Why we don't use facial recognition for conference photos

When we started building Pictag.IO, the first thing everyone asked was "why not just use face recognition?" It's the obvious answer. Every other platform does it. Memzo, Photier, FotoOwl, all of them.

We looked hard at face recognition. We nearly built it. Then we didn't. Here's why.

The GDPR problem is real, not theoretical

Face templates are biometric data under Article 9 of GDPR. That puts them in the same legal bucket as medical records and fingerprints. To process them lawfully at a UK or EU conference, you need explicit consent from every single person whose face you scan.

Not implied consent. Not a checkbox in the event terms. Explicit, informed, freely given consent, with the ability to withdraw it at any time.

We talked to a data protection lawyer about this. Her response was essentially "you can do it, but your customers will hate the paperwork and one complaint can turn into a regulator investigation." We decided we didn't want to be in that business.

Our competitors do operate in the EU, which tells you the risk is manageable if you put the work in. It also tells you they're putting in that work, and passing the cost and complexity to the organiser.

Attendees don't actually want this

A lot of conference attendees are uncomfortable with their employer's face scan ending up in a third party database. Especially in regulated industries. Especially in Europe.

We heard the same thing from three different events we talked to early on: "our attendees would not be happy with a selfie wall." Executives at financial firms. Healthcare professionals. Government staff. These are your best-paying customers at B2B conferences, and they're the most privacy-conscious.

A badge marker is different. It doesn't feel invasive. Nobody has ever objected to a small square on their lanyard.

Face recognition fails in the places you need it

The demo videos always look great. In reality:

• Group photos with 30 people crammed together produce unreliable matches
• Side angles and profiles drop accuracy a lot
• Masks, glasses, bad lighting, all degrade results
• The person facing away from the camera gets nothing, ever

Markers work at any angle the badge is visible. Back of the head shot? If the badge is clipped to the backpack, we catch it. Someone leaning over a table? We catch it. The marker doesn't care about facial expression or lighting direction in the same way a face embedding does.

The economics get weird

Commercial face recognition APIs like AWS Rekognition charge per face detected. A conference with 5,000 attendees and 10,000 photos can burn through a serious bill just on processing. Running your own face embedding model locally is possible but adds GPU costs and infrastructure headaches.

Markers cost us almost nothing to detect. A CPU can process hundreds per second. That difference lets us keep pricing simple and predictable.

The opt out is trivial

Someone doesn't want to be in the photos? They peel the marker off their badge. That's the entire opt out flow. Compare that to face recognition, where opting out means "please find and delete every photo I appear in, plus the face template you built from my other photos, and confirm when you're done."

We liked the simpler version.

What we gave up

There are real downsides. The badge marker has to be visible in the photo, which means candid shots of someone sipping coffee with their badge hidden under a jacket don't match anyone. Face recognition would catch those. We accept that tradeoff.

We also have to ship physical markers or printable templates to organisers. That's a real friction point in the sales process. Face recognition products just say "upload photos" and that's a much shorter pitch.

Would we ever add it?

Maybe. As an opt in feature, where the organiser actively decides to turn it on and takes responsibility for the consent flow. Probably never as the default.

For now, if you want to share conference photos with attendees without a GDPR headache, we think the marker approach is the better trade. You might disagree, and that's fine. There are plenty of face recognition options out there.

Try Pictag.IO for your next event.