Catalog / Everyday Automation & AI Micro-Tools / AUTO-013

Business card & contact capture into CRM or address book

Photos of business cards, email signatures or trade-fair notes become clean contact records with company, role and follow-up note in the CRM or address book, for sales teams and small offices.

Quick win: proven, manageable effort, first results within weeks, low regulatory risk.

Parts needed to run it yourselfAgents & automationDocument readingLanguage model

After a trade fair, a pile of business cards sits on a desk for weeks, and most contacts never reach the CRM. A phone photo, a capture mailbox or a short note is enough input: an OCR or vision step reads the card, an AI step normalises it into fields and a flow creates the contact after a duplicate check. Typical benefit: contacts are in the CRM the same day, with the follow-up note attached, and nobody types them in.

Problem & value

  • Today: cards sit in a box, are typed in later by an assistant, or never entered. Notes like "offer for 20 licences, call in March" stay in someone's head.
  • AI value: reading a card is solved; the value is the clean-up: splitting fields, normalising phone numbers, spotting existing contacts, attaching note and event tag.
  • Honest scale: a small helper, not a lead engine. Published time savings are mostly vendor claims, so measure your own (see KPIs).

Variants

  • Trade-fair mode: one shared flow, every colleague photographs cards, all contacts get the event tag.
  • Signature capture: forward an email to a capture address; the flow extracts the signature block into a contact.
  • Private use: phone address book with the phone's built-in text recognition.
  • Public sector: network and partner contacts kept in Nextcloud.

How it works

  1. photo / forwarded mail / typed note
  2. OCR or vision model reads the text
  3. AI step maps text to fields (name, role, company, email, phone, address, language)
  4. normalise (phone format +49, title case, website) and validate (email syntax, domain)
  5. duplicate check against CRM / address book
  6. human review screen (confirm, edit, merge)
  7. create or update contact + note + event tag
  • Reading: a prebuilt card model (for example the AI Builder business card reader in Power Automate), a CRM mobile scanner, or a vision-capable LLM that returns JSON against a fixed schema.
  • Mapping: a small LLM or the flow tool's extraction node resolves ambiguous lines (title vs. department, bilingual cards).
  • Human-in-the-loop: the person who took the photo confirms the record on the phone. Nothing is written silently, and no email is sent automatically.

Data preparation

Which data: the target field list of the CRM (required fields, picklists, owner), existing contacts for the duplicate check, and a consent or source field.

Where it lives: CRM (HubSpot, Zoho, Dynamics 365, Salesforce), Outlook or Google contacts, Nextcloud Contacts, or Excel.

Typical gaps: duplicates, inconsistent company names ("Müller GmbH" vs "Mueller GmbH & Co. KG"), missing owner and source fields, no rule on the legal basis.

Preparation steps: 1. Decide the minimum fields and the picklists (event, source, owner). Half a day. 2. Export the current contacts and remove obvious duplicates (PREP-011); one to two days for a few thousand records. 3. Add "source" and "capture date" fields. 4. Write a one-page rule on what may be captured and how long it is kept (PREP-025). 5. Test with 20 to 30 real cards, including bad lighting.

Tools: Excel or Power Query, the CRM import and dedupe features. Effort is low and mostly contact clean-up. See GUIDE-003.

Implementation paths

Which path for whom

  • SMB: Path A. Mature scanner features in CRMs and phone apps exist at SMB prices and the process is standard. Use B only if you need a custom CRM or a strict review step, C if you want zero cloud for contact data.
  • Enterprise: Path B. One shared flow (Power Automate or n8n) with the central CRM and an EU-region model instead of private apps.
  • Developer: Path B. Small flow with a vision-model API and a webhook into the CRM.
  • Public: Path C. Keep contacts in Nextcloud or a sovereign cloud; local OCR and a small model suffice.
  • CRM-native scanners: HubSpot and Zoho both offer card scanning in their mobile apps; Zoho Scanner exports to Zoho CRM [1]. Dynamics 365 users can use the AI Builder business card reader [2]. EU hosting: HubSpot, Zoho and Microsoft offer EU data centres or regions, but confirm the region of your own account and sign the AVV.
  • Standalone scanner apps: Blinq, Wave Connect, BizConnect and similar sync to HubSpot or Salesforce, usually in paid tiers [1]. Many are US vendors: check data location first.
  • Choose A when: your CRM already has a scanner and you need no special field logic.

Path B: Build on cloud APIs

  • Flow tool: Power Automate (cloud flow with the business card action [2]), or n8n with its Information Extractor node for free-text signatures [3].
  • Model: a vision-capable model from an EU region (for example Azure OpenAI in an EU region, Mistral from France, or Google Gemini with EU processing) with a JSON schema for the contact fields.
  • Integration: CRM API or Nextcloud CardDAV (vCard 3.0 or 4.0 only [4]).
  • Effort: one to three days with review step.

