Lead scoring without sending data to OpenAI

Why we use TensorFlow.js for in-browser ML, and what it can and cannot do.

Modern AI features usually mean uploading customer data to a third-party API. For a CRM, that is a privacy footgun: you are exporting the most sensitive information your business owns.

OrbitCubs CRM runs lead scoring in the browser, on your device, using TensorFlow.js. The model takes signals — recency of contact, deal size, engagement count, source channel — and returns a score from 0 to 100. The data never leaves the browser.

The trade-off is model size and complexity. We are not running GPT-4. The model is a small gradient-boosted regressor trained on synthetic CRM data; you can fine-tune it with your own historical conversions through the admin panel. For the typical "is this lead hot" question, it is more than enough.

For deeper analytics — sentiment on call transcripts, multi-document summarization — you can opt in to a server-side model later. But the default is private.

Published 2026-04-15 by OrbitCubs Team.

OrbitCubs CRM — Lead scoring without sending data to OpenAI