Claims Maturity Model for Bodily Injury Claims: What to Fix First

A claims maturity model gives a bodily injury claims operation a file-based way to score itself. This one sets out five levels, six dimensions, a scoring rule with a worked example and the order in which to fix gaps.

‍

What a claims maturity model is for

‍

A claims maturity model is a structured way to judge how well a claims operation works and what it should improve next. For a bodily injury (BI) claims operation, the judgement should come from a sample of real claim files, scored against a written rubric.

‍

The model below has five levels and six dimensions. It is a model to adapt, not an industry standard. It comes with a scoring rule, a worked example and a view on which gaps to close first.

‍

Why bodily injury claims management needs a maturity view

‍

Injury claims have become harder and more costly to handle. The Insurance Research Council (IRC) studied more than 7.4 million auto injury claims closed with payment between mid-2017 and mid-2022. Over that period, attorney representation rose from 40% to nearly 50%, and litigation rates rose from 10% to 18%. In the same release, average BI payments rose from about $14,000 to more than $20,000.

‍

The pressure reaches beyond auto. In 2025, the Insurance Information Institute (Triple-I) and the Casualty Actuarial Society published a joint analysis. It estimated that legal system abuse and related trends added $231.6 billion to $281.2 billion to liability insurance losses over 2015-2024. The analysis pointed to rising claim severity, not claim frequency, as the driver.

‍

No claims operation controls these forces. What it does control is how its files are built, reviewed, reserved and negotiated. A maturity view asks two questions. Is that work consistent from one adjuster to the next? And does the operation learn from the files it closes?

‍

Where the five levels come from

‍

Maturity models are older than claims technology. The best known is the Capability Maturity Model for Software, published by the Software Engineering Institute at Carnegie Mellon University in 1993. It set out five levels: Initial, Repeatable, Defined, Managed and Optimizing.

‍

The model here borrows that staged structure for BI claims, with adapted names:

‍

1. Ad hoc. Results depend on the individual adjuster. Good files exist, but no one can say why.
2. Repeatable. Some teams meet a written standard for core tasks. Others do not, so practice varies by team and office.
3. Defined. The operation has one written standard for each core task. Files show that the standard was followed.
4. Measured. Every team meets the standard, and leaders track that with data the files produce. Teams are compared on the same terms.
5. Improving. Closed-file results feed back into guidelines, training and tools on a fixed cycle.

Level 5 is not the right target for every dimension. A regional third-party administrator (TPA) and a national carrier may need different targets. The useful question is which level each dimension must reach, and in what order.

‍

Six dimensions of a BI claims operation

‍

Intake and triage

‍

Intake decides how a new file is captured and routed. In a defined operation, first notice records injury type, venue, representation status and coverage facts in structured fields. Routing rules for complex files are written down and applied the same way in every office.

‍

File review and medical evidence

‍

File review is how the file is read before anyone puts a value on it. The NAIC's Unfair Property/Casualty Claims Settlement Practices Model Regulation (#902) says each relevant document should be noted as to date received, processed or mailed. It also expects documentation detailed enough to rebuild the insurer's activities on the claim. States adopt their own versions, so check the rule that applies to you.

‍

A mature operation goes further than the minimum. Every medical record is logged with its date of service, provider and date received. A treatment chronology exists before the evaluation, not after it. A related guide explains how to build a medical chronology from a long claim file.

‍

Reserving

‍

A reserve is the amount set aside for the expected cost of a claim. Good practice starts with the first reserve. It carries a written reason tied to file evidence. Each change is logged with its trigger and reviewed above set thresholds.

‍

Reserving practice also matters to the actuary. Actuarial Standard of Practice (ASOP) No. 43 directs actuaries to consider significant changes in conditions, particularly with regard to claims. A change in how a carrier sets reserves may be such a change, so confirm with the appointed actuary. The link between early file review and reserve accuracy from day one is covered in its own post.

‍

Negotiation and litigation handling

‍

Negotiation runs from evaluation and authority to settlement or suit. Each evaluation should state a range and the evidence behind it. Authority levels are written. Every demand is logged with its date received and its deadline, as in this 30-day playbook for time-limited demands. Referral to defense counsel follows written criteria, not habit.

‍

Data and feedback loops

‍

The test here is whether the facts in the file exist as data. Representation date, venue, diagnosis and evaluation range should be fields, not lines buried in notes. Closed files are then reviewed on a fixed cycle, and the results change guidelines and training.

‍

Technology and AI governance

‍

Governance covers the tools adjusters use and the controls around them. The NAIC adopted its model bulletin on insurers' use of AI systems on 4 December 2023. It expects insurers to keep a written AI systems program, and it lists documents a state may request in an examination. The bulletin is a model, so each state decides whether to adopt it. As of 31 August 2026, 26 jurisdictions had adopted it.

‍

A well-governed operation keeps an inventory of every tool that touches a claim. Each output can be traced to the source page behind it. The professional makes the decision and records it. A related guide shows how to check a claims AI answer against the source page.

‍

The scoring table: observable evidence at each level

‍

Score each dimension on what an outside reviewer could see in the files. The table shows Levels 1, 3 and 5. Levels 2 and 4 follow from the scoring rule in Step 4 below.

