Troubleshooting

Why Your IIoT and SCADA Numbers Are Wrong: 8 Costly Mistakes

The eight most common errors that throw off IIoT ROI, tag counts, edge coverage, and data availability calculations, each with a symptom, root cause, and a fix backed by a number.

Symptom: your MQTT data bill comes in 4x the estimate. Root cause is almost always confusing publish rate with sample rate. A tag sampled at 10 Hz does not have to publish at 10 Hz. If you left report-by-exception off and set a 100 ms publish interval on 5,000 tags, that is 500,000 messages per second, versus maybe 2,000 per second on change-of-value with a 1 percent deadband. Fix: rerun the MQTT Data Cost calculator with realistic COV rates, not sample rates. A vibration tag changing on every scan still publishes 90 percent less traffic once you set a deadband above sensor noise floor, typically 0.5 to 2 percent.

Symptom: PLC Data Availability reads 99.9 percent but operators swear data is missing. Root cause is measuring poll success at the OPC UA server instead of end to end. The server can answer every request while the historian drops writes during buffer overflows. Availability is a chain: PLC uptime times gateway uptime times broker uptime times historian write success. Four links at 99.9 percent each yield 99.6 percent, not 99.9. Fix: instrument each hop separately and multiply. A single link at 99.0 percent silently caps your whole chain at 99.0, no matter how good the other three are.

Symptom: Tag Mapping Workload was quoted at 3 weeks and ran 9. The missed variable is almost always tag validation and naming, not the mapping itself. Teams budget 2 to 4 minutes per tag to map but forget the 6 to 10 minutes per tag to verify engineering units, scaling, and deadband against the PLC address. For 4,000 tags, that gap is roughly 400 hours. Fix: split the estimate into mapping, validation, and rework passes, and assume a 15 to 20 percent rework rate on the first pass. Rerun Tag Mapping Workload with those three line items instead of one blended rate.

Symptom: Edge Gateway Coverage says one gateway per cell, then you run out of ports and CPU. Root cause is sizing by device count while ignoring tag throughput per gateway. A gateway rated for 50 devices may only sustain 20,000 tags at 1 second updates before CPU hits 80 percent. Fix: size on the binding constraint, whichever hits first among device count, concurrent connections, and tags per second. Check the Edge Gateway Coverage calculator against both device fan-out and aggregate update rate. A plant with 30 slow devices and 30 fast ones needs the fast group split even if the count fits.

Symptom: Data Historian Storage Cost triples in year two. Root cause is estimating raw sample volume and forgetting compression ratio degrades on noisy analog tags. Swinging-door compression hits 20:1 on clean pressure loops but only 3:1 on noisy 16-bit vibration data. If you assumed a flat 10:1 across 8,000 tags, a mix skewed toward noisy tags can double actual bytes on disk. Fix: separate tags into clean and noisy buckets, apply per-bucket ratios in the Data Historian Storage Cost calculator, and add 30 percent headroom for retention policy changes that always creep upward over time.

Symptom: Machine Connectivity Rate shows 95 percent but the OEE project stalls. Root cause is counting connected machines instead of machines producing usable data. A machine can be pingable, register in the asset list, and still expose zero of the cycle-count and fault tags OEE needs. Fix: track two numbers, physical connectivity and useful-tag connectivity, using Machine Connectivity Rate and OPC UA Tag Coverage together. A fleet at 95 percent physical but 60 percent tag coverage is really 60 percent ready. Chasing the last 5 percent of physical links wastes money the tag gap dwarfs.

Symptom: IIoT ROI looked strong on paper, then payback slipped from 14 months to 30. The usual culprit is booking only hardware and license capex while omitting recurring integration labor and network hardening. Sensor and gateway hardware is often just 35 to 45 percent of total spend once you add commissioning at 1 to 3 hours per point and annual support at 18 to 22 percent of license value. Fix: load the IIoT ROI and Sensor Installation Cost calculators with fully burdened labor rates and a 5-year total cost of ownership, not a hardware bill of materials.

Symptom: OPC UA Tag Coverage reports 100 percent but dashboards show stale values. Root cause is confusing browse coverage with subscription health. Every tag can exist in the address space while half sit in a subscription whose publishing interval exceeds the sampling interval, so updates queue and drop. Fix: audit each subscription so sampling interval is at least half the publishing interval, and keep the queue size at 1 for latest-value tags. Also watch unit mismatches: a PSI tag mapped to a bar-scaled field reads 14.5x low and passes every connectivity check while being completely wrong.

Published 2026-07-01.