We ran two connected device programmes. The first cost a great deal and produced a dashboard nobody used. The second changed how we schedule maintenance and paid for itself inside two years.
The difference was visible in the sales process, and I did not know what to look for the first time. This is what I would ask an IoT company now.
The demonstration proves almost nothing
Every vendor shows the same thing. A device, a signal, a live chart on a screen.
That demonstration proves connectivity works in a room with good coverage on hardware somebody prepared that morning. It tells you nothing about the questions that decide whether the programme works.
Our first IoT services company gave an excellent demonstration. Eighteen months later we had four hundred devices, a third of which had stopped reporting, and no process for finding out why.
Ask about year three
The build is not where these programmes fail. The running is.
Devices go offline. Batteries die. Firmware needs updating on hardware bolted to a wall in a plant room. Somebody installs a new network and half your estate stops talking.
Ask what year three looks like. How do devices get updated? Who notices when one stops reporting? What is the expected failure rate and who replaces them?
A supplier without confident answers is selling you a deployment rather than a system.
The four questions I use
- What proportion of your existing deployments are still reporting reliably after two years?
- How is firmware updated on a device we cannot easily reach?
- What happens when connectivity is poor at a site, and how do you know it is poor?
- What data do we own, in what format, and how do we get it out?
The last one matters more than it sounds. Our first supplier held our telemetry in their platform, and extracting three years of history when we left was slow, expensive, and incomplete.
Who runs it after installation
The question I did not ask the first time.
Somebody has to watch the estate, chase silent devices, schedule battery replacements, and decide when a unit is faulty rather than reporting oddly.
That is a permanent role. Our first programme assumed it would be absorbed by an existing team, and it was not.
We now decide this before signing. Either an IoT services company runs it under a defined agreement, or we staff it internally. Assuming it will happen is how estates degrade.
Start with the decision, not the sensor
Our second programme began differently.
We asked which operational decisions we were making badly, and only then which measurements would improve them.
Maintenance scheduling came top. We were servicing equipment on a fixed calendar, which meant some assets were serviced too often and others failed before their turn.
That decision told us what to measure. Two sensors per asset, not fourteen. A far smaller programme, and it changed something.
The failed version worked the other way round. We instrumented everything, produced a great deal of data, and never established which decision it was meant to inform.
Count the whole cost
The device price is the smallest number in the business case.
Installation is significant, particularly in operational environments where access needs planning. Connectivity carries an ongoing fee per device. Platform costs scale with data volume. Replacement happens on a schedule you can predict.
And somebody has to run it. That is a permanent role, not a project.
We now model five years. Two of our proposals looked considerably less attractive that way, and one was correctly stopped.
Where the AI question fits
Every proposal now includes one, and it deserves a plain answer.
The useful applications in our estate are narrow. Anomaly detection against learned baselines catches slow degradation that fixed thresholds miss entirely. Forecasting failures a week out changes what our maintenance team does on Monday.
We are also being offered more ambitious things. Any generative AI architecture proposed on top of device telemetry should be assessed like any other component. What decision does it improve, what does it cost to run, and what happens when it is wrong.
Two of our proposals in that category had enthusiastic descriptions and no answer to the third question.
What good suppliers do differently
The one we kept asked to visit our sites before quoting.
They came back and told us two of the assets we wanted to instrument were not worth it, because the failure mode we were worried about would not be visible in the data we would collect.
That reduced their contract. It is also why they are still our supplier, and why I trust the rest of what they tell me.
An IoT company willing to shrink a proposal in week two is showing you something no reference call reveals. That is the IoT company I would shortlist.
Common questions
How large should a first deployment be?
Ours was thirty assets across two sites. Large enough to be real, small enough that being wrong was affordable.
Should we own the platform?
Own the data at minimum, in a format you can export. The platform can be theirs. Your telemetry history should not be.
Should the supplier run the estate too?
Often, yes. An IoT services company arrangement with defined response times costs less than discovering nobody was watching.
Do we need a specialist supplier?
Usually. General integrators underestimate the field and lifecycle work, which is where most of the difficulty sits.
How do we compare suppliers fairly?
Give each the same three sites and the same decision you want improved. Every IoT company looks similar against a generic brief.
How long before this pays back?
Ours took about two years on maintenance savings alone. Anything promising a return in months is describing the demonstration rather than the deployment.
What I would tell another chief executive
Ignore the demonstration. Ask about year three, ask what proportion of their existing devices still report reliably, and ask what happens to your data if you leave.
Then start with a decision you are making badly rather than with a set of things you could measure.
Our second programme instrumented far fewer assets than the first and produced considerably more value. That was not better technology. It was a better question at the start, and any competent IoT company will help you ask it if you give them the chance.
