On-Premise AI Lead Management: Business Cards to Prioritised Pipeline the Same Day
Event leads arrived as stacks of business cards, typed in days later. Our on-premise system digitises cards with OCR, captures conversations, prioritises leads and drives follow-up campaigns.
- Service
- AI Business Applications
- Industry
- Software & SaaS
- Region
- Japan
- Engagement
- Lab-type ODC (dedicated team)
Outcomes
- Same-day lead records
- OCR for JP/EN business cards
- Priority-driven follow-ups
Stack
On-premise · OCR · Speech-to-text · Python · Email automation
Written byOriental Nippon Team
A Japanese software company sells after-sales systems to car dealers. Much of its pipeline comes from trade events and field visits, where what you bring home is a stack of business cards and your memory of the conversations.
Its data-handling policy ruled out hosted lead-management tools, which removed most of the market.
The problem
- Cards and memories. Event leads arrived as paper and recollection.
- Late transcription. Contact details were typed in days later, after the context had faded.
- No prioritisation. Nothing separated a lead to call on Monday from one for next quarter.
- Manual campaigns. Follow-ups were assembled by hand, so they went out late or not at all.
- On-premise only. Hosted tools were not allowed.
| Before | After | |
|---|---|---|
| Business cards | Typed in days later | OCR into a contact record |
| Conversation notes | From memory | Transcribed, fields filled automatically |
| Follow-up order | Chronological | Priority-tagged |
| Hosting | n/a | Entirely on the company's own servers |
What we built
An on-premise lead-management system
Running entirely on the company's own infrastructure.
Business-card digitisation
OCR turns a card into a contact record, tuned to the languages and layouts the company actually sees.
Interaction capture
Transcription with automatic field population and a review step before saving.
Priority tagging and campaigns
An ordered follow-up list and bulk email campaigns driven by it.
Technology and the role it played
| Technology | Role in this project |
|---|---|
| On-premise deployment | Everything runs on the company's own servers |
| OCR | Turns business cards into contact records |
| Speech-to-text | Captures conversation context while it is fresh |
| Python | Extraction, tagging rules and campaign logic |
| Email automation | Bulk follow-up campaigns from the priority list |
How we delivered it
Our ODC team built the contact and interaction model first, because everything else writes into it. Then we tuned the OCR against real cards collected at the company's own events.
- Set the infrastructure boundary. And size the on-premise deployment.
- Build the data model. Contacts and interactions.
- Card digitisation. Tuned on real cards.
- Interaction capture. With a review step before a record is committed.
- Prioritisation and campaigns. Tested on a real event's leads.
01Infrastructure and data model
2–3 wks
On-premise deployment, contact and interaction structures
02Card digitisation
3–4 wks
OCR, field extraction, real-card tuning
03Interaction capture
3–4 wks
Transcription, auto-fill, review step
04Prioritisation
2 wks
Tagging rules, ordered follow-up list
05Campaigns and pilot
2–3 wks
Bulk outreach, tested against one real event
Risks and how we handled them
| Risk | How we handled it |
|---|---|
| Data leaving the company | On-premise only, with the infrastructure boundary set first |
| OCR errors on real cards | Tuned on cards from the company's own events |
| Bad data saved automatically | A review step before any record is committed |
Outcome
Leads from an event become working records the same day, ordered by priority, with follow-up going out as routine. All of it runs on the company's own infrastructure, which was a requirement, not a preference.
We cannot yet claim the pipeline converts better. What we can say is that the gap between meeting someone and contacting them has closed.
What the client keeps
- The system and its source
- On-premise deployment configuration
- Extraction and tagging rules
- Campaign templates
Sources and further reading
- Tesseract OCR, tesseract-ocr on GitHub
- AI Risk Management Framework, NIST
Facing something similar? This is the kind of work our AI, Data Science & Automation team delivers every day. Talk to us about your stack.
