A useful RevOps workflow has to survive the second run. It needs a definition of the input, a rule for exceptions, and a record of what changed. A plausible answer in a chat window is only the beginning when the output will affect ownership, pipeline reporting, or someone else's working queue.
The library below covers scoring, routing, enrichment, data hygiene, and recurring reporting. Many existing automation playbooks use Claude Code and name that stack explicitly. The supporting guides also include ChatGPT data analysis and Codex project work, so you can learn the tool you use without mistaking one implementation for another.
From an ICP to a reviewed routing rule
Write the qualification criteria
Define the account attributes and buying signals the team agrees to use. Include exclusions and examples that sit near the boundary.
Open the ICP prompt →Test scoring and routing
Use a small sample and compare the proposed score and owner with your expected decisions. Follow the playbook's named tool stack.
Open the scoring workflow →Review the exceptions
Identify incomplete and conflicting records before writing changes. Keep the evidence and approval record alongside each proposed correction.
Open the CRM hygiene workflow →
Define the decision before connecting the tools
Begin with a business rule you can explain without mentioning a model. For lead routing, that means territory, segment, ownership, exclusions, and a destination for ambiguous records. For a pipeline report, it means stage definitions, the reporting window, and the source of the totals.
Test the rule on a small export containing ordinary records and awkward ones: missing domains, shared accounts, duplicate contacts, and conflicting owners. Keep the original values beside the proposed output. You need to see which cases require judgment before you decide what a recurring job may handle.
Separate reading, proposing, and writing. A report can expose an exception queue before any automation changes the CRM. If the workflow later writes back, give each action a specific field, expected value, and reason. Keep an owner for the rule and a route for unresolved cases.
Make failure visible before scheduling the job
Check how the workflow behaves when it runs twice, receives incomplete input, or loses access halfway through. Repeated execution should not create duplicate notes or send the same alert again. The correct outcome for an uncertain record may be an explicit exception.
Keep a run log with counts of records read, proposals made, changes approved, and exceptions left open. Reconcile totals against the source system. A successful process exit tells you that the program finished; it does not tell you that the business result is correct.
Schedule a workflow only after its outputs, approvals, and failure path are understood. Start with a reviewable report and a named operator. Expand the permitted actions one at a time as the process earns that scope.
Use cases for RevOps
Complete workflows with their implementation tools listed below. Follow each playbook's setup and review steps.
The Ideal Customer Profile (ICP) Template, AI-Scored in Clay
Turn an ideal customer profile template into a live scoring rubric: pull a target account list, enrich it with the signals an AI can read, then score every account against your ICP with a reason and a risk you can audit.
Tools: Clay, Apollo.io, Claude, HubSpotHow to Identify Website Visitors and Trigger Same-Day Outreach
Identify the anonymous website visitors on your site, filter to ICP fits, and trigger a personalized touch the same day, while the intent is still warm.
Tools: RB2B, Clay, Slack, SmartleadSales Follow-Up Email Templates, Auto-Drafted From Every Call
The sales follow-up email is the highest-return touch in the cycle and the one reps most reliably botch, written three hours late from fading memory. Steal the templates below, then let a Claude agent draft each one from the actual call transcript in the rep's voice and write clean CRM notes to HubSpot over MCP. A human still reads every send. Same-day follow-ups, and RevOps finally gets comparable deal data across the team.
Tools: Gong / Fathom, Claude Code, HubSpot (via MCP)Security Questionnaire Template and AI First-Draft Answers
Start from a security questionnaire template, then use AI over your own approved documents to first-draft the answers, turning a multi-day deal-blocking slog into a few hours of focused human review.
Tools: Claude, Google Drive, Google SheetsWeekly Competitor Intel Digest, Run by a Claude Routine
You know the competitive-intel chore: someone is supposed to track what competitors shipped this week, and that someone is slammed. This hands the whole job to a Claude Routine that researches with web search, keeps only the moves that change a deal, and drops a cited digest in Slack every week. Nobody owns the chore, and there is no glue to keep alive.
