Free RevOps agent teardown
How to build a RevOps agent that qualifies leads, books meetings, and updates your CRM
This is the practical blueprint: the fields to collect, the routing logic to use, the prompts to start from, and the guardrails that keep an agent useful instead of noisy.
The workflow at a glance
Start with one job, not a fantasy agent
The mistake most teams make is asking an agent to “handle sales ops.” That is not a workflow; it is a vague department. A reliable RevOps agent needs one measurable job: when a new lead arrives, decide whether it is worth a fast sales response, prepare the handoff, and keep the CRM clean. If it cannot explain its decision in fields your team already uses, it is not ready for production.
Define the first version around inbound leads from forms, chat, webinar signups, referrals, and product usage signals. The agent should not invent strategy. It should apply your ICP rules consistently, reduce manual sorting, and make the next human action obvious. That keeps scope small enough to launch while still saving real RevOps time.
Step 1: capture a clean lead packet
Before prompts matter, inputs matter. Create a lead packet with the same fields every time: full name, work email, company, role, company size, source, page or campaign, stated pain, requested timeline, existing CRM owner, current lifecycle stage, and any product events. Add an empty field for agent notes and another for agent confidence. If your forms do not collect all of this, use enrichment or leave fields blank; do not let the agent guess facts that should come from systems.
Your intake should also include disqualifiers. Examples: student email, competitor domain, unsupported geography, agency researching for a client, no business email, or a company below your minimum account size. These are not moral judgments. They are routing shortcuts that protect the sales team from noise.
Step 2: score fit and intent separately
Do not collapse everything into one “good lead” score. Fit and intent answer different questions. Fit asks, “Could this account become a good customer?” Intent asks, “Should sales respond now?” A large perfect-fit account browsing a careers page is not the same as a smaller qualified account asking for a demo this week.
Use a simple 0–5 score for each. Fit can weigh industry, company size, role seniority, tech stack, region, and strategic segment. Intent can weigh demo request language, pricing-page visits, problem specificity, urgency, referral source, and recent engagement. Then create routing rules: 8–10 combined points gets an immediate sales meeting path, 5–7 gets a personalized nurture or SDR review, and 0–4 gets low-priority nurture or suppression.
Example qualification prompt
You are a RevOps lead qualification agent. Your job is to score inbound leads and explain the next action.
Inputs:
- ICP rules: {{icp_rules}}
- Lead packet: {{lead_packet}}
- Current CRM record: {{crm_record}}
Return JSON only with:
fit_score: 0-5
intent_score: 0-5
qualification_status: qualified | nurture | disqualified | needs_review
reasoning: 3 concise bullets grounded in the provided inputs
recommended_next_action: one concrete action for sales or marketing
missing_fields: fields that would materially improve confidence
crm_updates: lifecycle_stage, lead_status, priority, owner_notes
Rules:
- Do not invent missing facts.
- If confidence is low, choose needs_review.
- If the lead is qualified, write a one-sentence meeting angle tied to their stated pain.Step 3: book meetings without pretending to be human
Meeting booking should be direct and transparent. The agent can draft or send an email, but it should never fake urgency, invent relationships, or imply a human has reviewed something that only an agent processed. The best booking flow uses three branches: send a calendar link for highly qualified demo requests, ask one clarifying question when intent is strong but the use case is unclear, and route borderline accounts to nurture.
The meeting email should include the pain the lead already mentioned, the outcome of the call, and one frictionless scheduling link. Avoid a six-paragraph “personalized” essay. A useful agent writes like a sharp RevOps teammate: brief, specific, and easy to act on.
Example booking prompt
Write a meeting-booking email for this qualified lead.
Context:
- Lead pain: {{pain}}
- Meeting angle: {{meeting_angle}}
- Calendar link: {{calendar_link}}
- Sender: {{sender_name}}
Constraints:
- 90 words maximum.
- No fake familiarity.
- One clear CTA.
- Mention the business outcome, not generic AI automation.
- If the qualification_status is needs_review or nurture, do not include a calendar link; write a clarifying question instead.Step 4: update the CRM like an operator
CRM updates are where agents usually create mess. The fix is to make the agent write to a narrow field map, not a blank canvas. Map each output to fields your CRM already understands: lead status, lifecycle stage, owner, priority, qualification reason, next step, follow-up date, source detail, and last-agent-run timestamp. Keep the long reasoning in an internal note, not scattered across sales fields.
Add idempotency. The agent should check whether it already processed the same lead-source-event combination before creating a task or note. Otherwise one form resubmission can create duplicate tasks, duplicate emails, and a sales rep who disables the whole workflow by Friday.
Step 5: launch with human review, then remove it carefully
Run the agent in shadow mode for the first 25–50 leads. It should produce scores, notes, emails, and CRM updates without sending anything. Compare its decisions to your best operator. Track false positives, false negatives, missing-field rates, and whether sales actually accepts the meeting angle. Your goal is not perfect AI judgment. Your goal is a repeatable operating system that beats a neglected inbox.
When the agent is accurate, automate low-risk actions first: CRM notes, owner tasks, and nurture routing. Keep outbound sends or calendar booking behind approval until you trust the edge cases. Review the prompt every week for the first month, especially after pricing, ICP, or territory changes.
Step 6: measure the agent like a revenue system
A RevOps agent is not “working” because it runs without crashing. It works when it improves a revenue metric or removes a measurable operational drag. Track speed-to-lead, qualified-to-booked conversion, meeting no-show rate, disqualification accuracy, CRM field completeness, duplicate-task rate, and the percentage of agent decisions that humans override. Those numbers tell you whether the agent is helping sales or simply producing polished busywork.
Review outcomes by segment. The agent may perform well on demo requests and poorly on webinar leads, or it may over-score founder titles while under-scoring RevOps managers with budget authority. Feed those misses back into the rubric instead of writing a giant prompt. The best operating loop is small: inspect the run log, adjust one rule, test against recent leads, then ship the change. That is how the workflow compounds instead of drifting.
The minimum stack
You can build this with a form source, a spreadsheet or queue table, one LLM call for qualification, one LLM call for messaging, and CRM API actions. Keep logs from day one: input snapshot, model output, CRM write result, email draft, reviewer decision, and final outcome. Those logs become your QA dataset and make the workflow safer every time it runs.
The version worth shipping is boring in the best way: consistent inputs, strict JSON outputs, narrow CRM writes, and a clear escalation path. Once that works, you can add enrichment, territory routing, Slack alerts, account research, and closed-loop reporting. But the core should always stay simple: qualify, route, book, update.
Ready-to-run pack
Want the templates instead of rebuilding this from scratch?
The Operyxia starter pack includes the workflow definition, three production-ready prompts, a setup guide, CRM field map, lead intake CSV, and run-log template.
Skip the setup — get the ready-to-run RevOps Agent Starter Pack for $29