CRM AI Agent Workflow Examples and Templates (2026)

Written by: Doug Camplejohn, CEO & Co-Founder, Coffee

Key Takeaways for CRM AI Agent Workflows

• A CRM AI agent workflow automates the data-in step, monitoring triggers, applying AI reasoning, and producing scored outputs while humans keep final approval on high-stakes decisions.
• The ten production-ready templates cover the full revenue cycle, from inbound lead qualification and stalled-deal rescue to churn-risk alerts, pre-call research, and forecast roll-ups.
• Each template follows the same four-part pattern: trigger, AI reasoning, structured CRM updates, and a human approval gate, which keeps automation safe and auditable.
• Coffee ingests both structured CRM fields and unstructured data such as emails, transcripts, and notes into a built-in data warehouse, then writes clean outputs back to HubSpot or Salesforce without forcing reps to switch tools.
• Launch your first Coffee agent workflow this week and start removing manual CRM data entry.

The ten templates below share a common architecture with four parts: a clear trigger, AI reasoning steps, structured CRM updates, and a human approval gate. Each workflow focuses on a different friction point in your revenue process, from first touch to renewal. Use the comparison table to spot which agents match your current bottlenecks before you dive into the detailed templates.

Master Comparison Table: 10 Production-Ready CRM AI Agent Workflows

Inbound Lead Qualification Agent Template

Trigger: New form submission, inbound email, or lead created with status = “new.”

AI reasoning steps + confidence scoring:

1. Enrich the record with firmographic data such as company size, industry, and funding stage.
2. Score fit against ICP criteria using BANT or a custom rubric, then output a fit_score (0–100) and intent_score (0–100).
3. Assign a priority band: P1 (high fit and high intent), P2 (moderate), or P3 (nurture only).
4. Draft a personalized follow-up that references the lead’s specific inquiry.
5. Flag confidence below 60 for human review before any routing occurs.

Structured CRM fields updated: Lead Score, Priority Band, Fit Score, Intent Score, Routing Reason, Lifecycle Stage.

Human approval gate: The rep reviews the 5-line brief (company, role, BANT pass or fail, score, enrichment) and approves routing before the record is promoted or the draft is sent. The agent hands off immediately when the lead requests a human, mentions a committed competitor, or raises a compliance question.

HubSpot notes: Add custom properties for fit_score, intent_score, priority_band, and routed_reason. Update Lifecycle Stage and Lead Status by rule. Store the full AI decision JSON in an engagement note for audit.

Salesforce notes: Create custom Lead fields Priority_Band__c, Fit_Score__c, Intent_Score__c, and Routing_Reason__c. Use assignment rules or API-based owner setting. Create a Task for P1 leads with a due date tied to SLA.

Coffee implementation: Manual lead triage often relies on partial data and gut feel. In Companion App mode, Coffee writes fit_score, intent_score, and the routing brief directly back to the existing HubSpot or Salesforce record so reps see a clear recommendation in context. In Standalone CRM mode, Coffee owns the record entirely and applies the same scoring logic. Because Coffee ingests unstructured data such as the lead’s email text, form answers, and enrichment alongside structured fields, the scoring model uses context that rules-based workflows miss.

Deploy the inbound qualification agent to automate lead scoring without replacing your existing CRM.

Once a lead converts to an opportunity, deal momentum becomes the next challenge. Idle opportunities signal lost interest or weak follow-up, which an agent can surface and address automatically.

Stalled Deal Agent Template

Trigger: No activity logged on an open opportunity for a configurable number of days, or deal stage unchanged beyond the average time-in-stage threshold.

AI reasoning steps + confidence scoring:

1. Pull all emails, call transcripts, and meeting notes associated with the opportunity.
2. Identify the last meaningful buyer signal and classify deal health as Active, At-Risk, or Stalled.
3. Generate a deal health score (0–100) and a recommended next action such as re-engage email, executive escalation, or close-lost.
4. Draft a re-engagement message personalized to the last known objection or topic.
5. Flag confidence below 65 for AE review before any outreach is queued.

Structured CRM fields updated: Deal Health Score, Next Step, Last Activity Date, Stall Reason.

Human approval gate: The AE reviews the recommended action and draft message before any external communication is sent. Close-lost recommendations receive extra scrutiny because they carry high business impact.

HubSpot notes: Use a workflow enrollment trigger on “Days Since Last Engagement” property. Write deal health score to a custom Deal property. Log the AI reasoning summary as a Note engagement.

Salesforce notes: Use a scheduled Flow or Process Builder trigger on Opportunity Last Activity Date. Write to custom fields Deal_Health_Score__c and Stall_Reason__c. Create a Task assigned to the opportunity owner.