‍

• Intake and triage. Level 1: Ad hoc: Facts sit in free-text notes; routing depends on who picks up the file. Level 3: Defined: Structured first-notice fields; written routing rules applied in every office. Level 5: Improving: Routing rules are tested against closed-file outcomes and revised.
• File review and medical evidence. Level 1: Ad hoc: Records reviewed as they arrive; no chronology. Level 3: Defined: Every record logged with dates; chronology done before evaluation. Level 5: Improving: Review gaps found in audits change the standard and training.
• Reserving. Level 1: Ad hoc: First reserve set by formula or habit; changes not explained. Level 3: Defined: Written reason for first reserve and each change. Level 5: Improving: Reserve changes are analysed with the actuary on a fixed cycle.
• Negotiation and litigation. Level 1: Ad hoc: Evaluation exists only as a number. Level 3: Defined: Range and basis written; demands logged with dates; written referral criteria. Level 5: Improving: Settlement results are compared with evaluations and fed back.
• Data and feedback loops. Level 1: Ad hoc: Reports need manual file reads. Level 3: Defined: Key facts captured as fields for all open files. Level 5: Improving: Closed-file reviews run on a fixed cycle and change guidelines.
• Technology and AI governance. Level 1: Ad hoc: Tools adopted team by team; no inventory. Level 3: Defined: Tool inventory; outputs traced to source; human decision recorded. Level 5: Improving: Tools are tested and reviewed on a schedule under a written AI program.

Run a claims operations assessment on your own files

‍

A claims operations assessment does not need a consultant to start. It needs a sample of files, two scorers and the table above.

‍

Step 1: Pull a file sample

‍

Pick 20 recent BI files from each team. Twenty is a practical floor, not a statistical sample. Mix open and closed files, and include litigated ones. Choose them at random, not by reputation.

‍

Step 2: Score evidence, not intent

‍

Score only what the file shows. A rule that appears in the manual but not in the file scores as absent. This is the step most self-assessments skip.

‍

Step 3: Use two scorers

‍

Have a claims manager and someone from quality or audit score each file on their own. For each dimension, they record whether the file meets the Level 3 column. Then they compare. A large gap between scorers suggests the standard itself is unclear.

‍

Step 4: Turn file results into a dimension score

‍

A team meets the Level 3 column when at least 80% of its sampled files meet it. The 80% figure is a suggested threshold to adapt, not an industry benchmark. Then score each dimension:

‍

• Level 1: no team meets the Level 3 column.
• Level 2: at least one team meets it, but not every team.
• Level 3: every team meets it.
• Level 4: every team meets it, and leaders track that rate from file data.
• Level 5: Level 4 holds, and the Level 5 column is also in place.

Take reserving as an illustration. An operation has three teams and samples 20 files from each. Team A has a written reason for the first reserve and each change in 18 files (90%). Team B has 17 (85%). Team C has 11 (55%). Two of three teams meet the standard, so reserving scores Level 2. The fix is narrow: bring Team C up to the standard the others already meet.

‍

Step 5: Read the lowest score first

‍

An average hides the weak dimension. In this model, the lowest-scoring dimension is treated as the likely constraint on the rest. Use these six checks as a quick first pass:

‍

• Could someone outside the team rebuild the history of a file from the file alone?
• Does every medical record carry a date received?
• Does every reserve change carry a written reason?
• Is each evaluation range written down with its basis?
• Can representation status and venue be pulled for all open files without reading notes?
• For each tool that touches a claim, can the adjuster see the source page behind its output?

What to fix first in a claims transformation

‍

In this model's view, a claims transformation should not start with dashboards and predictive scores. Both depend on what the files produce, so the order of work matters.

‍

First: the file and the facts it captures

‍

Dated records, a chronology and structured key facts are the foundation. Without them, reserving cannot be explained and outcomes cannot be measured. This is also where tools can help early. Tools built for bodily injury claim file review, such as amaise, process unstructured claim documents, categorise medical records and apply ICD-10 diagnosis codes. The adjuster still reads the source and makes the call.

‍

Second: one standard for reserving and evaluation

‍

Once the file holds dated evidence, write one standard for reserve reasons and evaluation ranges. Tell the actuarial team when practice changes, so the reserve review can take it into account.

‍

Third: measurement and feedback

‍

With consistent files and standards, comparisons between teams become fair. Closed-file reviews can now show which practices lead to better outcomes.

‍

Last: predictive tools at scale

‍

Predictive scores learn from past files. If those files lack dates, reasons and structured facts, the model learns the gaps. Build the foundation first, then scale the predictions.

‍

A claims maturity model is only useful if it changes what the operation does next quarter. Score the files, find the lowest dimension and fix the foundations before adding prediction. Then score again with the same method a year later.

‍

Common questions

‍

How often should a claims operation reassess its maturity?

‍

Once a year is a sensible default. Reassess sooner after a major change, such as a new claims system, a reorganisation or a new AI tool. Use the same sampling method each time so the scores compare.

‍

Can one operation sit at different levels on different dimensions?

‍

Yes. A team can have strong reserving discipline and weak data capture. Start the improvement work with the weakest dimension.

‍

Should a carrier reach Level 3 before adopting AI tools?

‍

Not always. Tools that help adjusters review files can support the foundation, if every output links to its source and the professional makes the call. Predictive scoring built on weak file data is better left until the foundation holds.

‍

Key takeaways

‍

• A claims maturity model is only as good as its evidence: score what the files show, not what the manual says.
• Assess six dimensions: intake and triage, file review and medical evidence, reserving, negotiation and litigation, data and feedback loops, and technology and AI governance.
• A team meets a level when most of its sampled files meet it; this model suggests 80% as a starting threshold.
• Fix the foundations first: dated records, a chronology and key facts captured as data.
• Add predictive tools at scale last, because they learn from past files, gaps included.