GRABB & DURANDO — MEETING PREP AI Demand Letter Project ================================================================================ INTEL FROM PRE-MEETING CONVERSATIONS ================================================================================ Style/Judgment #5 — ANSWERED (API data / training): Robert does NOT want client files used for training or leaked. He's been using ChatGPT web UI (where you upload full documents) — he doesn't yet know the API works differently. KEY POINT TO MAKE AT MEETING: The app uses the Claude API, not the web UI. Under Anthropic's standard API terms, inputs are NOT used for training. Files are sent per-request and not retained. The only open item is whether we need a BAA for HIPAA compliance (medical records). This is solvable and shouldn't block the project. STILL NEED: His verbal sign-off once he understands the API distinction. Scope #7 — ANSWERED (file access): App should read directly from the existing F: drive. He doesn't want files uploaded or saved to a second location — the case folder stays the source of truth. Confirmed: direct SMB read access from the app. Vision / Scale — ANSWERED (critical): Asked: "What does this app do for you in a year?" Answer: "I see this replacing nearly all of my legal assistants entirely." Also: He has already tested AI on pleadings and was impressed with the quality. He is not afraid of the inaccuracy problem. IMPLICATIONS FOR THE MEETING: - Robert is thinking PLATFORM, not tool. Phase 1-3 delivers demand letters. It does not replace legal assistants. Make the scope explicit so he doesn't greenlight demand letters expecting full practice automation. - The bigger vision (intake, pleadings, discovery, correspondence) is real and achievable — but that's a separate conversation. Don't promise it in Phase 1-3. Do validate that the roadmap gets there. - His low fear of inaccuracy is a risk factor for a lawyer. The fact sheet review step (Phase 1) is the safeguard. Emphasize it as a feature, not a limitation. - Consider adding a "Phase 4" framing at the meeting — after demand letters are production-ready, what's the next module? Pleadings? Correspondence? This keeps the larger vision visible without overcommitting. QUESTIONS FOR ROBERT ================================================================================ STYLE AND JUDGMENT CALLS — these define the prompt: 1. What's your rule for full-narrative vs. short-form? Is it dollar threshold, hospitalization, institutional defendant, contested liability — or just your call case by case? 2. When the AI flags an injury mentioned in the intake notes but unsupported by records, what do you want it to do? Include and flag, exclude silently, or ask staff? 3. How do you want to handle unconfirmed allegations (DUI suspicion, disputed fault)? Always omit, or present and let you decide? 4. Do you want the UIM demand generated automatically alongside the BI demand when UIM coverage is identified, or as a separate step after BI limits are tendered? 5. [ANSWERED — see Intel above. Pivot this to: explain API vs. web UI, then ask specifically about the BAA: "Do you have a business associate agreement with your health insurer clients' carriers already, or do we need to add Anthropic as a covered vendor?"] SCOPE BOUNDARIES: 6. Is the scope limited to auto/PI MVA cases, or does it include premises liability, dog bites, commercial cases? 7. [ANSWERED — direct F: drive read. Confirm: is the app machine always on the same LAN as GND-SERVER, or does it need to work from outside the office (e.g., Robert at home)?] QUESTIONS FOR JEFF ================================================================================ WORKFLOW: 1. Walk me through what you do from intake to demand — at what point do you write Jeff's Notes, and at what point does the Eval Sheet get filled in? 2. Is Jeff's Notes always a PDF in the NOTES folder, or do older or unusual cases store it differently? 3. Is the Eval Sheet always complete before demand time, or does it sometimes get filled in after? 4. What's the one thing you always know at demand time that isn't in any document in the folder? OUTPUT: 5. When you get the demand draft today, what format is it in and what do you do with it? (Word -> edit -> print? Goes into a CMS?) 6. Would you rather review a fact summary first and approve it, or just get a draft letter to redline? WHAT YOU NEED ROBERT TO COMMIT TO BEFORE LEAVING ================================================================================ [ ] 5-10 closed cases for Phase 1 testing. Mix needed: - One simple limits case - One with contested liability - One minor client - One with no police report - One with large specials (>$30K) (SWAILIEH and ORTEGA already have — need 3-5 more) [ ] Sign-off on the AI API use question — but first EXPLAIN the API vs. web