Tools: Claude web search, Claude Routines, Slack (via MCP)Lead Enrichment and Routing with a Claude Agent
Lead enrichment is the other half of speed-to-lead, and most teams lose it in a queue. Fire every form fill into a Claude agent that enriches the lead with web research, scores it against your ICP, writes the record to HubSpot over MCP, and pings the right rep in Slack, in seconds. No n8n workflow, no Zapier, no human triaging a queue.
Tools: Claude web research, Claude Code, HubSpot (via MCP), Slack (via MCP)Automate Google Ads Management with AI
Most Google Ads accounts run on a human exporting a report, then clicking. Wire yours to a Claude agent instead: it reads the account through the Google Ads API, writes its own weekly diagnosis, mines the search-terms report for wasted spend, and executes only the changes you approve by ticking a checkbox, negatives, bids, budgets, and pauses, the same week, not next month.
Tools: Google Ads API (GAQL), Google Analytics 4, Claude Code, Claude Code Routines, Google Ads (mutate API)Automate LinkedIn Ads Management with AI
Most LinkedIn ad accounts run on a human reading a dashboard, then clicking. Wire yours to a Claude agent on a weekly routine instead: every Monday it ranks the changes worth making, and executes the ones you approve, bids, budgets, pauses, launches, and audience edits, the same day you approve them, not days later.
Tools: LinkedIn Marketing API, Claude Code, Claude Code Routines, LinkedIn Campaign ManagerThe RevOps Dashboard That Reconciles: A Claude Agent Builds One Number Set from CRM, Ads, Product, and Billing
Every team has three tools that each report 'pipeline' and none of them agree, so the revops dashboard becomes a weekly argument about whose number is right. This play makes the definitions the product: a metric definitions file names one formula, one source of truth, and one owner per metric, and a Claude agent reads the CRM, ad platforms, product usage, and billing over MCP every Monday, computes each metric exactly as written, shows the delta between what each tool claims and what the definition says, and writes one static page plus a Slack diff. No BI license, no dashboard nobody opens. One number set, with the disagreements on the page instead of in the meeting.
Tools: HubSpot or Salesforce (via MCP), LinkedIn and Google Ads exports, Product usage and billing (Stripe via MCP, or a warehouse), metrics.md (the definitions file), Claude Code, A static HTML page + Slack (via MCP)Run Your GTM Automations as Claude Agents with MCP
Most GTM automations are a pile of Zapier zaps and n8n flows nobody maintains. Move the logic onto one Claude agent with MCP connectors to your stack: it reads the context, reasons, decides, and writes, all in a brief you can actually read.
Tools: Trigger (webhook or schedule), Claude Code + MCP, HubSpot / Slack / DrivePut Recurring GTM Work on Autopilot with Claude Routines
Every GTM team has a list of recurring chores that quietly slip: the Monday pipeline-risk read, the competitor check, the CRM-hygiene sweep. A Claude Routine runs each one on a schedule as a cloud agent, so the work shows up whether or not anyone remembered it.
Tools: Claude web search / MCP, Claude Routines, Slack / your CRMTurn Your Best GTM Plays into Claude Skills the Whole Team Runs
Your best account brief, battlecard, and follow-up live in one person's head and one clever prompt nobody else can reproduce. Package them as Claude Skills: the play, the format, and the guardrails, versioned and shared, so every rep runs it at the same quality.
Tools: Your best prompts + docs, Claude Skills, Every rep's ClaudeAI Lead Scoring: Score Your Whole Database with a Claude Agent
Most lead scores are a points system someone built in 2021 and nobody trusts, so reps triage the queue by gut anyway. This play replaces it with a Claude agent that scores every lead in your CRM against a three-part model, fit, intent, and engagement, writes the score and a one-line reason back over MCP, re-scores nightly on a Routine, and routes by rule. The reason line is what makes reps stop ignoring the number.
Tools: HubSpot (via MCP), Claude Code, Claude Routines, Slack (via MCP)Pipeline Review on Autopilot: A Monday Digest from a Claude Routine
The Monday pipeline review is the most expensive meeting in sales, and most of it is spent establishing facts that were sitting in the CRM the whole time. This play hands the facts to a Claude Routine: every Monday at 7am it reads the pipeline over MCP, works out what moved, what went stale, what has no next step, and where coverage sits against quota, and posts a written digest to Slack. The meeting starts at the decisions.
