CRM-Learning Engine
AI that learns from your deals to improve targeting and messaging.
Overview
The CRM-Learning Engine connects to your HubSpot pipeline and analyzes every closed deal (won and lost). It extracts patterns about which industries, titles, company sizes, and messaging angles lead to wins, and adjusts your scoring and email generation accordingly.
Available on Pro and Scale plans.
How it works
- Connect — Select your HubSpot deal pipeline and map your won/lost stages.
- Sync — Every 6 hours, new closed deals are synced to GetSalesClaw.
- Analyze — When enough deals accumulate (5+), Claude AI extracts predictive patterns.
- Apply — Patterns feed into lead scoring (weighted blending) and email generation (winning angles).
Setup
- Go to
/app/icp-learning - Select your HubSpot pipeline from the dropdown
- Map your "Won" and "Lost" stages (these may have custom names in your CRM)
- Click "Sync now" to start the first sync
- Wait a few minutes for deals to import and analysis to run
What the engine learns
- Top industries — Which industries have the highest win rate
- Top titles — Which decision-maker titles convert best
- Company size — Your sweet spot (e.g. 51-200 employees)
- Best channels — Email vs LinkedIn vs referral performance
- Winning angles — Messaging themes that appear in won deals
- Avoid patterns — Segments with high loss rates
Confidence levels
The engine's confidence grows as it analyzes more deals:
| Deals analyzed | Level | Behavior |
|---|---|---|
| 0-10 | Learning | Uses your manual ICP, collects data |
| 10-20 | Emerging patterns | Suggests adjustments, hybrid scoring |
| 20-30 | Reliable ICP | Auto-scoring based on learned patterns |
| 30+ | Auto mode | Continuous refinement with every deal |
Scoring integration
When the engine has enough confidence (10+ deals), it blends learned scores with the existing AI scoring:
- Final score = 60% existing AI score + 40% CRM-learned score
- Prospects matching winning patterns get a boost
- Prospects matching avoid patterns get penalized
- Personalization hints from winning angles are added to lead files
Email generation integration
When winning angles are identified, they're injected into the email generation prompt:
- The AI uses your proven messaging themes when relevant
- Optimal email length from your best-performing sequences is enforced
- This happens transparently during the Write pipeline step
Data & privacy
- Only deal metadata is stored (industry, title, company size, deal value, stage)
- No email content, documents, or conversation history is stored
- Data is EU-hosted (Hetzner, Germany) and isolated per tenant
- You can delete all learning data from
/app/icp-learning
Limits
- Sync frequency — Runs every 6 hours automatically, or manually via "Sync now"
- Analysis cost — Pattern analysis uses Claude Sonnet (~$0.02-0.05 per analysis run)
- LLM budget — Analysis counts toward your daily LLM budget (Pro: $3/day, Scale: $5/day)
- API rate limits — HubSpot API calls count toward rate limits (throttled at 200ms between calls)
FAQ
How many deals does it need to be useful?
First patterns emerge at 10 deals. The system becomes reliable around 20-30 deals.
Does it work with Salesforce?
Currently HubSpot only. Salesforce is on the roadmap.
What if my ICP changes over time?
That's the point. The engine detects pattern shifts automatically. If your recent wins are in a new industry, scoring adjusts.
Can I override the learned ICP?
Yes. Your manual ICP definition always has priority. The CRM-learning blends with it, it doesn't replace it.