Written from deployments that are running, with the measurements, the failure modes and the costs left in. No vendor summaries.
Every vendor quotes you a per-sensor price. None mention that a managed platform's per-device fees can cost €36,000 over five years for a single deployment, before hardware. Here's where IoT budgets actually go, and why the platform decision makes or breaks ROI over five years.
The old advice was simple: sensor data is time-series, so use InfluxDB, because Postgres handles it badly. Both halves of that have since changed. TimescaleDB turned Postgres into a real time-series engine, InfluxDB rewrote itself twice and broke its query language both times, and a ten thousand device fleet turns out to generate about eleven rows per second anyway.
Thousands of IoT sensors exist, and many work flawlessly on the bench then fail in the field. The datasheet won't tell you which. Here are the criteria that actually decide a deployment, from firmware quality and battery chemistry to build materials, lead times and vendor support, and how to validate a device before you order five hundred of the wrong one.
Your 10-sensor pilot works perfectly. Six months later, the 500-device rollout is stalled. Different problems, different budget, different politics. 70% of enterprise IoT projects never escape pilot purgatory. Here's what actually goes wrong and how to architect for scale from day one.
Your IT security team has one question before the rollout is approved: where do the encryption keys live? Nearly every IoT protocol encrypts its traffic, and 'encrypted' and 'secure' are not the same thing. Here is what the protocols actually guarantee, where real deployments go wrong, and the architecture that earns sign-off.
Your utility bill shows total consumption, but not where the water actually goes. A leak runs for months before anyone notices. Wireless sub-meters pinpoint consumption by zone and detect leaks within hours: pulse-counter versus ultrasonic options, installation without shutting off supply, and typical payback periods.
The technician checks at 9am and 5pm; the compressor fails at midnight. Continuous monitoring catches equipment failures within minutes: detect problems before spoiled inventory, generate automatic audit trails for compliance.
InfluxDB is purpose-built for time-series data, so queries stay fast across millions of datapoints without asking you to think about partitioning. Setting up the data pipeline from ChirpStack to InfluxDB to Grafana, the tag and field decisions that determine query speed a year later, and retention policies that keep storage manageable.
A construction foreman spends half a morning calling other sites looking for a generator. GPS/GNSS trackers show where equipment actually is: theft alerts while recovery is possible, utilization data revealing which expensive machinery sits idle.
Most IoT projects stall at the same place: the data reaches the device platform, and then someone spends three to six months writing middleware to get it into the cloud. Native integration removes that step. Device twins, Thing shadows, message routing and serverless processing, with the path from sensor to storage down to two hops.
Which tenant used what? Which EV charger drew how much? Is your solar actually producing what the salesman promised? CT clamp sensors install without shutting off power, so you finally know where the electricity actually goes.
A rain gauge placed near a building undercounts by 30% because wind turbulence throws rain away from the funnel. A temperature sensor without a radiation shield reads 10-20°C too high in direct sunlight. Weather stations produce garbage data when installed wrong.
Most buildings run HVAC on timers, conditioning empty spaces. Occupancy sensors reveal when rooms are actually used, so you can cut energy waste, improve air quality, and catch leaks before they become expensive.
Most irrigation runs on fixed schedules: 4 hours every 3 days regardless of whether it rained yesterday. Soil moisture sensors change that, cutting water use by 30-50% by triggering irrigation only when the soil actually needs it. Sensor placement, valve control, and the logic that ties them together.
There is no best IoT radio, only the one that survives your site, your power budget and your five-year bill. Wi-Fi drains batteries. Cellular bills per device forever. A private LPWAN is yours but only covers ground you put gateways on. Here is how the choice actually gets made.
A rancher managing hundreds of cattle across thousands of hectares: where are they right now? Riding out to check takes hours. GPS trackers show herd locations, catch fence breaks, and spot behavioural changes that signal health issues.
Fifty panels showing every possible metric crammed onto one screen. Nobody looks at it. Good dashboards show 6-8 panels that actually drive decisions. Panel types, alerting strategies, and query optimization that keeps dashboards fast.
You need to mount an anemometer on 32mm pipe at a specific angle to clear a solar panel. No commercial bracket exists. Design it in CAD, print overnight in ASA, install the next day. Custom mounts for a fraction of commercial prices, ready when you need them.
That 1995-era water meter still works fine. That €200k CNC machine isn't going anywhere. Instead of replacing legacy equipment, add connectivity to what you already have: Modbus, RS485, 4-20 mA, pulse outputs. Custom firmware bridges the gap between decades-old industrial protocols and modern IoT monitoring.