Latest / The Edge Computing Podcast with Fexingo: Local Compute, CDNs, and Distributed Infrastructure / How Edge Computing Powers Real-Time Retail Geofencing
Transcript
- Lucas: You're walking past a Best Buy. Your phone buzzes — a notification for a discount on noise-cancelling headphones you looked at last week. That notification arrived in under 50 milliseconds. And it never touched the public internet. Luna: Wait — how do you know it didn't touch the internet? Push notifications typically go through a cloud server. Lucas: Normally yes. But more retailers are moving that logic to edge nodes inside their own stores. The geofencing — detecting when your phone enters a virtual boundary around the store — happens locally. The offer selection, the personalization, the send decision — all running on a compute rack in the back office or a closet. Luna: So the store itself becomes the data center. That's wild. What's driving this shift? Lucas: Latency and privacy. A cloud round-trip for geofencing can take 200 to 400 milliseconds. That's fine for a welcome message when you walk in. But for time-sensitive offers — say you're walking past a Starbucks and they want to hit you with a mobile order discount before you cross the street — 400 milliseconds is too slow. You've already moved on. Luna: Right — the window is literally a few seconds. So edge computing gets that down to what, 50 milliseconds? Lucas: Typically under 50. Some systems we've seen hit 10 to 20 milliseconds. The key is that the inference model — deciding which offer to send based on your past behavior — runs on a local GPU. That GPU can handle about 10,000 concurrent sessions. More than enough for a busy store. Luna: And the privacy angle — that's huge. Your location data never leaves the store's network. It's not being sent to a cloud provider. Lucas: Exactly. Retailers are increasingly wary of GDPR and CCPA liabilities. If the data never leaves the premises, they don't have to worry about a breach on a cloud server. The edge node processes everything and only sends back anonymized aggregates — like '42 people redeemed the headphone offer today' — to the central system. Luna: Which retailer is furthest along with this? I know Walmart has been investing in edge infrastructure. Lucas: Walmart is a good example. They've been deploying edge nodes — they call them 'Data Center in a Box' — in distribution centers and some supercenters. But for real-time geofencing, Target might actually be ahead. They started testing local geofencing in about 50 stores in 2024, and by early 2026 they've rolled it out to over 800 locations. Luna: Eight hundred stores with local edge compute for geofencing. That's not a pilot anymore. Lucas: No, it's production. And the results are interesting. Target reported that click-through rates on edge-triggered offers were 3 times higher than cloud-based ones. The obvious reason: timing. The cloud offer often arrives when you've already left the immediate area or opened another app. The edge offer hits you in the moment. Luna: Makes sense. So the hardware — what does an edge geofencing node actually look like? Is it a full server rack? Lucas: It can be. Most retailers use a compact edge server — about the size of a small desktop tower or a pizza box. It has a mid-range GPU, say an NVIDIA A16 or similar, 64 gigs of RAM, and a local SSD. Total cost per store is around $8,000 to $12,000. That's not trivial, but when you consider the cloud compute costs they're displacing, it pays for itself in 12 to 18 months. Luna: And you don't need a dedicated IT person at each store. The edge node can be managed remotely. Lucas: Exactly. That's the model — zero-touch provisioning, remote monitoring. The node self-updates. If it fails, the store just loses geofencing capability temporarily, but the rest of the POS system works fine. So it's low-risk. Luna: What about smaller retailers? A local boutique can't drop ten grand on a server. Lucas: They don't have to. There are now edge as a service offerings. Companies like Edge Impulse and some CDN providers offer a managed edge node for a monthly fee — roughly $200 to $400 per month per location. That includes the hardware, the geofencing software, and the model updates. You plug it into your store's internet and it just works. Luna: So edge geofencing is becoming accessible to any store with a broadband connection. Lucas: Exactly. And the use cases go beyond push notifications. Some stores use it for dynamic pricing on digital shelf labels — when a competitor down the street drops a price, the edge node adjusts the label in real time. Others use it for loss prevention: if a known shoplifter's phone enters the geofence, the system alerts security locally. Luna: That's a bit creepy — but also effective. Does the customer have to have the store's app installed for geofencing to work? Lucas: Usually yes. The geofencing relies on the app requesting location permissions. But there's a newer approach using Bluetooth Low Energy beacons and the edge node to detect devices without an app — just by listening for Bluetooth signals. That's more privacy-sensitive, so retailers are cautious. Most stick with the app-based model. Luna: Got it. So the edge handles the real-time decision, and the cloud handles the analytics later. That hybrid model seems to be the sweet spot. Lucas: Absolutely. And it's worth noting that this is not just for retail. Similar edge geofencing is being used in stadiums, airports, and museums. But retail is where the ROI is most direct — every millisecond of latency saved can mean a conversion. Luna: Speaking of conversions — if listeners are finding value in these deep dives into edge computing, there's a way to support the show that keeps it ad-free and focused on the tech. We've had a few people ask. Lucas: Yeah, and we appreciate that. If today's conversation about latency and local compute gave you something useful, you can support the show at buy me a coffee dot com slash fexingo. That's buy me a coffee dot com slash fexingo. No pressure — just helps us keep doing episodes like this. Luna: And honestly, the fact that we can spend 10 minutes on edge geofencing without shoehorning a sponsor message is exactly why listener support matters. So thank you to anyone who chips in. Lucas: Alright — back to the tech. One more piece of the puzzle: how does the edge node handle personalization without a cloud connection? It needs to have a local copy of the customer profile, or at least a segment model. Target does this by syncing profiles to the edge node overnight via a secure tunnel. During the day, the node uses that local copy to make decisions. If a customer hasn't visited in 30 days, their profile is evicted. Luna: So the edge node is essentially a tiny, ephemeral data center for each store's customer base. That's pretty elegant. Lucas: It is. And it scales beautifully. Each node only needs to know about the customers who frequent that store. A typical urban Target might see 5,000 unique devices per week. That's a small dataset — easily handled by a local database. Luna: What about the model update cycle? If the central marketing team wants to change the offer logic, how does that propagate? Lucas: They push a new model — usually a small ONNX or TensorFlow Lite file — to all edge nodes via a content delivery network. The node downloads it in the background and hot-swaps it without interrupting service. The whole process takes a few seconds. So you can have a campaign go live in all 800 stores within minutes. Luna: That's faster than updating a cloud API endpoint, honestly. Lucas: Often faster, yes. And more reliable because each node operates independently. If the central cloud goes down, the stores keep running. They just can't update the model until connectivity returns. Luna: So edge geofencing gives retailers both speed and resilience. That's a compelling combination. Lucas: It is. And I think we'll see this become standard in the next couple of years. The technology is proven, the cost is dropping, and the privacy benefits are real. If you're a retailer not thinking about this yet, you're probably leaving money on the table. Luna: Great angle. Thanks for breaking it down, Lucas. Lucas: My pleasure. That's all for this episode of The Edge Computing Podcast. We'll be back next time with another look at how distributed infrastructure is changing the world.