To digitize business cards you photograph or scan the card, let software extract the printed fields, then check and correct them before saving. Extraction is the easy part. Most methods fail afterwards — the fields that matter most were never printed on the card, and the ones the software infers often get silently dropped.
TL;DR
- Digitizing a card is roughly seven taps and 15-20 seconds on a clean card. Extraction is not the bottleneck and hasn't been for years.
- A card scan returns five fields: name, title, company, phone, email. That is everything a card physically contains.
- The fields you actually need six months later — how you met, what they asked for, whether it went well — are not on the card and cannot be extracted from it.
- Every stage after extraction is where digitizing fails: fields that need manual correction, values silently discarded, and information that never reaches storage at all.
- I found both failure types in my own system by testing it. They are documented below with the specifics.
How do I digitize a business card?
Here is the actual path in the system I built, counted honestly.
Open the menu, tap Quick Add, tap the CARD tab (it opens there by default), tap "Scan a card photo," the camera opens, tap the shutter, confirm the photo. The model reads it in about two to three seconds with no tap from me. The extracted fields appear for review. Tap Save.
That's five taps on a clean card, plus one for Save, plus however many it takes to correct a field. Realistic total on a clean card: around seven taps, 15 to 20 seconds start to finish.
On a messy card, add another 15 to 30 seconds of correcting.
Any competent scanner app is in this range. If someone is selling you on speed of capture, they're selling you a solved problem.
What does a business card scanner actually extract?
Five fields. Name, title, company, phone, email.
That is not a limitation of the model — it's a limitation of the card. A business card is a printed rectangle with contact details on it. It does not contain how you met, what you talked about, what the person is looking for, or whether you'd want to work with them.
Which is why I built a second input path. Instead of a card, you describe the person in a sentence or two, and the parse returns twelve fields rather than five: name, title, company, phone, email, industry, traits, how you met, a vibe rating, a summary, what they're asking for, and what they're offering.
- Fields returned
- Card scan: 5
- Describe it in a sentence: 12
- Where the data comes from
- Card scan: The printed card
- Describe it in a sentence: What you say about the encounter
- Time to capture
- Card scan: 15-20 sec (clean card)
- Describe it in a sentence: 30-45 sec to compose the note
- Parse time
- Card scan: ~2-3 sec
- Describe it in a sentence: ~2-3 sec
- Captures how you met
- Card scan: No
- Describe it in a sentence: Yes
- Captures what they need from you
- Card scan: No
- Describe it in a sentence: Yes, when it survives the save
- Best for
- Card scan: A card in your hand, right now
- Describe it in a sentence: Anyone you met without a card, or a conversation worth keeping
The card path is faster. The sentence path is the only one that captures the part you'll actually be missing later. That trade is the whole argument of this article, and it's why "digitize your cards" is worse advice than it sounds. The full six-method comparison is in how to organize business cards.
What do I have to fix by hand every time?
Four things, in descending order of how often.
Industry — basically every time. The card schema doesn't have an industry field at all, because a card never states one. On the sentence path the model does infer it, and infers it well; in one test it correctly derived "Logistics" from context. But my app only keeps the value if it exactly matches an eleven-item fixed list, and the prompt never tells the model what that list contains. So a correct answer gets silently discarded and the field displays as a dash. That's my bug, not the model's.
Preferred name — never filled. There's no way to extract it. A card never says what someone likes to be called. The interface says so in its own copy, which is at least honest.
Name and title cleanup — roughly one card in three or four. Real cards contain nicknames in quote marks and stacked job titles. A name printed as Jonathan "Jonty" Reyes-Whitfield comes through verbatim as a single string, quotes and all. A title reading Co-Founder & Creative Director / Interim Ops Lead arrives intact and needs splitting by hand.
Phone punctuation — near-certain on US-format numbers. The server strips hyphens and spaces but not parentheses, so a card printed as +1 (415) 555-0192 saves as +1(415)5550192. Minor, and it still needs a touch.
None of these are extraction failures. The model read the card correctly in every case. They're failures of what happens to the data afterwards — which is where digitizing actually breaks.
What happens to the things a card could never hold?
This is the part that changed how I think about the whole category, and it's a defect in my own system.
I described a contact who had told me two specific things: he was looking for a warehouse operations software vendor, and he wanted an introduction to someone in freight tech. The parse handled it correctly — those came back as structured "asks," properly separated, exactly as designed. The database has dedicated tables for asks and offers. There's an endpoint that accepts them.
And then the save routine never reads them off the parse response before building the payload. They don't get stored, they don't get folded into the summary text, they simply cease to exist. Anything of the form "she needs X, follow up with Y" is unrecoverable once you hit Save.
So my system extracted the single most actionable thing anyone said to me that week, and threw it away, and told me nothing.
I'm publishing that because it generalises. The interesting failures in digitizing contacts are almost never at the reading stage. Reading text off a card is a solved problem and has been for a while. The failures are in the plumbing between extraction and storage — enum mismatches that discard valid values, fields the schema has no room for, and payloads assembled without every piece the parser found. Those failures are silent by construction. Nothing errors. The record saves. You find out months later, when you go looking for something you were certain you'd captured.
If you're evaluating any tool for this, that's the thing to test. Put in something specific and useful. Save it. Come back the next day and see whether it's still there.
The gap between what a system extracts and what it keeps is why I built Habitus Mind the way I did: every person gets one record, every later encounter merges into the same dated history rather than starting a second entry, and you set a follow-up date yourself while you still remember why. It doesn't interpret anyone. It extracts what's there and files it — and where it fails at that, as above, I'd rather write it down than not.
Habitus Mind is invite-only and admin-approved. There's no public signup. Request access.
Which is faster, scanning or describing?
Scanning, comfortably — for a card you're holding.
A clean card scan is 15 to 20 seconds end to end, including the two to three seconds of parsing and a glance to check the fields. Composing a real three or four sentence note about someone takes 30 to 45 seconds, because the writing is the work; the parse afterwards costs the same two to three seconds either way.
So if the card is in your hand, scan it. The sentence path isn't competing on speed and never was. It exists for the person you met without a card, and for the conversation worth keeping — the part that was never printed on anything.
FAQ
How do I digitize business cards?
Photograph the card, let the software extract the printed fields, check them, then save. On my own system that's about seven taps and 15 to 20 seconds for a clean card. Extraction takes two to three seconds and is rarely the problem.
What information can be extracted from a business card?
Five fields: name, title, company, phone and email. That's what a card physically contains. Industry, how you met, what the person needs and whether the conversation went well cannot be extracted from a card, because they were never printed on it.
Can I scan a business card into Excel?
Yes — most scanner tools export to CSV, which opens in Excel. The limitation isn't the export. A spreadsheet of five printed fields per person is a tidier address book, and it still won't tell you why any of those people mattered.
Why do business card scanners get names wrong?
Usually they don't. They copy the card accurately, including nicknames in quote marks and stacked multi-role titles, because that's what's printed. Roughly one card in three or four needs the name or title split or trimmed by hand afterwards.
What's the biggest problem with digitizing business cards?
Silent data loss after extraction. Values get discarded by schema mismatches or dropped between parsing and saving, with no error shown. I found both in my own system. Test any tool by saving something specific and checking the next day that it's still there.