The reading that should worry you is rarely the one that fails. It is the one that passes this week, passed last week, and has been sliding the wrong way for two months while every dashboard stayed green. A single temperature is a snapshot. A trend is a story, and the story is where Legionella risk usually hides.

That, in plain terms, is what Legionella data analytics is for: reading the story before it becomes an incident. Not the dramatic alarm - the quiet slope. Here is how that plays out on a monitored system, and what the pattern teaches you to look for on your own site.

An illustrative quarter: three months of green, one quiet drift

The scenario below is a composite, not a real named incident - but every element is drawn from how monitored systems actually behave.

A facilities team looks after a refurbished four-storey office. Remote temperature loggers sit on the sentinel outlets - the nearest and furthest points the risk assessment told them to watch - and a weekly report lands in the responsible person’s inbox. For a quarter, every report came back inside limits. Green across the board.

Then someone plotted the top-floor cold-water sentinel across the full three months instead of glancing at each week in isolation. The line had crept from around 16C in spring to a touch over 21C by midsummer. No single reading had ever tripped the threshold. The slope had.

The cause was mundane. During the refit a cold riser had been boxed into a service void alongside a newly insulated heating flow, and through the warmer months that void ran warm. The cold water was warming in transit, climbing into the band where Legionella multiplies most readily - broadly 20-45C as a general expectation in HSE guidance [1]. A weekly pass/fail check would only have caught it the day it finally failed. The trend caught it weeks earlier.

Why every reading passed and the system still drifted

This is the gap analytics fills. Most monitoring is built around a limit: cold below a target, hot above one, flag anything outside. That is necessary, and it is not enough. Three patterns slip straight through a pass/fail filter.

  • The slow slope. A value moving steadily toward a limit is a forecast of a breach, not a clean result. Plot the same outlet over weeks and the direction tells you more than the dot.
  • The recurring dip. A hot return that reads low every Monday morning is not random; it is the weekend BMS setback bleeding into the first checks of the week. A pattern tied to a day or a season is a clue to a cause.
  • The silence. An outlet whose flush events simply stop appearing, or a sensor that quietly drops offline, reads as “no alert”. No alert then gets mistaken for “no problem”. Missing data is data.

None of this needs clever software. A column of figures and a willingness to look across time, rather than only down the week, gets you most of the way. The tools earn their place when the estate is large enough that no one can hold every outlet in their head.

The checks that made the data trustworthy

Before the team acted on the drift, they did the unglamorous thing: they confirmed the sensor was telling the truth. A trend built on a misplaced or out-of-calibration logger is a confident piece of fiction. They checked the probe was reading the water and not the warm void around it, and that it was inside its calibration date. Graphing bad data does not turn it into insight.

Only then did the trend become an action. And an action needs an owner. The drift went to a named person with a deadline, the riser was re-routed and lagged, and the fix was recorded against the asset so the next reviewer could see why the line changed. That last step matters more than the chart. L8 expects duty holders to keep records of monitoring and the management arrangements behind it [2], and a trend you spotted but never wrote up is a trend you cannot prove you managed.

What this transfers to your site

The building does not matter; the habits do. A few that travel anywhere:

  • Read slopes, not just snapshots. Once a quarter, plot your sentinel outlets over the whole period. Direction beats any single day.
  • Compare like with like. Same outlet, same time of day, same season. A Monday-morning hot reading is not comparable to a Friday-afternoon one.
  • Treat silence as a signal. Build a deliberate check for outlets that have gone quiet - no flush logged, no reading received - and chase them as actively as a failed value.
  • Confirm before you conclude. Calibration date and probe placement first; the conclusion second.
  • End every trend with a name and a date. Analysis that does not change what someone does this week is decoration.

None of it replaces the people. Remote sensors and dashboards sharpen the judgement of the responsible and competent person; they do not stand in for it. The monitoring frequency, the limits you measure against, and what counts as a meaningful trend all come from your site risk assessment, not from the software’s defaults [1].

What the data can and cannot tell you

Temperature and flow trends are an early-warning system, not a verdict on whether Legionella is present. A cold line drifting upward raises the probability that conditions favour growth; it does not confirm colonisation. Equally, a flat, in-range line does not prove a system is clean - biofilm and stagnation can sit behind perfectly compliant temperatures. Where you genuinely need to know whether bacteria are there, that is a sampling question, and sampling regimes follow the system and the risk assessment rather than a fixed calendar [3]. Use the analytics to point your attention and your maintenance effort, decided through a competent, site-specific assessment - not as a stand-in for one.

FAQ

Can sensor data alone tell me whether I have a Legionella problem?

No. Temperature and flow trends tell you whether conditions are drifting toward favouring growth, which is a prompt to investigate or correct. Confirming whether bacteria are actually present is a sampling question, and what to sample and how often is set by your risk assessment [3].

How far back should I look to spot a trend?

Far enough to see direction through the normal noise. A weekly glance catches breaches; a quarterly plot of the same sentinel outlets catches the slow slopes and seasonal patterns the weekly view hides entirely. Match the window to how much your building’s use and temperatures swing across the year.

A reading is technically in range but climbing every week. Is that worth acting on?

Usually, yes. A value moving steadily toward a limit is a forecast, not a clean result. It is far better to find the cause - a warming void, a sticking valve, a creeping return temperature - while you still have margin, than to wait for the threshold to trip and then react.

Sources

[1] HSE, “Legionnaires’ disease: Technical guidance (HSG274)”. https://www.hse.gov.uk/pubns/books/hsg274.htm [2] HSE, “Legionnaires’ disease. The control of legionella bacteria in water systems - Approved Code of Practice and guidance (L8)”. https://www.hse.gov.uk/pubns/books/l8.htm [3] HSE, “Testing and monitoring your water system for legionella”. https://www.hse.gov.uk/legionnaires/testing-monitoring-water-system.htm