The problem
Before a demand goes to the insurer, someone reads hundreds of pages of ER, imaging, physical therapy and specialist records, builds a treatment chronology and writes the facts and injuries sections of the demand letter. If a firm sends 10 demands a month and a case manager spends 4 hours on each chronology and 2 hours on the narrative, that's 60 hours a month.
What the AI does
Claude reads long documents whole: its help center says a chat takes up to 20 files and PDFs of up to 1,000 pages. On the Team plan, which Anthropic says doesn't use business data for training by default, the case manager uploads the records the client authorized and asks for a dated treatment chronology with the page number for every entry, plus a list of gaps in treatment and any earlier injuries mentioned. The case manager checks every entry against the cited page. Claude then drafts the facts, treatment and impact sections of the demand from the checked chronology and the attorney's notes, and the firm's CASEpeer demand template supplies the medical specials from the bills logged in the case. The attorney edits, sets the demand amount and signs. Claude never decides what the case is worth.
A worked example
A personal injury firm in Tampa with three attorneys and four case managers sends about 10 demand packages a month and already uses CASEpeer for case management and its demand template. Two case managers who prepare demands use Claude Team, billed annually.
Before
4 hours on each chronology and 2 on the narrative: 60 hours a month
After
About 1.5 hours checking the chronology and 1 hour editing the draft: 25 hours a month
What it's worth
About $1,220£917€1,092A$1,761C$1,725 a month of case manager time, against a $40£30€36A$58C$57 cost. Saving 3.5 hours on each of 10 demands is 35 hours, at an assumed $35£26€31A$51C$49 an hour. Claude Team is $20£15€18A$29C$28 per seat a month billed annually or $25£19€22A$36C$35 monthly, with a 2-seat minimum; the firm's existing CASEpeer plan (from $90£68€81A$130C$127 per user a month) is not counted. The 3.5 hours is an assumption to test on closed files first.
How to set it up
- Test on three closed cases: compare Claude's chronology with the one your team built and note what it missed.
- Use only Claude Team seats for client files, never personal or free accounts, and record the decision under ABA Formal Opinion 512, including whether your engagement letter tells clients the firm uses AI tools.
- Ask for every chronology entry with its date, provider and source page, and check each one against that page.
- Draft the narrative only from the checked chronology and the attorney's notes, and keep the CASEpeer template as the source for bills and liens.
- Have the attorney edit the full demand, set the amount and sign it; nothing goes to an insurer without that review.
What goes wrong
- A chronology can miss a pre-existing condition or a gap in treatment that the adjuster will find. Read the relevant records yourself before the demand goes out.
- Claude's help center says it reads images as well as text only in PDFs of 100 pages or fewer; from 101 to 1,000 pages it reads the text alone. Split scanned records into files of 100 pages or fewer, and check handwritten notes by hand.
- On a free or Pro plan, business data protections differ from Team; client records belong only on a plan with business terms. Team needs at least 2 seats.
- AI drafts can overstate injuries or invent a detail. Every factual statement in a demand must trace to a record, under the state bar's rules on truthfulness to others (ABA Model Rule 4.1).