Coffee implementation: Coffee’s Pipeline Compare feature surfaces week-over-week deal movement automatically, so stalled opportunities appear without manual CSV exports. The agent reads call transcripts and email threads, which are unstructured data that legacy CRMs cannot parse, and then produces a stall reason grounded in actual conversation history.

Meeting-to-CRM Follow-up Agent Template

Trigger: Call or video meeting ends and a transcript becomes available via the connected conferencing tool.

AI reasoning steps + confidence scoring:

1. Transcribe and summarize the meeting using the configured sales methodology such as BANT, MEDDIC, or SPICED.
2. Extract action items, commitments, and next steps with owner and due date.
3. Update deal stage or qualification fields if new BANT signals were captured.
4. Draft a follow-up email that references specific discussion points.
5. Score summary completeness and flag incomplete transcripts for manual review.

Structured CRM fields updated: Meeting Summary, Action Items, Next Activity Date, BANT or MEDDIC Fields, Follow-up Draft.

Human approval gate: The draft follow-up routes to the rep for review and editing before any message is sent. Stage changes derived from the meeting require rep confirmation before the opportunity record is updated.

HubSpot notes: Connect via HubSpot’s native calling or a Zoom or Teams integration. Use the “Meeting Outcome” property to trigger the workflow. Write the summary to the Notes engagement object.

Salesforce notes: Use an Event or Task record as the trigger. Write the summary to the Description field on the related Opportunity. Create child Tasks for each extracted action item.

Coffee implementation: Reps often forget details once a call ends. Coffee’s AI Meeting Bot joins calls on Zoom, Teams, or Meet, transcribes in real time, and posts the structured summary, including BANT or MEDDIC fields and a draft follow-up email, directly to the Gmail drafts folder for one-click review. In Companion App mode, all fields write back to HubSpot or Salesforce automatically.

Churn-Risk Alert Agent Template

Trigger: Usage intensity drops below a defined threshold, support ticket volume spikes, or sentiment analysis on recent communications turns negative.

AI reasoning steps + confidence scoring:

1. Aggregate usage data, ticket history, and communication sentiment into a unified account health view.
2. Score churn risk on a 0–100 scale by combining engagement signals, fit, and historical closure patterns.
3. Identify the specific product or service most at risk.
4. Recommend a retention play such as executive outreach, QBR acceleration, or feature enablement.
5. Flag accounts with risk score above 75 for immediate CSM escalation.

Structured CRM fields updated: Churn Risk Score, Health Status, Risk Reason, CSM Task.

Human approval gate: High-impact retention actions like executive escalation emails or contract concessions require CSM approval before customer contact. The agent drafts and the human decides.

HubSpot notes: Use the Breeze Customer Health Agent (beta) signals or a custom workflow on a calculated Health Score property. Write churn risk to a custom Contact or Company property.

Salesforce notes: Use Einstein Prediction Builder or a custom Apex trigger on usage data. Write to Churn_Risk_Score__c and Health_Status__c on the Account object.

Coffee implementation: Coffee maintains a built-in data warehouse with full interaction history, so the churn-risk agent compares current engagement against a baseline without manual data pulls. The agent surfaces the risk reason in plain language, grounded in actual email and call transcript evidence.

Pre-Call Research Brief Agent Template

Trigger: A meeting is added to the rep’s calendar within a configurable lead time, such as 24 hours before the scheduled start.

AI reasoning steps + confidence scoring:

1. Identify all attendees and resolve their CRM records.
2. Pull prior interaction history, open opportunities, and last known objections.
3. Enrich attendee profiles with current title, LinkedIn, and company news.
4. Produce a structured brief with attendee roles, deal context, recommended talking points, and risk flags.
5. Score brief completeness and flag missing data fields for rep awareness.

Structured CRM fields updated: Briefing Note logged as an engagement, with no destructive field changes at this stage.

Human approval gate: The brief is read-only for the rep. No external action is taken, so no approval gate is required and the rep simply reviews and enters the call prepared.

HubSpot notes: Trigger via the HubSpot Meetings tool or a calendar sync event. Deliver the brief as a Note on the associated Contact and Deal record.

Salesforce notes: Trigger via an Event record. Deliver the brief as a Task or Chatter post on the related Opportunity.

Coffee implementation: Manual pre-call research often requires toggling between LinkedIn, the CRM, and email threads. Coffee’s “Today” page surfaces the pre-call brief automatically each morning and pulls from the built-in data warehouse to include historical context from past calls and emails, not just the most recent CRM note.

Support-Triage Routing Agent Template

Trigger: A new support ticket is created via email, chat, or web form.

AI reasoning steps + confidence scoring:

1. Classify the ticket by category such as billing, technical, onboarding, or escalation.
2. Assess priority using account tier, SLA terms, and sentiment score.
3. Match to the appropriate support queue or agent based on skills and workload.
4. Draft an acknowledgment response personalized to the issue type.
5. Flag tickets with ambiguous category or high-distress sentiment for immediate human review.