Tools: HubSpot or Salesforce (via MCP), Claude Routines, Claude Code (optional), Slack (via MCP)CRM Data Cleanup on a Schedule: A Claude Agent That Proposes, You Approve
CRM data cleanup is the job every RevOps team owns and nobody schedules, so it happens as a panicked spreadsheet purge the week before the board deck. This play makes it a weekly run: a Claude agent reads the CRM over MCP, finds duplicates, non-standard values, empty fields, orphan records, and dead deals, proposes every fix as a line in a change list with the evidence beside it, and writes only the lines you tick. The database gets cleaner every week, and nothing merges without a human looking at it.
Tools: HubSpot or Salesforce (via MCP), Claude Code, FullEnrich or Clay, Slack (via MCP)Churn Risk, Scored on Evidence: A Weekly Claude Agent for Renewals 90 Days Out
Most churn is visible ninety days out, in a support thread nobody in sales read, a champion who quietly changed titles, or a usage line that has been falling since spring. This play makes a Claude agent read all of it every week: every renewal due in the next 90-120 days gets scored against a scorecard you set, on evidence with sources, and every at-risk account gets a one-page brief with the signals, the quotes, and a proposed save plan, routed to its CSM. The renewal that surprised everyone stops happening.
Tools: HubSpot or Salesforce (via MCP), Gong or Fireflies (via MCP), Usage + ticket exports, Claude Code, Slack (via MCP)AI Sales Coaching from Every Call: A Weekly Claude Agent That Scores Reps Against Your Rubric
Most sales coaching happens on the two calls a manager had time to listen to, and the rep picked both. This play puts a Claude agent on every call instead: each week it reads the transcripts over MCP, scores each call against a rubric you own, with the quote that earned each score, and writes every rep a one-page note with a pattern across their calls and one thing to practice. The manager gets a team view that shows where the whole team is weak. Coaching stops being a lottery of which calls got heard.
Tools: Gong or Fireflies (via MCP), HubSpot or Salesforce (via MCP), The rubric file, Claude Code, Slack (via MCP)AI RFP Response with Provenance: A Claude Answer Library That Cites Its Sources
An RFP lands with 180 questions and a deadline in nine days, and the first thing an AE does is open the last RFP and start copying. Half those answers were written for a different buyer, a few are a year stale, and nobody remembers which. This play puts Claude over your approved sources instead: it drafts every answer from the security docs, the product docs, and the answer library, tags each with the source it came from and when that source was last verified, flags anything it could not source as 'needs a human', and exports in the buyer's numbering. The AE spends the nine days on the twenty answers that need a person.
Tools: Notion or Google Drive (via MCP), The buyer's RFP file, Claude Projects, Claude Code, The answer libraryWin-Loss Analysis on Evidence: A Quarterly Claude Agent That Codes Every Closed Deal
Every CRM has a loss-reason field, and in most of them 'price' wins by a mile, because it is the first option and the least embarrassing. The real reasons are on the calls. This play puts a Claude agent on every deal closed in the quarter: it reads the CRM fields, the transcripts, and the buyer interviews, codes each deal against a taxonomy you own with the buyer's quote behind every code, and quantifies the patterns by segment and competitor. The report that lands in the QBR says why you actually lost, in the buyer's words, and how often.
Tools: HubSpot or Salesforce (via MCP), Gong or Fireflies (via MCP), Buyer interview notes, Claude Code, Notion + Slack (via MCP)Sales Forecasting with AI: A Weekly Claude Agent That Builds the Bottoms-Up Forecast Brief
The Monday forecast call spends its first forty minutes reconciling: the CRM says one number, the rep's spreadsheet says another, the manager's gut says a third. This play hands the reconciliation to a Claude agent. Every Monday at 6am it reads every open deal in the quarter, checks each committed deal against a written evidence checklist (a dated next step, a named champion, a paper process that has started, activity in the last ten days), flags deals whose stage disagrees with their evidence, rolls it up by rep against quota and against last week, and writes the forecast call brief. The call starts at the deals worth arguing about. The number stays the human's.
