full audit trail.

What you'll learn
  • Which sales ops tasks are highest-leverage for an AI agent
  • How to design the agent — model, tools, guardrails
  • How to chain enrichment, scoring and CRM writes into one workflow
  • How to set this up step-by-step in Dezifi

The agent design

Sales ops work is mostly read-heavy with targeted writes. A mid-tier reasoning model plus tightly scoped CRM access covers 90% of the workload.
  1. 1

    LLM choice

    GPT-4o or Claude Sonnet for account briefs and pipeline risk reasoning. Use a cheaper model (Haiku, GPT-4o-mini) for high-volume enrichment.
  2. 2

    Tools

    Salesforce (read accounts, opportunities, contacts; update fields and notes), HubSpot or Apollo for enrichment, LinkedIn data source if licensed, Slack to post briefs, Gmail or Outlook to draft outreach.
  3. 3

    Guardrails

    Block PII leakage in outbound drafts. Require approval before any Salesforce write that changes stage, amount or close date. Allow free-form notes and activity logs without approval.
  4. 4

    Workflow shape

    Trigger on new lead or stale opportunity. Branch on lead source or deal size. Loop over open opportunities for weekly pipeline review. Approval gate before high-impact CRM updates.

Tools to connect

  • Salesforce — primary CRM. Read/write opportunities, accounts, contacts, activities.
  • HubSpot — alternative CRM or marketing automation source.
  • Apollo or ZoomInfo — lead enrichment with firmographic and contact data.
  • Slack — deliver account briefs to deal rooms, post pipeline alerts to #sales-ops.
  • Gmail / Outlook — draft personalized outreach for AE review.

How to set this up in Dezifi

  1. 1

    Connect Salesforce

    Go to Integrations, pick Salesforce, authorize OAuth. Choose a scoped service account so the agent only sees the territories it should.
  2. 2

    Create the agent

    New Agent → "Sales Ops Copilot". Attach Salesforce, Apollo and Slack. Pick Claude Sonnet as the default model.
  3. 3

    Apply guardrails

    Attach the Standard guardrail profile. Add a policy: any Opportunity update on Stage, Amount or CloseDate routes to an approval queue.
  4. 4

    Build the pipeline-review workflow

    Open Workflows → New. Trigger: weekly cron. Step 1: list opportunities with no activity in 14 days. Step 2: enrich each. Step 3: agent drafts risk summary. Step 4: post to Slack #pipeline-review.
  5. 5

    Test and publish

    Run the workflow against a sandbox Salesforce org. Inspect the trace in Monitor — confirm tool calls, token spend and approval gates fired correctly. Promote to production.

Frequently asked questions

Will the agent overwrite reps' manual edits in Salesforce?
No. Field-level policies let you mark fields like Stage and Amount as approval-required. The agent can suggest a change but a human must accept before the write lands.
How does the agent handle leads with missing data?
The workflow calls the enrichment tool first. If enrichment fails or returns low-confidence data, the agent flags the lead for manual review instead of writing back.
Can we restrict the agent to one sales territory?
Yes. Use a Salesforce service account scoped to the territory, or apply a row-level policy on the integration. The agent only sees records it is authorized to read.
How do we measure ROI?
The Monitor and Analytics modules track every run, cost and reviewer decision. Compare reviewer time saved and CRM data completeness before and after the agent goes live.