SparkSense Agent Mode
How SparkSense Agent Works
SparkSense Agent connects to your CRM, retrieves account, contact, and opportunity data, then uses AI to identify patterns across your successful deals. It produces a complete ICP model with firmographic profiles, contact personas, and behavioral signal weights -- with minimal manual input.
Prerequisites
- Connected CRM (Salesforce or HubSpot) with active authentication
- Field mapping configured between your CRM and TrailSpark
- Paid plan (free tier cannot use SparkSense)
- Owner, Admin, or Editor role
Setup Validation
When you select SparkSense Agent, the system checks three prerequisites:
| Requirement | What it checks |
|---|---|
| CRM Integration | Active, authenticated CRM connection (Salesforce or HubSpot) |
| Field Mapping | CRM fields mapped to TrailSpark data structure for accounts and leads |
| Score Destination | At least one destination configured for score delivery |
If anything is missing, SparkSense provides a guided setup flow that walks you through each requirement without leaving the wizard.
Build Mode
After setup validation, you select Build Mode to create a new ICP from scratch. (Conversion Audit mode for refining existing models is planned for a future release.)
Build Mode Steps
The Build Mode wizard has three steps:
| Step | What happens |
|---|---|
| Fetch & Analyze | Agent retrieves CRM data and runs AI analysis on accounts, contacts, and opportunities. You can toggle individual insights on or off. |
| Behavior Signals | You describe positive engagement signals and negative signal patterns in free text (optional -- can be skipped). |
| Review & Create | Review selected suggestions, add free-text refinements, name the model, and create it. |
Step 1: Fetch & Analyze
The agent connects to your CRM and retrieves accounts, contacts, and opportunities. It then runs AI analysis to identify:
- Firmographic patterns -- common industries, company sizes, regions across closed-won deals
- Contact personas -- job title, department, and seniority patterns among decision makers
- Deal patterns -- characteristics that correlate with wins vs. losses
Analysis typically takes 2-5 minutes depending on data volume. Progress updates display in real time. You can leave the page and return -- progress is saved.
Step 2: Behavior Signals
Describe the engagement behaviors that indicate buying intent and poor fit. This step is optional -- if skipped, the model focuses on firmographic and persona matching.
Step 3: Review & Create
The review step shows selected suggestions from the analysis and lets you create the model. You can:
- Review suggestions selected during the Fetch & Analyze step
- Remove individual suggestions before model creation
- Add free-text refinements to supplement the AI analysis
- Name the model and add a description
Click Create Model to generate and save the model. After creation, the model becomes active and lead scoring begins.
Data Requirements
SparkSense works best with sufficient CRM data:
- At least 20 accounts with firmographic fields populated
- At least 10 closed-won opportunities
- Contacts associated with deal records
If your CRM data is sparse, consider Upload & Analyze or Describe Manually instead.
Ensure closed-won deals are properly marked in your CRM and that contact-to-account associations are current before running SparkSense. The quality of the generated ICP directly reflects your CRM data quality.
Next Steps
- Upload & Analyze Mode -- Alternative for teams with CSV exports
- Describe Manually Mode -- For teams without historical data