Structured CRM fields updated: Ticket Priority, Category, Assigned Queue, SLA Flag, Sentiment Score.

Human approval gate: Tier-2 escalations, large credit issuances, and compliance-sensitive requests require human confirmation before action. Routine acknowledgments can send autonomously after a configurable confidence threshold is met.

HubSpot notes: Use Service Hub ticket pipelines. Trigger via ticket creation webhook. Write category and priority to native Ticket properties.

Salesforce notes: Use the Case object. Trigger via a Flow on Case creation. Write to Case Priority, Type, and Owner fields via assignment rules.

Coffee implementation: In Companion App mode, Coffee reads the incoming ticket text as unstructured data alongside the account’s CRM history as structured data. The triage decision then reflects full customer context instead of relying only on the ticket body.

Data-Cleanup Agent Template

Trigger: Scheduled daily scan or a duplicate-detection event fired when a new record is created with a matching email or domain.

AI reasoning steps + confidence scoring:

1. Scan for duplicate contacts and companies using email, domain, and name fuzzy-matching.
2. Identify records with missing required fields such as phone, title, or company.
3. Enrich incomplete records from licensed data partners.
4. Produce a merge proposal with a confidence score for each duplicate pair.
5. Flag low-confidence merges below 80 for human review before execution.

Structured CRM fields updated: Duplicate Flag, Enriched Fields, Data Quality Score, Merge Status.

Human approval gate: Record merges and deletions require ops team approval before execution. Enrichment of missing fields below a risk threshold can proceed autonomously with logging.

HubSpot notes: Use HubSpot’s native duplicate management API combined with a custom workflow. Log merge proposals as Notes. Write Data_Quality_Score to a custom Contact property.

Salesforce notes: Use Duplicate Rules and Matching Rules on the Contact and Account objects. Write merge proposals to a custom Duplicate_Flag__c field. Require admin approval for Account merges.

Coffee implementation: Coffee’s agent continuously enriches contacts and companies from licensed data partners once connected to Google Workspace or Microsoft 365. This ongoing enrichment slows data decay so periodic cleanup projects become lighter and less risky.

Visitor-to-Lead Conversion Agent Template

Trigger: Tracking pixel fires when an anonymous visitor lands on a high-intent page such as pricing, demo request, or product comparison.

AI reasoning steps + confidence scoring:

1. Resolve the visitor’s identity, including name, title, email, LinkedIn profile, and company.
2. Score fit against the defined buyer persona using firmographic and behavioral signals.
3. Identify the two or three most relevant contacts inside the visiting company.
4. Draft a personalized outreach message that references pages visited and time on site.
5. Flag visitors with incomplete identity resolution for manual review before outreach.

Structured CRM fields updated: Visitor Identity, Company, Pages Visited, Time on Site, Lead Source, Fit Score.

Human approval gate: The rep reviews the identified visitor and the suggested contacts before the record is created in the CRM or outreach is initiated. No contact is made without rep confirmation.

HubSpot notes: Use HubSpot’s tracking code alongside a custom workflow triggered by a Contact property set by the identification layer. Write visitor data to Contact and Company records.

Salesforce notes: Create a Web-to-Lead record from the pixel payload. Write visitor behavior to custom Lead fields. Assign to the appropriate rep via territory rules.

Coffee implementation: Coffee’s Visitor Identification feature resolves anonymous traffic to named individuals, not just companies, and surfaces Suggested Leads that match the configured buyer persona. A single click adds the prospect to Coffee with all enrichment pre-filled so the rep can start LinkedIn outreach or enroll the contact in a Campaign sequence immediately.

Campaign Performance Review Agent Template

Trigger: Scheduled weekly event or campaign-end signal from the marketing automation platform.

AI reasoning steps + confidence scoring:

1. Pull campaign metrics such as sends, opens, clicks, replies, meetings booked, and pipeline influenced.
2. Compare performance against historical benchmarks and identify anomalies.
3. Score campaign health (0–100) and classify as Performing, Underperforming, or Anomalous.
4. Recommend specific adjustments such as audience refinement, subject line variant, or budget reallocation.
5. Flag recommendations that involve budget changes above a defined threshold for human approval.

Structured CRM fields updated: Campaign Score, Anomaly Flag, Recommended Adjustment, Performance Summary.

Human approval gate: Budget or audience changes require marketing lead approval before execution. Informational summaries can be distributed autonomously.

HubSpot notes: Trigger via a HubSpot Campaign workflow or a scheduled report. Write the performance summary to the Campaign record’s Notes field.

Salesforce notes: Use the Campaign object and Salesforce Reports API. Write Campaign_Score__c and Anomaly_Flag__c to the Campaign record. Route budget recommendations to a Campaign Approval process.

