Where a face is processed shapes your latency, your breach surface, your data residency and your privacy posture. An honest look at the trade-offs — and why Ottica runs everything on a local server.
Most conversations about facial recognition focus on the model — how clever it is, how accurate it scores. But for a venue, an equally consequential decision is architectural and almost invisible: where does the processing actually happen? A face captured at your door can be analysed on a local server in your back room, or it can be packaged up and sent to a data centre somewhere else, and that single choice ripples through latency, privacy, cost and reliability. This is not a one-sided argument — cloud architectures have real advantages, and it is worth being honest about the trade-offs rather than pretending the answer is obvious. But for biometric workloads in Australian venues, the balance lands firmly on-premise, and it is worth understanding why.
Facial recognition for self-exclusion or access control is a real-time task. A patron is at the door for a few seconds; a useful alert has to arrive inside that window. On-premise processing keeps the entire loop local, so a match can be flagged in well under a second, whereas a cloud round trip adds network latency, depends on a stable internet connection and introduces a point of failure that has nothing to do with the algorithm's quality. For occasional, non-urgent analysis that delay is irrelevant; for a discreet alert that has to reach a staff member before someone reaches a gaming machine, it is the difference between a system that works and one that is merely interesting.
Every place biometric data travels to, or rests in, is a place it can be exposed. Sending faces to the cloud means the data crosses networks and sits on infrastructure outside the venue's direct control. That expands the breach surface and the number of parties who could, in principle, access or be compelled to produce the data. Processing on-premise keeps the most sensitive information you handle inside the four walls you already secure.
The most defensible biometric system is one where the sensitive data simply never leaves the site it was collected on.
Biometric data is among the most sensitive information a venue handles. Cloud deployments raise questions about where data is physically stored and whether it crosses borders. On-premise processing removes most of that complexity: if the data never leaves your venue, data residency is simply not in question and your accountability maps cleanly to hardware you control. When you can state plainly that no biometric data leaves the site, several of the hardest risk questions answer themselves.
Cost is where the honest trade-offs are clearest. Cloud processing often means lower upfront spend but recurring per-camera or per-query fees that scale with usage, plus dependence on bandwidth; on-premise means investing in a local server, but with predictable ongoing costs and no metered charges for every face analysed, which over the life of a deployment a busy venue frequently finds more favourable. Reliability follows a similar logic — a cloud system is only as available as the venue's internet connection and the provider's uptime, whereas an on-premise server keeps working through an outage, which for a safety-critical function like self-exclusion enforcement is not a small consideration. To be fair to the cloud, it genuinely shines for some workloads: elastic scale across many sites, rapid model updates pushed centrally, and analytics that aggregate data where real-time response is not required. The point is not that on-premise always wins; it is that the right answer depends on the workload, and biometric venue safety is one that favours keeping things local.
Ottica is built around a local server. Faces are processed on-premise, in real time, on Ottica's smart cameras; only enrolled individuals are matched, and people who are not enrolled are never identified or recorded. Because the system is Australian-made and engineered in Melbourne, and because the data never leaves your venue, the architecture is designed to make latency, residency and breach-surface concerns straightforward rather than something to manage after the fact. Where you process a face is not a technicality — it quietly sets the ceiling on how fast, how private and how reliable the whole system can be, and for venues handling biometric data in Australia, processing it on hardware you control is the choice that keeps the most options open and the most risk off the table.
Self-exclusion only protects people if it is enforced. Here is how facial recognition matches a face against a self-exclusion list at the door — and why an on-premise design keeps the program privacy-first.
Read insightA single accuracy percentage tells you almost nothing. Here is how to read independent face-recognition evaluations — verification versus identification, FMR and FNMR, and demographic differentials — and what to actually ask a vendor.
Read insightBook a private demo and we will show you Ottica AI working on real venue scenarios — entirely on-premise, with nothing leaving your site.