Business Cards to CRM Records, Automatically

The challenge

Manual entry was the bottleneck, and it failed silently. Nobody could tell which cards had been entered and which were in a drawer, so the leads that were never entered were also never missed. The cost was invisible, which is why it persisted.

The approach

1. 01

   Made capture a photo, because anything that needs an app installed at a trade show stand does not get used.

2. 02

   Ran OCR over the image and extracted name, company, role, email, phone, and website into structured fields rather than one blob of text.

3. 03

   Enriched the contact against company data, so the record arrives with more than what fits on a card.

4. 04

   Created the CRM record, categorised it, notified the owner, and queued the first follow-up in the same run.

5. 05

   Routed anything with low extraction confidence to a human review queue instead of writing a half-read record into the CRM.

Architecture

What was hard

• OCR confidence is not optional metadata. A card photographed at an angle in bad light produces a plausible-looking wrong email, and a wrong email in a CRM is invisible until a bounce three weeks later.

• Cards are inconsistent by design. Two phone numbers, a QR code, a job title in a second language, and no clear company name is a normal card, not an edge case.

• Enrichment has to fail softly. When the company lookup returns nothing, the record still has to be created, because a missing firmographic is not a reason to lose the contact.

Related

• Workflow automation
• AI lead pipeline