Somewhere in your organisation is an investigator who has seen three versions of your next serious incident. They know which isolation procedures get rushed when the shutdown window is tight, which site's permit system drifts under pressure and why the 2019 recommendation on lifting plans never stuck. In five years they will be gone, and most of what they know will go with them. Safety knowledge retention is becoming one of the quieter risks in high-risk industry, and few organisations have a plan for it.

The numbers are not abstract. A third of the UK rail workforce is aged 50 and over, and up to 70,000 workers could leave through retirement and attrition by 2030. Workers in energy and utilities expect to retire at 59 on average, the youngest of any sector surveyed and seven years ahead of state pension age. The practical safety expertise that took thirty years to build sits in the heads of people in their late fifties and early sixties, and the handover plan in too many organisations consists of a lunch and a laptop wipe.

The institutional safety knowledge that never gets written down

The formal record survives, investigation reports stay in the system, procedures stay on the intranet and the training matrix shows who was certified for what. What leaves is the connective tissue. The experienced investigator knows that the incident in the northern region in 2021 and the near miss at the depot last spring were the same failure wearing different clothes. The report for each sits in the archive, written by different people in different formats, and nothing links them except the person who remembers both.

That pattern recognition is the most valuable thing a senior safety professional does, and it is almost entirely undocumented. When they retire, the organisation keeps the incidents and loses the pattern.

The knowledge transfer gap: why investigation quality drops when experienced people leave

Experienced practitioners investigate consistently because they have internalised a method. A new investigator, however capable, produces reports that vary in depth, structure and causal language until they have done enough of them to settle. Across a team with high turnover that variation means the data from one investigator cannot be compared with the data from another, and the organisation’s collective learning fractures into individual experience.

Learning communities, mentoring and structured handover all help, and the organisations that run them well are better placed than those that do not. But none of them scales to the volume of knowledge that leaves a large operator every year, and none of them can read 20 years of investigation reports on behalf of the person who has just joined.

What the ageing workforce problem looks like at scale, and at a single site

One global power technology manufacturer runs COMET across its operations, with around 2,300 active users and an average of 11,500 investigations a year over the past five years. At that volume, no individual holds the pattern. The investigator in one plant has never read the reports from another, and the person who could once connect a failure in one region to the same failure elsewhere may have retired two reorganisations ago. For an organisation of that scale, the question is no longer whether knowledge will leave but whether the system can hold what the people cannot.

The same challenge presents itself differently at a single site or a small operator. There, one or two people often hold the entire investigation history in their heads. There is no archive large enough to need a system, until the day the person who was the archive hands in their notice. A small operation loses a higher proportion of its knowledge in a single retirement than a large one loses in a year, and it has fewer people left to notice the gap.

How AI helps keep investigation knowledge in the business

When an experienced investigator retires, the hardest thing to replace is their ability to connect one event to another. That is where AI makes the biggest difference to knowledge retention. Built into COMET, it searches years of investigations for the patterns a retiring expert used to carry in their head. It also supports newer investigators in the workflow itself, drafting timelines and proposing causal factors so their work builds on everything captured before them. The judgements that matter most, identifying root causes and analysing Human Factors, stay with the trained investigator by design, so the AI strengthens their expertise rather than replacing it.

Three things make that work.

• Every investigation is captured against the same coded taxonomy, so the AI works from a consistent record rather than inheriting the fragmentation of the past.

• Every AI suggestion is reviewed by the investigator before it is accepted, because a suggestion is not a decision.

• The questions it answers are the ones a senior practitioner would ask, not just the ones a dashboard happens to answer.

How COMET keeps investigation expertise in the business

COMET was built by career investigators, and the methodology encodes the way they think. The five Root Maps (Communication, Operating Environment, Management, Equipment and Training) and the coded root cause taxonomy give a new investigator a structure to work within from their first case, so the data they produce compares with everything captured before them. Accredited Human Factors Analysis asks the questions the experienced investigator would have asked about the person at the centre of the event.

COMET AI Assistant turns the archive into something the organisation can question. Insight AI, delivered through COMET Companion Plus, lets a safety leader ask in plain language which root causes recur most often, which human factors keep appearing and which preventive actions have correlated with fewer incidents. The questions a retiring expert used to answer from memory, the organisation can now answer from its data. Input AI supports the investigator in the workflow itself, drafting timelines and proposing causal factors for the investigator to review so that someone with two years’ experience produces work much closer to the consistency of someone with twenty.

For organisations whose historical records were captured before any consistent method existed, COMET Signals processes large unstructured datasets from HSEQ and other operating systems and maps them back to the organisation’s taxonomy, whether that is COMET or not, so the knowledge in a decade of free-text reports becomes part of the shared record rather than a filing cabinet nobody opens.

Your most experienced people will retire. The question is whether their safety knowledge retires with them. Book a discovery call to talk about keeping it.

References

1. National Skills Academy for Rail, Workforce Survey 2025.

2. Osborne Clarke, Understanding an ageing workforce: Energy and Utilities sector findings.‍