How to start an IoT project without losing your mind

Most IoT projects that fail do not fail because of the hardware. They fail because nobody defined which decision the data was going to drive.

Start with the decision, not the sensor

Before buying anything, answer this: when this number goes out of range, who does what? If there is no answer, you do not need sensors yet, you need to define the process. A beautiful dashboard nobody looks at is the most common way to burn an IoT budget.

The minimal architecture

A project that works has four pieces and not one more:

• The node: the device that measures and transmits, with local storage so nothing is lost when the network drops.
• The transport: MQTT or HTTP over wifi or cellular. MQTT when there are many nodes or the link is poor.
• The storage: a time-series database, not an ordinary table growing without control.
• The rule: threshold, duration and who to notify. This piece is what turns data into value.

The three mistakes we see most

The first is installing too many sensors up front. Thirty points on day one produce thirty times the noise and zero learning. Start with three to five, at the places where a deviation costs money.

The second is alerting without hysteresis. If the rule fires on a single reading, the first time someone opens a door ten alerts go out, and within a week the team mutes all of them. Require two consecutive readings out of range.

The third is interpolating gaps. When there is no data, the chart must not draw a tidy straight line: it must show that no data arrived. Faking continuity is worse than admitting the gap, especially in an audit.

What to expect in time and cost

A pilot of three to five nodes with dashboard and alerts takes four to eight weeks, including circuit design when the hardware is our own. Cost concentrates in the first installation; growing from five to twenty-five nodes costs considerably less than starting.

We do both halves: electronic design and firmware on one side, dashboard and rules on the other. That is how we built Edgemon, our temperature and humidity monitoring, which is now in a pilot program.