UI distinction. His concern (training/leakage) is valid for the web UI; the API doesn't work that way. Then get his verbal on proceeding under API terms, and flag the BAA question for HIPAA. [ ] His answer on the style rule (short-form vs. full-narrative trigger) — this is the single most important architectural decision for the prompt [ ] Explicit scope conversation: confirm Phase 1-3 = demand letters only. Acknowledge the larger vision (practice automation). Don't let him sign off on "demand letters" while expecting "replace legal assistants." Ask: "After demand letters are solid, what's the next module you'd want?" THE PLAN ================================================================================ PHASE 1 — Fact Extractor (3-4 weeks) Point it at a case folder. Reads Jeff's Notes, Eval Sheet, and all text-readable medical records. Outputs a one-page structured fact sheet: - Liability summary - Injury list - Medical chronology - Specials table (provider by provider) - Lien status - List of everything it couldn't read (scanned docs flagged by name) Staff reviews and corrects. Nothing goes to an adjuster yet. PHASE 2 — Demand Generator (3-4 weeks) After staff approves the fact sheet, select style (short-form or full-narrative) and generate. Also generates UIM demand variant where applicable. Output is a Word doc matching the firm's current format. Compare output to 10-15 actual closed-case demand letters to tune. PHASE 3 — Live Refinement (2-3 weeks) Run on real active cases. Attorney reviews each output and flags anything off. Tune omission rules, specials presentation, and style triggers based on actual feedback. End state: attorney signs off that output is production-ready. TOTAL: 8-11 weeks. Phase 1 gives them something to react to in a month. KEY MESSAGE FOR THE MEETING ================================================================================ "The AI's job is to do the tedious reading so Robert and Jeff can focus on judgment. It will never make the call on what to include or exclude — that's always a human decision. The fact sheet review step is the checkpoint where attorney judgment replaces document parsing." GLOSSARY — ABBREVIATIONS ================================================================================ LEGAL / INSURANCE BI Bodily Injury The liability coverage carried by the at-fault driver's insurance policy that pays for injuries caused to others. This is always the primary demand target. BI limits are typically written as 25/50 ($25K per person / $50K per accident). UIM Underinsured Motorist Coverage on the client's own auto policy that pays when the at-fault driver's BI limits are insufficient to cover the damages. A UIM demand is sent to the client's own carrier after the BI carrier tenders (pays) their policy limits. UM Uninsured Motorist Similar to UIM but applies when the at-fault driver has no insurance at all. Also on the client's own policy. MedPay Medical Payments Coverage First-party coverage on the client's own auto policy that pays medical bills regardless of fault, up to the policy limit (often $5K-$10K). No subrogation right — carrier cannot recover it back. Usually collected before or alongside the BI demand. PIP Personal Injury Protection Similar to MedPay but broader; includes lost wages. Required in no-fault states; Arizona is an at-fault state so PIP is rare here. PD Property Damage Coverage for vehicle damage. Settled separately from the injury claim, usually early. PD settlement does not release the BI claim. LOA Letter of Authorization / Letter of Attorneys The rep letter sent by the firm to the insurance carrier putting them on notice that the firm represents the claimant. Required before the carrier will communicate with the attorney. DEC Page Declarations Page The summary page of an insurance policy showing coverage types, limits, named insured, policy period, and premium. Key document for identifying available coverage. LOP Letter of Protection A document the firm issues to a medical provider promising to pay the provider's bill from any settlement proceeds. Allows the client to treat without upfront payment. Creates a lien on the settlement. ERISA Employee Retirement Income Security Act Federal law governing employer-sponsored health benefit plans. An ERISA health plan has strong subrogation rights — it can recover what it paid in medical bills from any settlement. ERISA lien holders (like BCBS of Florida, Cigna) are harder to reduce than government liens. AHCCCS Arizona Health Care Cost Containment System Arizona's Medicaid program. Has a statutory subrogation right under A.R.S. § 36-2915. Lien is managed through Optum/Katch. Generally more negotiable than ERISA. A.R.S. Arizona Revised Statutes The codified laws of Arizona. Demand letters cite specific A.R.S. sections for traffic