Path C: Self-host (open source)

  • Stack: n8n self-hosted, PaddleOCR or Tesseract for reading, a small local model through Ollama (a model of 7 to 8B parameters is enough for field mapping), Nextcloud Contacts as target.
  • Hardware: HW1 (CPU server) is enough for a few hundred cards per event.
  • Choose C when: contacts are sensitive (lawyers, doctors, public bodies) or n8n and Ollama already serve flows such as AUTO-007 and AUTO-008.

Path D: Custom training

Not worth it. Card layouts are easy for current models.

State of the art (2026)

  • Prebuilt card readers and vision models read printed cards well; fields like email and phone are reliable, while titles, department lines and unusual layouts are the weak spots.
  • Limits: stylised fonts, dark backgrounds, vertical text, handwritten notes, and non-Latin scripts. Models may invent a missing field (for instance a domain-based email address). Always show the original photo next to the extracted data.
  • The prebuilt Microsoft card reader was reported in 2024 to support only English and Japanese; check the current documentation before relying on it for German cards [5].
  • Web enrichment of the record is a different use case (AGENT-011).

Shared resources & combinations

  • Same stack as AUTO-007 (attachment filing), AUTO-008 (forms to rows) and DOC-003 (form extraction): one n8n or Power Automate environment, one extraction pattern.
  • Next steps: AUTO-014 (follow-up mail merge), AUTO-026 (photo to table), AUTO-025 (voice notes); AGENT-011 keeps contacts clean.

Effort, cost & KPIs

  • Path A: hours to a day; often included in the CRM licence.
  • Path B: one to three days build, low API cost (measure your own).
  • Path C: two to four days including n8n and Ollama, then little maintenance.
  • KPIs: share of cards in the CRM within 48 hours, share of records needing correction, duplicate rate, follow-up contact rate after events, minutes saved per contact (time a sample).

People & change

  • Involve: the sales or marketing lead (process owner), event attendees (end users), IT for accounts and CRM, the DPO for legal basis and retention; management approves the tool.
  • Works council: usually not triggered, as the flow processes third-party contacts. It applies if capture counts are reported per person (§87(1) no. 6 BetrVG; inform under §90(1) no. 3). See GUIDE-004.
  • Data protection: business contacts are personal data. Check the legal basis, sign an AVV with every tool (Art. 28 GDPR), keep a source field. A DPIA is normally not needed at this scope; ask the DPO. See GUIDE-005.
  • AI literacy: AI Act Art. 4 requires measures that support staff AI literacy; a short briefing on the review screen is enough. See GUIDE-006.
  • Pilot: one event, 2 to 3 colleagues; people stay responsible for the record. See GUIDE-008, GUIDE-009.

Risks, regulation & compliance

  • EU AI Act: minimal risk; no Art. 50 duty for pure extraction. See GUIDE-014.
  • GDPR: information duty (Art. 14) towards the card owner, usually by a line in the first email; erasure on request; no photos left in phone galleries or private cloud backups. Provide an approved app (GUIDE-007).
  • Risks and mitigations: wrong fields (review screen), duplicates (check before create), shadow scanner apps with US storage (approved tool), prompt injection in forwarded signatures (text is data, no automatic actions; GUIDE-013), marketing without consent (keep consent status separate).

Getting started

  1. Pick the target (CRM or Nextcloud) and write the minimum field list with an event tag.
  2. Test the built-in scanner of your CRM with 20 real cards; note the error types.
  3. If it falls short, build a Power Automate or n8n flow with a review step (one to two days).
  4. Check the AVV, data location and retention with the DPO before the next event.
  5. Pilot at one event, count corrections, then roll out (GUIDE-009).

Sources

  1. Comparison of business card scanner apps and CRM integrations (HubSpot, Zoho, Blinq, Wave Connect, BizConnect) - https://wavecnct.com/blogs/best-business-card-scanner-app
  2. Microsoft Learn - Use the business card reader prebuilt model in Power Automate - https://learn.microsoft.com/en-us/ai-builder/flow-business-card-reader
  3. n8n Docs - Information Extractor node - https://docs.n8n.io/integrations/builtin/cluster-nodes/root-nodes/n8n-nodes-langchain.information-extractor/
  4. Nextcloud documentation - Contacts / CardDAV - https://docs.nextcloud.com/server/stable/admin_manual/groupware/contacts.html
  5. Inogic - Prebuilt business card reader model in Microsoft Power Automate (2024) - https://www.inogic.com/blog/2024/04/utilize-the-power-of-prebuilt-business-card-reader-model-in-microsoft-power-automate/