Every Action AI Can Take on Your CRM: A Working Catalog

Most AI-in-CRM marketing is abstract: "AI-powered insights," "intelligent assistance," "agentic workflows." This post is the opposite. It is a concrete catalog of actions AI in a modern CRM can actually take, grouped by the record type they operate on. Every one is a proposal that lands in a review queue, not an autonomous "send and hope" action.

Email

Draft a reply in your voice. "Draft a reply to Sarah's email about Q3 pricing." The AI reads the thread, the sender's contact record, any deals or tickets they're on, and produces a draft that sounds like how you actually write.

Generate a subject line. "Give me three subject-line options for the follow-up to Acme." The AI produces variants matched to the content and the relationship.

Triage the inbox. "What needs my attention in the next hour?" The AI surfaces the emails that actually move work forward today.

Write a re-engagement email. "Draft a warm re-intro to our dormant contacts at Globex." The AI reviews the last interaction, the deal history, and drafts a non-generic re-intro.

Schedule a follow-up. "Remind me to follow up with James if he hasn't replied by Friday." The AI creates the task with the right trigger.

Contacts

Create a new contact. "Add Priya Patel at Initech as a new contact, VP of Ops."

Update an existing contact. "Update Mike Chen's title to CFO."

Enrich a contact. "Pull firmographics for Stark Industries and fill the missing fields."

Find likely duplicates. "Are there duplicate records for J. Smith at Acme?"

Score a contact. "Score my top 10 open-opportunity contacts by engagement."

Research before a call. "Brief me on Maya Rivera before my 2 PM with Globex." The AI summarizes recent company news, the contact's role history, the relationship activity, and your open deals.

Companies

Create a new company. "Add Umbrella Corp, enterprise segment."

Update fields across records. "Set industry to 'Manufacturing' for every account in our West region."

Link parent and subsidiary. "Link Acme Europe as a subsidiary of Acme Corp."

Research company news. "What's happened at Initech this quarter?"

Merge duplicates. "Merge the two Globex records into the one with the most activity."

Deals and Opportunities

Create a new opportunity. "Create a $45K opp with Acme, Q3 close, mid-market tier."

Advance stage with approval. "Move the Stark deal from Proposal to Negotiation."

Score risk with reasoning. "Which of my open deals are at risk and why?" The AI returns ranked deals, each with a short explanation: activity decline, stage-exit timing, champion changes, competing deals.

Draft the next follow-up. "Draft a check-in for the Acme opportunity. They've been quiet for 10 days."

Summarize the deal timeline. "What's happened on the Globex deal since we met in May?" Timeline summary pulled from activities, emails, and notes.

Forecast probability. "Update the probability on the Stark deal based on recent activity."

Tasks and Activities

Create tasks in bulk. "Create follow-up tasks for every deal with no activity in 7 days."

Assign a task. "Assign the Initech renewal prep to Priya."

Reschedule. "Push this week's follow-ups to next Monday."

Summarize call notes. "Summarize my call notes from yesterday's meeting with Acme."

Log an activity. "Log a 30-minute discovery call with Maya Rivera."

Quotes

Draft a quote. "Draft a quote for Acme: 25 seats on Business, annual." The AI builds line items, applies standard terms, and produces a draft ready for review.

Generate a share link. "Give me a public link to the Globex quote." Prospects can view and accept without logging in.

Campaigns

Draft a campaign email. "Draft a Q3 re-engagement email for contacts who haven't opened anything in 90 days."

Suggest audience segments. "Who should get the Q3 upsell offer?"

Propose an A/B test. "Give me two subject-line variants for the renewals campaign."

Pipeline and Forecasting

Summarize pipeline health. "How's the pipeline this week?" A written summary in plain English, with the material changes highlighted.

Identify deals at risk. "What deals look like they're slipping?"

Suggest next actions. "What should I prioritize today?" A ranked list of specific actions based on deal state across the book of business.

Reports

Draft a report. "Build a report of closed-won revenue by source for last quarter."

Explain a chart. "Why did conversion drop in April?" The AI correlates the chart with activity, pipeline movement, and record changes to produce a sourced answer.

Export as PDF. "Send the win-loss analysis as a PDF to the leadership team."

Search

Search in plain English. "Find everyone at Initech that I've emailed in the last month."

Filter by any axis. "Open deals over $50K, owned by Priya, closing this quarter."

What Happens When You Approve

Every action above is a proposal. When you approve one (individually or in bulk) it takes effect in the CRM. A proposed task is created. A proposed email goes out once you approve it. A proposed record update happens. Every approval is logged in the audit trail.

The approval queue is what makes this list safe. Without it, an action catalog this broad would be dangerous: AI making silent changes to records you didn't sign off on. With it, every entry in the catalog is staged work, not committed work, until you say yes. That staging step is the whole reason a list this long can stay in one product without becoming reckless. It is also the line between AI that does the work and AI that merely talks about it, which we draw in full in what an AI CRM does that advice tools don't.

What's Not in This Catalog

Honest scope matters. AI in a modern CRM cannot:

• Delete records without a second confirmation.
• Send emails without you approving the draft.
• Bypass your role's permissions.
• Act on behalf of a teammate without impersonation being explicitly enabled.
• Make contract or pricing commitments on your behalf.

These limits aren't bugs. They're deliberate. The action catalog is useful precisely because the things it can't do are bounded clearly.

How to Think About This as a Buyer

If you're evaluating AI in a CRM, this catalog is a buying checklist. Ask the vendor for a live demo of five specific actions from the list above, picked at random by you, not by them. Watch what happens when the model finishes thinking: does the record change, the draft appear, the task land in the queue? Or does the answer arrive as a paragraph describing what you could go do next? A vendor who can run five random actions end to end has built the thing this post catalogs. A vendor who narrates instead has built something else. The unannounced live demo is the cleanest test there is.