AI and malpractice case reserves: support, not the number
Answer in brief: AI can gather the facts that move a malpractice case reserve (injury severity, treatment since the event, billed and paid amounts, wage records) and watch the file for events that should trigger a reserve review. The reserve amount and its written rationale stay with the claims professional who has authority, and aggregate reserves stay with the actuary.
Limits: We read each vendor's own web pages and the cited studies on 2026-11-14 and did not upload records to any tool, ours included. Nothing here is legal, actuarial or clinical advice; privilege and reserving rules vary by state and by program.
Case reserves and aggregate reserves: who sets what
2 different numbers get called "the reserve", and AI touches them differently.
| Case reserve | Aggregate reserves, including IBNR | |
|---|---|---|
| What it is | The estimate of what 1 known claim will cost to resolve, usually split into indemnity and expense | The estimate for the whole book: known claims, development on them, and claims not yet reported |
| Who sets it | The claims professional with reserve authority, often approved by a committee above a threshold | The actuary, for the captive, trust or carrier |
| What drives it | The facts of this claim: injury, liability view, damages, venue, stage | Claim history, payment and reporting patterns, exposure |
| Where AI helps | Gathering and updating the facts from the file | Outside the scope of this site; actuarial software is a separate category |
This page covers case reserves on hospital-side malpractice claims. If your program is self-insured or runs a captive, the actuary uses your case reserves as an input, so better-documented case reserves help them too.
The inputs AI can gather from the file
Every item below is a fact in the records. A record review tool can find it, cite the page and keep it current as new records arrive. None of them is the reserve.
| Input | Where it sits in the file | What to check |
|---|---|---|
| Nature and severity of the injury | Discharge summary, later provider notes, imaging reports | That the tool reports the latest status, not the worst day |
| Baseline before the event | Prior primary care and specialist records | That prior conditions are dated and cited, not inferred |
| Treatment since the event | Rehab, home health, readmissions, surgery recommendations | Recommendations vs care actually given |
| Medical expenses | Bills, EOBs, lien notices | Billed vs paid; whose bills are included |
| Work and earnings | Employment records, disability forms, tax documents if produced | Requested vs received |
| Life expectancy markers | Diagnoses and prognosis statements in the records | Leave the projection to a qualified expert |
| Liability signals | Reviewer opinion, deposition testimony vs chart | Use the reviewer's and counsel's words, not the tool's |
Worked example: a reserve review sheet
This is a synthetic example, no real patient or claim. It continues the synthetic PCA case from the incident review page. Every dollar figure is invented. The point is the format: facts with pages, then a human rationale.
| Item | At ECA (month 2) | After depositions (month 14) | Source |
|---|---|---|---|
| Condition | Discharged home after 3 ICU days; no lasting deficit charted | New memory complaints; neuropsychology testing ordered | Discharge p. 210; clinic note p. 1,912 |
| Medical expenses, billed | $61,400 | $88,900 | Bills pp. 2,401 to 2,488 |
| Medical expenses, paid | $23,750 | $31,200 | EOBs pp. 2,489 to 2,530 |
| Work | Retired | Retired | Social history p. 14 |
| Liability signals | Reviewer: monitoring gap concerning, causation unclear | Nurse testimony contradicted on pump change by 2 entries | Reviewer letter; transcript 44:19; pp. 89, 96 |
| Reserve and rationale | Written by the claims lead, not the tool: what changed, why the estimate moves, who approved it, and the date. | ||
2 lessons from the format. First, the deposition contradiction is a liability signal only because counsel and the reviewer say it is; a tool that flags it is doing its job, but the weight is a judgment. Second, the billed and paid columns differ by more than half. A tool that reports only billed amounts can push the estimate in the wrong direction, so ask for both.
Events that should trigger a reserve review
This is the part of reserving where AI may help most: watching a growing file for changes a busy claims lead might not see for weeks. Ask whether your tool can alert on these when new records arrive:
- A new diagnosis, surgery recommendation or readmission in later records.
- A death, a transfer to long-term care or a new disability determination.
- A demand, a changed demand or a mediation date in counsel's report.
- Deposition testimony that conflicts with the chart on a point the reviewer flagged.
- A new expert disclosure, or your reviewer changing their view.
- A trial date set or moved.
Litigation management tools already pull demands and dates out of counsel's reports; CaseGlide describes doing this and syncing the facts into the claims system (read 2026-11-14). Record review tools cover the medical side. Few hospitals need both if counsel reports are short.
Benchmarks, models, and where to be careful
Some data products offer claims benchmarks for reserving. Candello, the malpractice claims database from CRICO, lists analytics to "set reserves" and "price cases for settlement" for its subscribers, and RLDatix's claims module tracks reserves, settlements and case expenses (both read 2026-11-14). Benchmarks tell you how similar coded claims resolved. They don't read your chart.
Be careful with any tool that offers a predicted reserve or settlement number for a single claim:
- Ask what it was trained on. Auto and workers' comp claims behave differently from hospital malpractice. Several claims AI vendors on the tool table use auto and workers' comp as their examples.
- Ask for the reasons. A number without the facts behind it can't go in a reserve rationale, and your auditors and actuary will ask.
- Watch for anchoring. Once a number appears on screen, people adjust from it. Look at the facts first and the prediction last, if at all.
- Keep the context. Published payment figures age fast. The Studdert study found average payments of $521,560 on claims involving error and $313,205 on claims without, but those are 2006 figures from 5 insurers (N Engl J Med), not a benchmark for today.
Questions about AI and reserving
Can AI set a malpractice reserve?
It shouldn't. AI can gather the facts and flag changes, but the case reserve is an estimate that someone with authority has to own and explain. Keep the tool's output as the fact base and write the rationale yourself.
How often should a case reserve be reviewed?
At the early case assessment and whenever a material fact changes, plus on your program's fixed schedule. The trigger list above is where a tool watching the file can help.
Does better record review change reserves?
It changes the facts behind them. A complete, cited picture of injury, baseline and expenses at month 2 means fewer surprises at month 14, which is what reserve stability depends on.