What the HSE Fatigue and Risk Index can and cannot tell you
I built a product on the HSE Fatigue and Risk Index, so every commercial instinct says this post should sell the model hard. It is going to do the opposite, and I want to be honest about why.
When you spend months inside a methodology, building software around it, testing it against real rotas, you develop a relationship with it that marketing copy cannot survive. You learn where it is genuinely clever. You also learn, with uncomfortable precision, where it is guessing. I know exactly which of my product's numbers rest on solid modelling and which rest on assumptions about a statistical average human who does not exist. The vendors in this space who never tell you that second part are not lying, exactly. They have just decided you do not need to know. I think you do, because trust in a safety tool that is not calibrated to what the tool can actually do is itself a hazard.
So here is the honest account.
What it does well
The FRI is a biomathematical model. Feed it a working pattern, shift start and end times, rest gaps, the cumulative sequence, and it estimates two things: the probability that a person working the pattern experiences high sleepiness (the fatigue index), and the relative risk of a fatigue-related incident compared with a reference pattern of twelve-hour shifts on a two-day, two-night, four-off cycle (the risk index).
Its strengths are real. It is published, transparent and free, three words that describe almost nothing else in this market. It has been evaluated in the open literature, including outside its home industry: a 2019 study applied it to 95 foundation doctor rotas and found it workable and informative. It captures the things that matter most at population level: time of day, shift length, rest opportunity, cumulative load. And it is comparative by design, which is its most defensible use. If pattern A scores materially worse than pattern B, that difference is worth taking seriously even if you distrust both absolute numbers. I trust the model most when it is ranking and least when it is pronouncing.
What it cannot see
Work as done. The model scores the pattern you type in. Every evaluation of the tool makes the same point: planned rotas systematically understate real exposure, because overtime, swaps, on-call disturbance and late finishes all happen after the plan is published. I used to write rotas for a living, in a coffee shop rather than a control room, and not once did the published week match the worked one. A one-off assessment at roster design is an audit of a document, not of anyone's actual fatigue. The honest response is to score worked hours as well as planned ones, continuously. This single limitation shaped RotaPulse more than any other.
The individual. The FRI models a statistical person. It knows nothing about age, chronotype, health, a new baby, a second job or a ninety-minute commute each way. Two people on the same pattern can be in completely different physiological states, and the tool will score them identically. This is why using FRI outputs to judge individuals, in either direction, is a misuse. It describes patterns, not people. If a score is ever quoted in a disciplinary meeting, something has gone badly wrong upstream.
Sleep itself. The model infers sleep opportunity from the gaps between shifts. It does not know whether anyone slept. Wearable-based systems attempt to close this gap and bring their own problems: cost, consent, surveillance concerns and data quality. I have mixed feelings about strapping monitoring devices to workers in the name of their own safety, and I suspect many workers do too. The FRI's inference is a limitation to be aware of, not a scandal, but it is a real one.
It is worth knowing that the HSE itself withdrew the FRI spreadsheet from its website in June 2021, citing two reasons: the Excel platform could no longer be supported, and the tool's design needed improvement to promote better understanding of its outputs, its limitations, and its role within a wider fatigue risk management system. Read that second reason again. The regulator that commissioned the model concluded that the way it was packaged was leading people to misunderstand it. That is the clearest official endorsement of this post's argument I can offer, and the ORR has since documented cases of the FRI being misused to justify work patterns that clearly required further action.
A verdict. There is no pass mark. The thresholds used in the rail industry were explicitly indicative, meant to flag patterns for attention. The tool's own guidance says risk should be minimised and a tolerable level agreed for the specific work. A score below threshold is not a safety case, and treating it as one converts a screening tool into false assurance, which is worse than no tool at all.
The uncomfortable general point
Every fatigue model, the FRI included, faces the same critique: it produces a precise-looking number from an imprecise reality, and precise-looking numbers invite overconfidence. I write poetry in my other life, so I spend a certain amount of time thinking about the gap between a description and the thing described. A fatigue index of 22.4 has one decimal place more confidence than anyone should feel about a human being's Tuesday night. The literature calls this the risk of using biomathematical models as compliance instruments rather than design aids. The number is the beginning of a conversation between the pattern, the task and the people working it. It is never the end of one.
Why I built on it anyway
Because the alternative, in most UK shift operations, is nothing. Not a better model: nothing. Rotas get checked for legality and cost, and fatigue gets managed by exception report and incident investigation, after the fact. A transparent, published, imperfect model applied to every rota beats a perfect model applied to none.
My position is simple. Use the FRI for what it is good at: comparing patterns, flagging outliers, making the fatigue consequences of a rota visible at design time and against worked reality. Refuse to use it for what it cannot do: judging individuals, certifying safety, replacing judgement. RotaPulse is built around exactly that boundary, and I would rather tell you where the boundary is than pretend it does not exist. If that costs me a sale to someone shopping for false assurance, that is a sale I am content to lose.
The model, its published evaluations and the rail industry threshold guidance are all linked below. Read them. Then score a rota with the free calculator and see what the numbers start to show you.
Sources
- HSE Research Report RR446, Spencer MB, Robertson KA, Folkard S. The development of a fatigue / risk index for shiftworkers (2006). The primary methodology source.
- Fatigue and Risk Index Calculator Version 2.2 User Guidance (HSE), mirrored copy, including the Cautions for Users section.
- ORR, Managing Rail Staff Fatigue, Appendix D: Benefits and limitations of fatigue risk assessment tools.
- Health and Safety Laboratory (2008), Evaluation of the UK Rail Sector Initial Fatigue and Risk Index Thresholds.
- Cumber E, et al. BMJ Open 2019;9:e023470, application of the FRI to foundation doctor rotas, including the work-as-planned versus work-as-done caveat.
- RSSB, Fatigue Risk Management Systems.
- HSE, Human factors: fatigue.