Stop bolting AI onto your old CRM

We ran the experiment. For six months we sold an "AI sidekick" for Salesforce. Customers loved it. Then we measured the actual productivity lift and it was 1.2x at best.

The ceiling came from three structural issues:

1. Read-only by default. Agents could suggest updates but couldn't make them without a human round-trip. So humans became the bottleneck on every update.
2. Object inflexibility. Adding a custom field for "objection cited on call" required an admin. Agents that can't change the schema can't grow.
3. No durable memory. Every chat session started cold. Compounding learning was impossible.

Moving the database — yes, doing a real migration — unlocked all three. The new ceiling is closer to 4x.

The migration playbook:

• Week 1–2: Export your accounts, contacts, deals, and notes as CSV.
• Week 3: Run them through Dench's bulk import. Schema fields auto-create.
• Week 4: Run shadow operations: writes go to both systems, reads from Dench.
• Week 5: Cut over. Salesforce becomes archive.
• Week 6: Cancel Salesforce.

The one-time cost is real (~80–120 person-hours for a 250-seat org). The ongoing savings start the day after cut-over.