Sales Agent Playbook
Attach sales agents to CRM and seller workflows with clear autonomy and quality controls.
Key takeaways
- Sales agents should work inside the sales process, reducing research, prep, follow-up, coaching, and CRM friction while preserving seller judgment.
- Map use cases from account research and meeting prep to next-best action, follow-up, coaching, and CRM hygiene, each with defined inputs and outputs.
- Use five autonomy levels from L1 assistant to L5 autonomous; most teams should start at L1 to L3 and only reach L4 to L5 with clean permissions, logging, and rollback.
- Agents must operate in CRM, show recommendations on the record with source links, and create tasks only after approval.
- Guard quality against hallucination, over-automation, CRM pollution, recommendation fatigue, and playbook mismatch, feeding outcomes back into scoring.
Sales agents should work inside the sales process, not beside it. Their value comes from reducing research, preparation, follow-up, coaching, and CRM friction while preserving seller judgment.
Sales Agent Use Cases
| Use case | Input | Output |
|---|---|---|
| Account research | CRM, website, news, hiring, usage logs | Account summary, trigger, hypothesis |
| Meeting prep | Prior conversations, tickets, product usage, persona | Agenda, questions, expected objections |
| Next-best action | Stage, engagement, intent, risk | Recommended action and reason |
| Follow-up | Call transcript, commitments, product assets | Email draft, task, mutual action plan |
| Coaching | Call transcript, playbook | Improvement points and role-play questions |
| CRM hygiene | Email, meeting, activity log | Field update draft and missing-field alert |
Autonomy Levels
| Level | Description | Example |
|---|---|---|
| L1 Assistant | Responds when asked | Call summary |
| L2 Recommender | Suggests next action | "Send the security FAQ" |
| L3 Drafting Agent | Creates execution drafts | Email, proposal, CRM update |
| L4 Human-approved Agent | Executes after approval | Add to sequence, create task |
| L5 Autonomous Agent | Executes within policy | Low-risk nurture, lead routing |
Do not skip levels
Most teams should start at L1 to L3. L4 and L5 require clean permissions, logging, rollback, and approval policy.
Must Operate in CRM
| Requirement | Why it matters |
|---|---|
| Show recommendation on account/opportunity record | Seller can judge in context |
| Provide source links | Trust and verification |
| Create tasks only after approval | Recommendation becomes action |
| Capture outcome feedback | Recommendation quality improves |
| Enforce permission-based access | Protect sensitive customer data |
Quality Control
| Failure mode | Prevention |
|---|---|
| Hallucination | No unsourced facts |
| Over-automation | Approval required for high-risk actions |
| CRM pollution | Show field-update diff |
| Recommendation fatigue | Limit number of recommendations per account |
| Playbook mismatch | Tie recommendations to official methodology |
Operating Checklist
- Sales agents run inside CRM or the seller's primary workflow.
- Every recommendation includes source and reason.
- CRM updates are drafts or diffs before approval.
- High-risk customer actions require human approval.
- Outcomes feed back into scoring and playbook tuning.