AI for RevOps teams

Most bad AI pipeline numbers aren't hallucinations. They're real records, correctly fetched, answering a slightly different question: the wrong date field, the wrong rollup, two currencies added together. Your CRM settles each of these one way, and an assistant that doesn't know which way picks one silently. Write the differences down once, in the project, and Claude can say which one each figure uses.

The definitions themselves are on pipeline coverage ratio, sales forecast accuracy and lead-to-opportunity conversion rate. The project instructions in the setup below turn this table into rules Claude follows.

Most RevOps teams don't need another platform. The CRM, billing and the warehouse stay the systems of record. Claude sits in the middle: a project holds the context above and a glossary of your own stages and fields, and official connectors let it read the tools that hold the data. Count sits beside it and checks the answers.

If your company runs ChatGPT instead.

ChatGPT Business costs the same ($25 a user monthly, $20 annually) and adds custom MCP servers from its plugins page, but a server has to be reachable at a public HTTPS address or through a Secure MCP Tunnel (OpenAI Developers). Stripe publishes an official ChatGPT plugin (Stripe Docs). The Salesforce setup above is Claude's, so check Salesforce's own documentation for ChatGPT, and keep the context file in a ChatGPT project.

You don't need to be technical for most of this. The steps marked You happen in Claude's settings and take about half an hour between them. Salesforce and the warehouse need an admin, and CSV exports cover the gap until that's done. Step 1 gives the project the context from the top of this page, so every workflow below starts from your definitions instead of a guess.

Technical steps for Salesforce, Snowflake and BigQuery (for admins)

• Salesforce: request beta access through AgentExchange; the organisation needs the latest Sales Cloud enterprise edition. Turn on the Salesforce MCP server and give the Claude Owner the External Client App's consumer key and secret, which go in Organization settings > Connectors > Salesforce (Beta) (Claude setup guide).
• Snowflake: create the MCP server (needs the CREATE MCP SERVER privilege) and an OAuth security integration with OAUTH_CLIENT_TYPE = 'PUBLIC' and the redirect URI https://claude.ai/api/mcp/auth_callback, then add the server URL in Claude as a custom connector. Leave the SQL tool's read_only at its default of true (Snowflake docs).
• BigQuery: the remote server is at https://bigquery.googleapis.com/mcp. Grant MCP Tool User, BigQuery Job User and BigQuery Data Viewer, and block execute_sql with an IAM deny policy so only execute_sql_readonly runs (BigQuery MCP docs).

Keep Claude reading, not changing, until you trust it.

Salesforce, HubSpot, Stripe, Chargebee and Slack can all make changes. Salesforce asks before each write by default, and Stripe asks for confirmation before refunds and payouts; for the rest, approve each change one at a time and never choose "Allow always". Snowflake's SQL tool is read-only by default. Claude's Research mode can act without asking, so turn off those tools' editing options in the chat's tools menu before you start a Research run.

The question every pipeline review starts with: is there enough to hit the number? Claude reads open deals from the CRM and the targets from your quota sheet, and works out coverage by segment against what's left to close. Deals with a close date already gone are listed apart, so they don't pad the ratio.

Most forecast calls spend their first ten minutes on what changed since last week. Claude compares this week's forecast categories with last week's snapshot, lists every deal that moved in or out of Commit, and checks last quarter's week-4 call against what actually closed. It doesn't replace your forecasting tool; it explains it.

Stale close dates, missing amounts and duplicate companies quietly bend every number on this page. Claude runs the same checks every Monday and returns each problem as a list of record IDs with an owner, so the fix is a person's job, not the assistant's. It also matches closed-won deals against Stripe, so a deal that never turned into a subscription shows up. The CRM-specific traps are in AI for CRM data.

Planning season starts with one sum: will the team you'll have carry the number finance wants? Claude takes accounts by territory from the CRM, the rep roster from a spreadsheet and last year's attainment from the warehouse, and lays capacity against target territory by territory. The ramp for new hires is the input that moves the answer most, so the prompt makes you state it. Quota attainment is defined on sales rep quota attainment.

Marketing and sales argue about the handoff because they count it differently. This workflow follows one cohort of leads through the lifecycle stages by source, using the dates HubSpot records when a contact enters each stage, so both teams look at the same people. Revenue by ad campaign is a different question, covered in Claude for marketing.

Prompts reduce errors; they don't remove them. In a 2026 study of 14 models, the best produced working links more than 94% of the time, yet only 39 to 77% of their claims were accurate (Onweller et al., Cited but Not Verified, May 2026). A query that runs is the easy part.

Common pitfalls

• Created date for close date. "Q4 pipeline" built on create date counts deals that won't close for a year.
• Adding currencies together. Amount is in each deal's own currency. Summed as one, a dollar deal counts as pounds.
• Adding Closed to Commit twice. In a cumulative rollup Commit already includes it.
• Period rates instead of cohorts. SQLs this quarter divided by MQLs this quarter aren't the same people.
• Trusting a capped result. HubSpot returns 200 records a page (HubSpot Developers). A count of exactly 200 is one page.
• Mixing booked and billed. CRM ARR and Stripe MRR answer different questions, and both are right.

Rules for the team

• Definitions live in the project. Change them there, once, not in individual chats.
• Read-only by default. Claude lists record IDs; a person merges, closes or reassigns them.
• Board figures come from the approved snapshot. Claude can explain them, not restate them.
• Every reason gets a record. "Slipped because of legal" needs the note or the message that says so.
• Customer data stays in the Team workspace, never a personal account.

One check before anything leaves RevOps

Count audits every answer automatically, inside the chat, and it checks more than the numbers. For each claim it asks: did this come from a source, and does the source really say it? Does the reasoning hold, or does a conclusion go further than the data? Was a definition or a formula assumed without saying so? Is a statement from the model's memory rather than anything in the chat? It replies with a canvas where each claim is labelled a citation, an assumption, model knowledge or a hallucination, with the reasoning chain under it. A Commit figure that counts closed deals twice, a currency summed as another and a slipped deal explained by a reason nobody recorded all show up for what they are. Each workflow above shows an example.

How do I use AI for RevOps?

How much does it cost to set up AI for a RevOps team?

Can Claude connect to Salesforce?

Why does AI give a different pipeline number from my CRM report?

Can AI do sales forecasting?