Tools: HubSpot or Salesforce (via MCP), Gong (via MCP), Quota + snapshot files, Claude Code, Slack (via MCP)Account Planning Template + a Claude Agent That Builds Every Rep's Territory Plan
Territory planning is a ritual most teams run once a quarter and abandon by week two: a template goes out, the reps fill in the top of it, the plans go in a folder, and the quarter is worked from the inbox. This play makes the plan build itself and stay alive. At the start of the quarter a Claude agent reads each rep's territory from the CRM, enriches the missing fit fields, tiers every account by fit, signal, and history against a rubric you wrote, drafts a one-page plan for each tier-A account with why now, who to reach, the play, and the risk, cited to sources, and writes the manager a territory summary. The rep edits and commits. A monthly refresh re-tiers on new signals and checks the plan against what actually happened.
Tools: HubSpot or Salesforce (via MCP), Clay, Apollo, or ZoomInfo, Claude web research, Claude Code, The CRM + SlackHow I’d take a cold email draft through QA to a Smartlead import
A finished email draft still needs evidence review, suppression checks and a reliable handoff. I built a small offline workflow for that last mile, with a fictional batch you can reproduce before using your own prospects.
Tools: Python, SmartleadSkills for RevOps
Reusable procedures for the tool or runtime named on each card. Check the setup before adding one to your workflow.
Sales Battlecard Template, Built From Live Competitor Intel
Fills an honest, rep-ready sales battlecard template against a named competitor: positioning, where they win, discovery traps, objection handling, and landmines to avoid.
LinkedIn Ads Optimization Skill for Claude
Reads a LinkedIn ad account by layer and format, flags what moved week over week, and returns a ranked, evidence-backed list of changes to approve. It classifies before it counts and proposes, never executes.
Google Ads Optimization Skill for Claude
Reads a Google Ads account through the API, writes a dated diagnosis, mines the search-terms report for negatives, and returns a ranked, evidence-backed list of typed changes to approve. It categorizes before it counts and proposes, never executes.
MEDDICC Scorecard Template, Scored From Evidence by a Claude Skill
Turns call notes, transcripts, and CRM fields into a MEDDICC scorecard (or BANT, SPICED, or your own framework) where every score is backed by a buyer quote or action, the gaps are ranked by threat to the close date, and one next action is named.
Sales Call Summary Template, Filled From the Transcript by a Claude Skill
Turns a call transcript or raw notes into a sales call summary template filled with only what the buyer actually said, plus a CRM-ready field block (stage, next step, close date, framework fields, each tagged Confirmed or Inferred) and a to-do list ranked by deadline.
ICP Framework Builder: Turn Your Wins and Losses Into a Scoring Rubric
Feeds Claude what you sell plus a dozen recent won and lost deals and gets back an ICP framework a new hire could apply without judgment: measurable disqualifiers, weighted attributes that sum to 100, tier cut-offs, and a back-test against the deals you gave it.
Lead Routing Rules That Explain Themselves: A Claude Skill That Assigns the Owner
Turns a written lead routing rules table into a router that shows its work: for every lead it names the owner, the segment, the SLA, the rule that fired, the out-of-office backup, and the fields it could not see, and it flags catch-all leads so the table gets better every week.
Renewal Playbook Skill: A One-Page Brief and Save Plan for Every Renewal
Runs your renewal playbook against one account: hand it the usage, support, call, and stakeholder evidence and it returns a one-page brief with a Green/Amber/Red risk read, a sequenced save plan drawn from the plays you allow, and the renewal ask. Includes a renewal playbook template to define those plays.
RFP Response Template Skill: Every Answer Cites Its Source
Drops an RFP or security questionnaire onto your approved answer library and returns a draft in the buyer's numbering, with a source and a last-verified date on every answer. It is the RFP response template that refuses to guess: anything unsourced comes back as an exception with an owner, not a plausible paragraph.
