Most organisations know how many claims they had last year and what the average cost was. These figures have value at renewal and in the budget, but on their own they can't tell you where your next loss will come from.
Your claim files can often provide clues. Investigation reports, adjuster notes, and supporting documents show what was happening, when and where, who was involved, what supervision was in place, and what evidence was available afterwards. These details may sit outside searchable fields, but developments in AI and data analytics are making it increasingly possible to analyse this unstructured information at scale, identify patterns across claims, and highlight areas that may warrant closer investigation. Technology can help surface the signals, but claims and risk specialists are still needed to test what the data appears to show, understand the operational context, and determine what action is appropriate.
Start with the information that's recorded consistently, then review a selection of files to look for patterns. Taken together, these files show how an organisation's controls work in practice.
Reading Past the Incident Type
Look beyond the incident type for patterns. Four manual handling injuries across 20 sites tells you very little, but four in the same dispatch bay, all involving inbound pallets, is worth investigating. The same applies to controls, such as a permit that wasn't raised, an isolation that wasn't applied, or a pre-use check signed off after the event. If the same control failure keeps showing up behind incidents that otherwise seem unrelated, find out whether that control is practical, understood, and consistently applied.
Losses also concentrate. Particular sites, vehicles, roles, and shifts carry more than their share. Check whether patterns coincide with certain shifts, agency staffing, recently acquired sites, or changes in contractor arrangements. Comparing similar locations is often more useful than broader benchmarking. Differences between comparable depots warrant investigation into local conditions, supervision, and working methods.
What the Claim Itself Tells You
Some of the strongest signals come from how the claim behaves, rather than the incident itself.
- The first is notification lag. Delays in reporting an incident can make it harder to preserve evidence and investigate what happened. Where delays recur, check whether reporting procedures are clear, staff know when and how to report, and supervisors pass information on promptly.
- The second is thin evidence. For example, overwritten CCTV, a missing risk assessment, no witness account, or an investigation written up a fortnight later. Any one of these weakens a single file, but a run of them shows how consistently the system is operated day to day.
- Then there are claims that look routine at notification but become expensive later. Review what was known at notification, what emerged later, and how that affected the handling and cost of the claim.
Predictive analytics can help identify characteristics shared by claims that have previously escalated, allowing new notifications with similar features to be highlighted for earlier attention. This should not be treated as a prediction of what will happen in an individual case, but as an additional source of insight to support experienced claims professionals in deciding where closer review may be needed.
Near misses and complaints should also be reviewed alongside claims, as they may reveal recurring hazards before another injury occurs.
Testing What You Find
Patterns must be tested. First, compare the numbers with exposure: headcount, hours worked, miles driven, volumes handled. Second, check the pattern against how the work is actually done, not just the protocol. If a pattern survives both tests, act on it. If not, investigate whether exposure, recording differences, or other factors explain it before selecting an intervention.
The same discipline applies to patterns identified through AI or analytics. Technology can process large volumes of information and highlight possible relationships, but the quality of its output depends on the completeness and consistency of the underlying data. Any finding must therefore be checked against the claim file and the operational circumstances before conclusions are drawn.
Even a real pattern identifies only the symptom, not the cause, and this is where many claims reviews come unstuck. A recurring incident is often seen as evidence of a training need, which can be addressed with a course. But just as often, people already know the procedure and simply can't follow it with the equipment or time they've been given. In this case, it's the work that needs to be redesigned.
In other cases, the problem is that when a contractor arrangement changed, responsibilities became unclear and nobody recorded who was now responsible for what. Training is only one option, alongside equipment, supervision, staffing, and working methods. Organisations that choose it by default often conclude that training doesn't work.
For carriers, the next step is to agree how relevant findings will reach underwriting and risk specialists, and how proposed improvements will be discussed with the broker and insured. Each agreed action should have a named owner, a completion date, and evidence of implementation. Subsequent reviews should check outstanding actions as well as claims outcomes. Identifying a recurring hazard is only the beginning; someone must be responsible for addressing it.
Knowing Whether It Worked
Before making changes, agree how you'll measure success. Early checks can be modest. Are the new arrangements being followed? Do the people involved know what changed? Are incident records now complete? Over a longer horizon, the measures are recurrence, notification times, frequency, and severity.
- First, recent claims may still develop, so an early fall in reported cost is not, by itself, evidence of a lasting improvement.
- Second, when several things were changed at once, the improvement cannot honestly be attributed to any one of them. Recognising this gives the next review its credibility.
A Different Kind of Claims Relationship
A third-party administrator is measured by how well it handles the loss once it has occurred, and that work still has to be done properly. But the file that settles a claim also holds the evidence of why the claim arose, and that evidence only has value if somebody is in a position to act on it.
Gallagher Bassett is built to do both. Claims handling, data analytics, risk consulting, and training sit within the same business, so patterns identified through the data can be interpreted by specialists and translated into a practical response without being handed between suppliers. Technology can help bring the relevant information together and surface emerging trends, while claims and risk expertise determines what those findings mean and how they should be addressed.
In one published example, a large logistics client with a high collision rate had its incident, collision, and claims data reviewed alongside telematics data, and the findings were used to build a bespoke accredited Driver CPC module aimed at the risks identified in the review. The collision rate fell by 60% afterwards.
Your claims data has already cost you a great deal. It's worth asking what it's being used for beyond explaining last year.
To discuss a claims-led review of where your losses are coming from, get in touch with the Consulting Solutions team.
Author
Make Gallagher Bassett your dependable partner
When making the right decision at the right time is critical to minimise risk for your business, count on Gallagher Bassett's extensive experience and global network to deliver.