Why 40% of your tags are orphaned, and why that is good news
Across 240 site surveys the median share of signals that nothing reads is 38%. Most plants read that as waste. It is usually the opposite.
4 September 20269 min
The first thing a signal survey produces is an inventory, and the first thing the inventory produces is an argument. Somebody always says the number is wrong.
It is not wrong. On the median site we survey, 38% of configured tags are written by equipment that nothing downstream reads. On the worst site we have seen it was 71%. The instinct is to treat that as an indictment of the instrumentation budget, and to start asking who approved all those sensors.
What orphaned tags actually mean
Almost none of it was over-instrumentation. In nearly every case the tag was read by something once — a reporting tool, a vendor dashboard, an optimisation package — and that something was decommissioned, replaced or simply stopped being opened. The sensor kept writing. Nobody turned it off, because turning things off in a plant requires a change request and there was no incentive to raise one.
So the instrumentation outlived its consumers. That is a software lifecycle problem wearing a hardware costume.
Why this is the cheapest work available
Orphaned tags are not free. They consume historian write throughput, they inflate every backup, they slow every query that scans a time range, and — the expensive one — they pollute every conversation about data quality, because nobody can tell at a glance which signals matter.
Retiring a dead tag costs a change request. Keeping it costs a slice of every query your plant runs for the next decade.
At Đông Hải we retired 41% of 340,000 tags before writing a line of integration code. Ingest throughput requirements dropped by a third, which changed the hardware specification, which took roughly USD 200k out of the deployment before it started.
How to find them
You cannot do this from the historian configuration alone, because configuration tells you what is written, not what is read. You need read-side telemetry — query logs, dashboard definitions, report bindings — and then a join that most plants have never performed.
That join is the single most valuable artefact of a survey, and it is why we hand it over whether or not the client engages us afterwards. It is their data estate. They should have the map regardless of who draws it.
- Signal survey
- Data quality
- Deployment