violations (e.g., failure to yield) to establish statutory negligence. TPD Tucson Police Department Most common police agency in G&D's cases. Police reports are TPD Reports. Other agencies: PCSD (Pima County Sheriff), DPS (Department of Public Safety), Marana PD, etc. PCSD Pima County Sheriff's Department DPS Arizona Department of Public Safety (highway patrol) SSCIP Social Service Contractors Indemnity Pool A specialized liability pool insuring group home operators and social service contractors in Arizona. Appeared in the Ortega case (FHL Group Home). Policy limits tend to be high ($1M+). DOCUMENT CATEGORIES R&B Records and Bills The combined file of medical records and itemized billing statements from a provider. Requested together; received together. Standard naming: "[Provider] R&B.pdf" IML Integrated Management Log HP iLO server log — records hardware events including thermal shutdowns. Referenced during VWP server investigation. OCR Optical Character Recognition Technology that converts scanned images of text into machine- readable text. Critical issue for G&D because police reports and intake forms are always image-only scans. SMB Server Message Block The file-sharing protocol used on Windows networks. G&D's case files are on an SMB share at F:\Shares\Company Data\CLIENTS\. CASE / DEMAND TERMS DOL Date of Loss The date the accident occurred. Central reference date for the entire case — all medical chronology, statutes of limitations, and insurance correspondence are measured from DOL. MVA Motor Vehicle Accident Standard abbreviation for an auto accident case. PI Personal Injury The practice area. Encompasses MVA, premises liability, dog bites, etc. G&D's primary practice. Specials Special Damages Quantifiable economic losses: medical bills, lost wages, out-of- pocket costs. Contrasted with "generals" (pain and suffering). Demand letters present gross billed specials by provider. Generals General Damages Non-economic losses: pain and suffering, loss of enjoyment of life, emotional distress. Not itemized; argued as a multiplier of specials or as a standalone value. Eggshell Eggshell Plaintiff Doctrine Plaintiff Legal principle that a defendant takes the plaintiff as they find them. If the plaintiff had a pre-existing vulnerability (prior injury, thin skull, fragile health), the defendant is still fully liable for all harm caused, even if a "normal" person would have been less injured. Robert invokes this in cases with pre-existing conditions or unusual severity. Subrogation The right of an insurer or lienholder to recover what they paid (in medical bills or benefits) from a third-party settlement. Health insurers (BCBS, AHCCCS) assert subrogation rights against PI settlements. Lien A legal claim against settlement proceeds. Common lien holders: health insurer subrogation, MedPay, hospital LOP, attorney fees. Must be resolved before net settlement can be distributed to client. Tender When an insurance carrier pays out their policy limits in full. "Mendota tendered their limits" means Mendota paid their full $25K BI policy. This triggers the UIM demand to the client's own carrier. CMS Case Management System Generic term for practice management software. G&D uses a system (DDT = DUI Defense Team is a separate module) for tracking cases, contacts, and documents. Demand The settlement demand letter sent to the at-fault carrier's Letter adjuster. Summarizes liability, damages, medical treatment, and specials. Does not state a dollar demand — asks the carrier to respond with an offer. Standard G&D practice. Adjuster The insurance company employee who evaluates and negotiates the claim. The demand letter is addressed to the adjuster by name (or claim number when no adjuster is assigned). TECHNOLOGY API Application Programming Interface A way for software systems to communicate. The Claude API is what the app uses to send documents to the AI and receive the fact sheet / demand letter output. BAA Business Associate Agreement A HIPAA-required contract between a covered entity (or its business associate) and a vendor who handles protected health information (PHI). Required if sending medical records to cloud services like Azure or Anthropic. HIPAA Health Insurance Portability and Accountability Act Federal law protecting the privacy of medical information. Relevant because G&D's case files contain medical records. JSON JavaScript Object Notation A structured data format used for the Stage 1 fact extraction output. Allows staff to review extracted facts in a readable table before the demand letter is generated. ================================================================================ Generated: 2026-05-13 | Arizona Computer Guru LLC | Grabb & Durando Project ================================================================================