Cold email campaign QA: the skill I’d run before importing leads
I want to know exactly why a prospect and a message are ready before either reaches a sender. This skill produces a review queue, ties decisions to the final copy, and exports approved records to a Smartlead import CSV.
Prompts for RevOps
Individual tasks with required inputs and example outputs. Each prompt lists the assistants it supports.
Define Your ICP and Buyer Personas
Distill real win/loss evidence into an ICP, a buying-committee map, and a scoring model your whole team can act on, built on disqualifiers, not aspiration.
Works with: Claude, ChatGPT, GeminiBattlecard: Build a Rep-Ready One vs Any Competitor
A prompt that builds a rep-ready, no-spin battlecard against a named competitor: where you win, where you honestly lose, the traps to set, and the landmines to avoid.
Works with: Claude, ChatGPT, Gemini, PerplexityGuides for your chosen assistant
Follow the guide for the tool you use. A shared workflow directory does not change the setup requirements of an individual playbook.
Claude
- What is MCP? The Model Context Protocol, explained →MCP, the Model Context Protocol, is an open standard for connecting an AI model to external tools and data through one consistent interface.
- Claude Code: install and your first agent run →Claude Code is an agentic tool that runs in your terminal, your editor, a desktop app, or the web, and does multi-step work against your real files and tools.
- Claude Routines: scheduled cloud agents, explained →Claude Routines are scheduled jobs that run in the cloud on a cadence you set, without you or your machine involved.
- Claude Connectors: link Slack, Drive, Calendar & more →Connectors link Claude to the tools you already use, Slack, Google Drive, Google Calendar, your CRM, and more, so it can read your real data and take actions in those systems from inside a chat.
- The best AI for RevOps: whoever owns the loop →RevOps buys systems, so the best AI for RevOps is the one that should own the loop you are automating.
- The AI security review a RevOps lead runs before an agent touches the CRM →An AI security review for a GTM agent answers five questions before the agent gets anywhere near the CRM: what data it can see (classified, with PII named), what it can do (read freely, write only behind a gate), whose permissions it runs under (the connector inherits the user's, so the user matters), what gets logged (audit logs and the usage dashboard on business plans), and how you switch it off.
- Claude cost control for a GTM org: seats, Code, and routines on one ledger →AI cost control for a GTM org means putting three different bills on one ledger: app seats (a fixed price per person, standard or premium), Claude Code (usage that scales with how much reps and builders run it), and API routines (tokens, priced per model).
- Build a shared prompt and skill library your GTM team actually uses →A team prompt library works when it is organized around plays, each with one owner, and split across the two containers Claude gives you: Projects hold the context a role shares (ICP, positioning, tone), and Skills hold the repeatable plays (the pre-call brief, the follow-up note, the weekly digest) as versioned files anyone can invoke.
ChatGPT & Codex
- ChatGPT file uploads and data analysis: check the spreadsheet →Give ChatGPT the source file, metric definitions, desired output and validation checks.
- ChatGPT apps and plugins: connect the right tools →Apps and plugins let ChatGPT work with connected services and reusable capabilities.
- Getting started with Codex: your first checked project change →Install a supported Codex client, sign in, open the intended project and begin with a small task.
- Codex code review and Git: inspect the change before it ships →Use Codex review to inspect a defined diff, then verify the resulting behavior with appropriate checks.
Frequently asked questions
Which RevOps workflow should come first?
Choose a recurring task with clear inputs and an output you can check manually. A report or exception queue is a useful start because it reveals problems before a workflow changes the system of record.
Should I use Claude Code or Codex?
Start with the tool your team can inspect and maintain. A published playbook names its implementation; the Claude Code and Codex guides cover their respective setup and review processes. Moving a workflow between them requires checking its instructions, dependencies, and connections.
How do I connect the CRM safely?
Follow the connector setup in the chosen playbook and begin with the minimum access the task needs. Validate reads on known records, inspect proposed writes, and keep a way to identify and reverse incorrect changes.
What must be in place before an unattended run?
A tested input contract, bounded actions, a record of each run, a failure notification, and a named owner. Decide what happens to ambiguous records and repeated runs before putting the job on a schedule.