Coffee implementation: Coffee’s Campaigns feature tracks per-step stats such as emails sent and replies received natively. The performance review agent draws from the same data warehouse that ran the campaign, so teams avoid reconciling data across separate tools.

Forecast Roll-up Agent Template

Trigger: Scheduled pre-QBR cadence or a pipeline-review event initiated by the RevOps team.

AI reasoning steps + confidence scoring:

1. Pull all open opportunities with stage, amount, close date, and last activity.
2. Apply a confidence score to each deal based on engagement signals and historical win rates.
3. Categorize deals into Commit, Best Case, and Pipeline forecast buckets.
4. Flag deals where rep-submitted stage conflicts with activity-based confidence score.
5. Produce a roll-up summary with risk-adjusted forecast and deal-level commentary.

Structured CRM fields updated: Forecast Category, Confidence Score, Risk Flag, Forecast Commentary.

Human approval gate: RevOps approves the final forecast before executive distribution, with extra attention on deals where the agent’s confidence score diverges significantly from the rep’s submitted stage.

HubSpot notes: Use the Forecast tool in Sales Hub. Write Confidence_Score and Risk_Flag to custom Deal properties. Deliver the roll-up as a Dashboard or exported report.

Salesforce notes: Use the Forecasting object and Opportunity fields. Write AI_Confidence_Score__c and Risk_Flag__c to the Opportunity. Route the roll-up summary to the RevOps Chatter group.

Coffee implementation: Coffee’s Pipeline Compare feature visualizes week-over-week changes automatically, including progressed deals, stalled opportunities, and new additions, so teams replace manual CSV exports. Because the agent captures activity from emails and calls into the built-in data warehouse, the confidence score reflects actual buyer engagement rather than self-reported deal stages.

Use Coffee to remove manual forecast prep from your next pipeline review.

Frequently Asked Questions

Does Coffee integrate with existing HubSpot or Salesforce instances?

Coffee integrates cleanly with existing HubSpot or Salesforce instances. Coffee operates in two modes. As a Companion App, it authenticates with your existing HubSpot or Salesforce instance via a simple OAuth connection and writes enriched data, meeting summaries, lead scores, and pipeline intelligence directly back to your system of record. Reps stay inside their primary CRM while the agent works in the background. For teams not yet committed to a legacy CRM, Coffee also functions as a Standalone CRM where the agent manages the entire system of record. Broader third-party integrations beyond HubSpot and Salesforce are currently available via Zapier, with deeper native integrations on the product roadmap.

Is Coffee secure and compliant?

Coffee is SOC 2 Type 2 certified and GDPR compliant. Customer data is not used to train public AI models. For teams in regulated industries, Coffee recommends a direct conversation with the sales team to assess fit, because the product is optimized for small-to-mid-market companies rather than organizations that require multi-year enterprise security reviews.

How does Coffee’s data quality compare to ZoomInfo or Apollo?

Coffee’s enrichment data, including job titles, funding information, LinkedIn profiles, and firmographic details, is sourced from licensed data partners and is broadly on par with standalone enrichment tools for most small-to-mid-market use cases. The key difference is consolidation. Coffee builds enrichment directly into the agent workflow, so reps do not need a separate ZoomInfo or Apollo subscription to access prospect data. The Lead Finder feature allows natural-language searches against Coffee’s own database, and results live in the same system that runs outreach sequences, which removes the CSV export step entirely.

How does Coffee’s seat-based pricing work?

Coffee uses straightforward seat-based pricing. Organizations pay for the number of human users, and the agent’s labor such as data entry, enrichment, meeting summaries, pipeline monitoring, and outreach drafting is included without additional metering on LLM usage or workflow executions. There are no per-process fees or complex consumption tiers. This model allows RevOps leaders to forecast the cost of the agent the same way they forecast a headcount addition.

Conclusion: Stop Being Data Clerks

B2B sales reps spend an average of 2.3 hours per day (11.5 hours per week) on CRM input, and RevOps leaders often cite CRM hygiene as a significant recurring time drain. That time goes into manual data entry, enrichment lookups, and post-meeting note-taking, which all require context but little judgment. The 10 templates above reclaim those hours by automating the data-in step while preserving a human approval gate for every decision that carries meaningful business risk.

Coffee is the only agent that ingests both structured CRM fields and unstructured data such as emails, call transcripts, and meeting notes into a built-in data warehouse, then writes clean, context-rich outputs back to HubSpot or Salesforce without forcing reps to toggle between tools. Whether the team needs a Standalone CRM or a Companion App layered on top of an existing instance, the agent handles the busywork so humans can focus on selling.

Put your first Coffee agent into production this week and give your team relief from CRM data entry.