The card should fill the frame
Optical character recognition works on pixels of text. A card photographed from across a desk, occupying a third of the frame, gives the engine perhaps forty pixels of height for a name printed in nine-point type. That is not enough, and no amount of processing afterwards recovers detail that was never captured.
Get close enough that the card's edges nearly touch the edges of the frame. This single change fixes more scanning problems than anything else you can do.
Watch the background, not just the card
A cluttered background is not neutral. Text on a laptop screen, a poster or a keyboard behind the card competes for the engine's attention, and words from the background can end up in your contact record.
Put the card on a plain surface if you can. A table, a notebook, even your palm — anything without text on it.
Light beats resolution
Even, diffuse light matters more than megapixels. Direct overhead light on a glossy or foil-stamped card produces a reflection that erases whole lines of text.
Tilt the card slightly to move the glare off the print, or step away from the light source. If you can read every line comfortably on your screen before you capture, the scanner can too.
Always check before you save
Treat extraction as a first draft. Unusual layouts, vertical text, logos rendered as words and non-Latin scripts all trip up even good engines.
Ten seconds of correction at the moment of capture is worth an hour of confusion in six months, when you no longer remember whether that phone number was theirs or their office's.
