Why the Smartest Buildings Won't Have More Devices in Them

Ask an IoT team how to make a physical space smarter and the answer usually starts with a bill of materials. Devices to mount. Power to run. A network for them to join. A maintenance schedule someone has to own. That instinct built this industry, and for a great many problems it is still the correct answer.

But it carries a cost that rarely survives contact with a real deployment, and in large physical environments that cost has become the thing standing between a working pilot and an actual rollout.

Call it the install tax.

The Pilot Works. The Rollout Doesn't.

The install tax is not the price of the hardware. It is everything downstream of the hardware.

It is the site survey. The electrical work. The union labor. The property manager who needs a change request. The device that fails in month fourteen in a location nobody wants to reach with a ladder during business hours. The second site that turns out to be laid out differently than the first. The finance team asking why a capital project is required to answer a question that feels like it should be software.

None of this shows up in a pilot. A pilot is one site, high attention, motivated staff, and a vendor in the building. That is exactly why so many IoT programs look excellent at ten locations and stall at five hundred. The technology was never the constraint. The install was.

I have watched this pattern play out across enterprise retail for years, and I would argue retail is the clearest place to study it, because retail runs the largest, messiest, most frequently rearranged indoor environments in commercial use. If a deployment model survives a thousand grocery stores, it will survive almost anything.

What Is Actually Missing Is Not More Detection

Here is the part that took me a long time to say clearly.

The problem in most large physical spaces is not that we cannot detect things. Detection has been more or less solved for a decade. Cameras see. Sensors sense. Scanners scan. Signals are everywhere.

The problem is that almost none of it is spatially grounded, which means almost none of it can be acted on.

Consider what a large retailer already knows. It knows every item it sells. It knows the price of each one. It knows, with reasonable confidence, how many units are somewhere in a given building. What it does not know, in any structured, machine-readable form, is where any of it physically is. Not approximately. Not last quarter. Right now, on a floor rearranged three weeks ago by a team that didn't file anything.

So the software sitting on top of all that data is running blind. An app knows a shopper's list but not the aisle. A task system knows the job but not the distance. A picking system knows the order but not the route. Each of those gaps gets absorbed by a human who compensates with memory and walking.

That is what I mean when I say the physical space is not machine-readable. It is not a detection failure. It is a missing layer.

The Counterintuitive Move: Use What the Building Already Emits

Most teams reach for hardware because they assume the space contains no usable signal. It contains an enormous one.

Every building distorts the earth's magnetic field in a way that is specific to that building. Steel, concrete, structure, and shape produce a pattern that is stable over time, unique to each location, and readable by the magnetometer that already sits inside every smartphone in the world. You don't install the sensor. Your associates and customers already carry it.

Once you accept that, the deployment model inverts. Instead of instrumenting the space, you read it. Nothing to mount, nothing to power, nothing to fail in month fourteen, and nothing to renegotiate with a landlord.

We run this across more than 850 million square feet and over 6,000 locations globally, accurate to within a meter, with zero hardware installed in any of those buildings. The number I find more telling is different: we have activated more than 300 sites in a single week. That number is only possible because there is no install step. It is not a technology benchmark. It is a deployment-model benchmark, and deployment is where most IoT programs actually die.

Why This Matters Beyond One Vertical

I am describing retail because that is where we work, but the underlying argument is not a retail argument.

Every conversation happening right now about AI in the physical world runs into the same wall. Agents, automation, robotics, and predictive systems all assume the environment they operate in is legible. Indoors, it usually is not. You can hand a model perfect inventory data and perfect task data, and it will still produce instructions that ignore the fact that two things sitting in the same database are two hundred feet apart.

Hospitals have this problem. Warehouses have it. Airports, campuses, manufacturing sites, and any large facility where people and assets move have it. The spatial layer is the input those systems are missing, and adding intelligence on top of a space you cannot read only produces confident, expensive guesses.

The Question Worth Asking

If you are scoping an indoor deployment right now, I would push on one question before the architecture discussion starts.

It's not, "What should we install?" Ask instead, "What signal does this environment already produce that we are not reading?"

Sometimes the honest answer is that you genuinely need hardware, and you should install it. But the default assumption that physical context requires physical infrastructure is worth interrogating, because it is the assumption that turns a promising pilot into a capital project, and a capital project into a program that quietly does not scale.

The smartest buildings of the next decade will not be the ones with the most devices in them. They will be the ones whose software finally knows where it is.

Mickey Balter is Co-Founder and CEO of Oriient, an indoor positioning platform that operates entirely in software with no in-store hardware. He has spent the last decade working on why physical environments remain unreadable to the systems built on top of them.