“Predictive” is the word vendors reach for when they want to sell a sensor. The pitch is seductive: stop reacting to last quarter’s survey and start seeing risk before it happens. There is a real idea buried in that pitch, and there is a great deal of marketing wrapped around the real idea. Telling the two apart is the whole job.
Stripped of the gloss, a predictive approach means using continuous data — temperature trends, water turnover, asset condition — to catch risk building before it trips a control limit, rather than finding out at the next walk-round or the next failed sample. Done well, it shortens the gap between “something drifted” and “someone acted”. Done as a box-ticking purchase, it adds a dashboard nobody reads. The foundations of UK Legionella compliance have not moved: the site-specific risk assessment and written scheme are still a duty, not an optional extra [1].
Where the gains are genuine
Between scheduled surveys, plenty can drift. A trace-heating circuit fails on a long run, a TMV starts passing warm water back to the cold side, a wing of rooms is taken out of use and quietly stagnates. Continuous and remote temperature monitoring catches that kind of drift in days rather than at the next quarterly visit. That is the legitimate core of the idea — “predictive” is really “earlier”, and earlier is worth paying for.
Trend analysis adds the second piece: it turns a pile of isolated readings into a direction of travel. An outlet that is slowly warming, a calorifier taking a little longer each month to recover, a cold feed creeping up across summer — none of those shows up in a single spot check, but all of them show up in a line on a chart. This sits comfortably inside the plan-led, preventive model that water safety plan guidance encourages [5] and the monitoring that HSG274 already expects [2]. It changes the cadence, not the principles.
So the gains are real but narrow. Most of the trouble starts when the cadence improvement gets sold as something larger.
Five claims worth a second look
Myth: a predictive system replaces the risk assessment
The assessment is the decision document; the sensors are evidence that feeds it. Under L8 the site-specific risk assessment and the written scheme are a duty in their own right, not a survey you can automate away [1]. An algorithm can rank outlets by how stagnant they look. It cannot decide that a redundant spray tap should be removed, or that a newly vulnerable occupant on the third floor changes your priorities. That judgement stays with a competent person. The myth sticks because “continuous” sounds more thorough than “annual” — but continuous monitoring supplements the assessment, it does not supersede it.
Myth: the algorithm predicts an outbreak
What these systems actually predict is conditions associated with growth: a calorifier slow to recover, a cold feed warming, an outlet whose turnover has fallen to nothing. They cannot see biofilm on a pipe wall, and they cannot confirm the organism is in the water. Only laboratory sampling gives direct evidence that Legionella is present [3]. A good model tells you where to look harder; it does not do the looking. Treat a prediction as a prompt to investigate, never as a verdict on whether the system is safe.
Myth: more data means a safer system
A platform logging thousands of readings an hour is worth precisely as much as the response it triggers. If an alert lands in an inbox nobody owns, you have not added a control — you have created a record showing you knew and did nothing, which is a worse position than not knowing. The value lives in the loop: who is alerted, what they are authorised to do, and how fast they do it. Design the response before you buy the sensor. The myth persists because data volume is easy to show in a demo, while the response process stays invisible until something fails.
Myth: continuous monitoring lets you scrap flushing and sampling
A temperature sensor measures temperature, usually at a single point. It does not flush a dead leg, descale a shower head, lift a tank lid or culture a sample. Monitoring scope and frequency are still set by your risk assessment, and HSE is explicit that sampling frequency should follow the system and the assessment rather than a fixed schedule [3]. Where data analytics earns its place is in targeting physical work — flush the outlets the readings say are stagnating, sample where a trend looks wrong — not in retiring the work. Beyond the survey goes further into running assessment as a living process rather than a once-a-year event.
Myth: predictive means AI, and you must buy it now
Most of the “prediction” working on real sites today is threshold alerting and trend analysis: explainable, proven and cheap to act on. Machine learning layered on top adds little when the asset register and schematics are wrong, because the model inherits every error in them — confident output built on a bad inventory. Get the foundations right first. An accurate system inventory and an up-to-date schematic are what every method, old or new, depends on [4]. The myth persists for a simple reason: “AI” sells, and “keep your asset register current” does not.
Before you sign anything
Pick the part of your estate where you already struggle — the wing of intermittently used rooms, the plant you cannot easily reach, the outlets that fail temperature checks most often — and run a small trial there rather than across the whole site. Ask the supplier to show how their system would have caught your last three control failures, and what action each alert would have triggered, named to a real person on your team.
If they cannot map alerts to owners and authorised actions on your site, you are being sold a dashboard, not a control. And before the first sensor goes live, write the response routine into your written scheme so the data has somewhere to go. The cost case for any of this is genuinely a cost-versus-risk decision rather than a gadget purchase — Justifying Legionella control measures: cost vs risk sets out how to make that argument to a budget holder.
What this kit can and cannot prove
Continuous data and analytics are monitoring aids. They are not a regulatory shortcut, and they are not a diagnosis. None of it replaces a competent, site-specific risk assessment, clinical judgement about illness, or engineering judgement about your plant. Any threshold an alert fires on — a temperature, a turnover figure, a frequency — should come from your own assessment and the current HSE guidance for your system, not from a supplier’s factory default. Where a specific number would change a decision, confirm it against the source before you rely on it.
FAQ
Does continuous monitoring satisfy our legal duty to carry out a risk assessment?
No. Monitoring shows whether controls are holding; the risk assessment decides what the controls should be and who could be exposed. You need both, and the assessment comes first [1].
Can a predictive system tell us whether Legionella is actually present?
No. It can flag conditions linked to growth, but only laboratory sampling confirms the organism in your water [3]. Use the alert to decide where and when to sample, not as a substitute for sampling.
Is a predictive system worth it for a small, low-risk site?
Often not. The payback is largest where you have many intermittently used outlets, hard-to-reach plant, or a history of missed checks. For a simple system with disciplined manual monitoring, the money may be better spent on removing dead legs and keeping records straight.
Sources
[1] 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 [2] HSE, “Legionnaires’ disease: Technical guidance (HSG274)”. https://www.hse.gov.uk/pubns/books/hsg274.htm [3] HSE, “Testing and monitoring your water system for legionella”. https://www.hse.gov.uk/legionnaires/testing-monitoring-water-system.htm [4] BSI, “BS 8580-1:2019 - Risk assessments for Legionella control. Code of practice”. https://knowledge.bsigroup.com/products/water-quality-risk-assessments-for-legionella-control-code-of-practice-1 [5] BSI, “BS 8680:2020 - Water quality. Water safety plans. Code of practice”. https://knowledge.bsigroup.com/products/water-quality-water-safety-plans-code-of-practice