One Pattern, Three Industries.

A timestamp and a value. That's the entirety of what comes off an enterprise handheld scanner. A timestamp and a lat/long. A timestamp and a battery reading. Millions of records per day, across thousands of devices, spread across warehouses, retail floors, and distribution centers.

Three of the client's largest customers saw that data and made the same request: just give it to us raw. Walmart said it. FedEx said it. Lululemon said it. Each had data science teams. Each had analytics infrastructure. Each assumed that access to the data was the hard part.

It wasn't.

THE PROBLEM

Our client built its business on hardware. A hardware and software company serving retail supply chain and distribution, their devices were excellent and the customer relationships were strong. But the sales motion was stuck in a product catalog.

When customers asked for data access, the client said yes. That was the easy answer. The harder question, which nobody was asking yet, was: data to do what, exactly?

The raw event streams were not interpretable at a glance. A device ID, a timestamp, a coordinate. Another device ID, a timestamp, a battery percentage. Cross-reference enough of these and something might start to make sense. Maybe. But six months of analyst time and a lot of assumptions would be required to get there.

The client didn't yet have the language to explain why their customers needed more than raw access. They knew their devices. They didn't yet know how to position themselves as the only people who actually understood what their device data meant.

We came in as data product and go-to-market advisors. A team of two, embedded alongside the client for a year starting in late 2019. The goal was to understand the event data well enough to build pilots, then translate that into a repeatable motion the company could take to any account.

WHAT WE FOUND

Three customer pilots. Three completely different operational contexts. The same underlying data problem in every one.

Walmart's use case centered on handheld devices used for RFID scanning across store operations: moving items from storage, stocking shelves, running inventory. The business was buying replacement devices on a regular cycle because it assumed lost devices were gone. It had no visibility into whether a device was sitting in a locker, abandoned in a break room, or walking out the door in an employee's pocket. These were smartphones in a ruggedized case. There was real financial incentive to take one home, factory reset it, and put it to personal use.

FedEx was running a similar device fleet across its warehouse and distribution operations, with camera sensors added at loading docks pointed at trucks. The business problem was loading efficiency: were trucks being loaded on schedule, was volume being packed correctly, and were crews in the right place? The cameras generated surface area and volume readings at regular intervals. The handhelds generated location pings throughout shifts.

Lululemon's setup was different in form but identical in structure. Security cameras and Wi-Fi routers across retail stores. Every device in the store, customer or associate, generated a MAC address ping with a timestamp and a location coordinate. The question was about store performance: were associates deployed effectively, were high-value sections of the store getting appropriate attention, and could any of that be tied to sales?

In all three cases, the customer had looked at the raw feed and concluded they could work with it. In all three cases, the answer was: not without knowing what the data means.

WHAT WE BUILT

The foundation was the same across all three. We modeled the event stream into something interpretable: a semantic layer that translated device behavior into business states.

For Walmart, that meant building a device status model on top of battery telemetry and location data. A device that hadn't moved in an extended period and was showing a stable battery pattern wasn't lost. It was sitting somewhere idle. That insight alone changed the procurement conversation. Alerts could go out before a device was ever declared missing. Geofencing logic flagged when a device left the store perimeter, which surfaced both theft risk and a more useful angle for sales: you don't actually need more devices. You have devices that haven't moved in weeks.

Software version data added another layer. Cross-referencing which version of the client's operating software was running against battery performance patterns made it possible to draw conclusions about software-related drain. Proactive maintenance recommendations could go to IT teams before devices degraded. Employee utilization analysis followed the same logic. If a device assigned to an associate was stationary for the majority of a shift, that was a training signal. Store managers could see which associates were engaging with their tools without needing to shadow anyone.

For FedEx, the geofencing model applied to the warehouse floor. Sections of the facility mapped to business functions: loading zones, staging areas, dock assignments. Device location over time told the story of whether crews were in the right places during the right windows. Camera data from the docks added volume and surface area readings. A truck that hadn't changed in several hours wasn't being loaded. The crew responsible for that dock was somewhere else.

Then COVID hit.

In early 2020, FedEx had to keep operating under lockdown conditions. Contact tracing became an immediate regulatory and operational requirement. The question was: which employees had been within six feet of each other, and when?

It took one week to adapt the existing model to answer that question. The same mechanism that told us a crew had drifted from their assigned dock area was the mechanism that told us two device holders had passed within close proximity. The event data was already structured. The semantic layer was already in place. We mapped a new set of business rules onto an existing foundation and handed FedEx a functional contact tracing tool in the middle of a global crisis nobody had planned for.

For Lululemon, the modeling challenge was identity. A MAC address ping tells you nothing about who is holding the device. We built a classification layer using behavioral rules. A device that appears in the store after closing hours is almost certainly an associate. A device that visits three or more times in a single day is likely an associate. Everything else is probably a customer.

Once classified, movement patterns became meaningful. We mapped common paths through the store onto heat map visualizations. We identified zones with high customer traffic and low associate presence and looked at whether those imbalances correlated with lower sales per foot traffic. Well-staffed stores averaged 15% higher sales per foot traffic than locations where staffing patterns were disorganized. That wasn't a hypothesis. It was in the data.

OUTCOMES

For the client's customers, the immediate results were operational. Proactive geofencing flagged roughly 10% of the device fleet as at-risk before any device was ever declared missing. FedEx gained warehouse efficiency tooling and, within one week of COVID reaching US operations, functional contact tracing infrastructure. Lululemon had a measurable, product-integrated way to connect staffing decisions to sales outcomes.

For the client, the outcome was structural. The initial Walmart MVPs were delivered in one month. Walmart's own estimate for getting there from raw data was six months. That gap wasn't about speed. It was about knowing what questions to ask and how the data needed to be shaped to answer them.

By the end of the engagement, the client had a repeatable data product playbook and a go-to-market motion built around solving operational problems rather than selling device subscriptions. That motion created customer loyalty hardware sales alone could not. When your vendor is the reason your operations adapted to a global crisis in a week, the renewal conversation is different.

WHAT THIS MEANS FOR OTHERS

A timestamp and a value is not data. It's a signal. Whether it becomes useful depends entirely on whether you know what it's signaling and what it means in the context of your business.

The client's customers were not wrong to want raw access. But access to data and the ability to derive answers from it are different things. The event data across retail, logistics, and supply chain was structurally similar. The same modeling approach applied across three completely different operations. When circumstances changed overnight, the model adapted in days because the foundation was already sound.

Any organization sitting on raw IoT or event data and waiting for the right team to figure it out should ask a simpler question first: do you know what operational problem you are actually trying to solve? If the answer is vague, the data will be too.

The product was always in the data. But so was the complexity.