A five-attorney PI firm in Indianapolis has a good problem. The intake phone rings. Referrals come in from former clients. The Google Ads campaign works. Cases walk through the door every week.
The bad problem: the firm refers out a third of them.
Not because the cases lack merit. Because the firm lacks processing capacity. A multi-defendant auto wreck with six treating physicians, two surgeons, an IME report, and a subrogation lien from a self-funded ERISA plan requires 40 to 60 hours of paralegal time before an attorney even touches the demand. A single mass tort screening with 200 potential claimants buries the office for a month. A med-mal adjacent PI case with complex causation analysis needs someone who can read radiology reports and cross-reference treatment timelines against the mechanism of injury.
The managing partner does the math. Refer it out. Take the 33 percent referral fee. Move on.
That math is wrong. And private AI changes it.
The cases you are referring out
Every small PI firm has a threshold. Below the threshold, the case is routine. Soft tissue. Clear liability. One or two treating physicians. The paralegal pulls records, builds the demand package, and the attorney sends it out. Settlement or litigation. The firm handles it in-house.
Above the threshold, the case gets complicated.
Multi-defendant cases. Three vehicles. A trucking company with an independent contractor driver. A municipality that failed to maintain the intersection. Each defendant has separate counsel. Each files cross-claims. The discovery load triples. Your two paralegals cannot track three sets of interrogatories, three deposition schedules, and three separate demand packages simultaneously.
Med-mal adjacent PI. A surgical complication following a car accident. Was the nerve damage caused by the impact or the surgery? Answering that question requires reviewing the complete surgical record, cross-referencing pre-operative imaging against post-operative findings, and mapping the treatment timeline against the injury mechanism. That analysis takes a paralegal with medical records experience 20 to 30 hours. Most small PI firms do not have that paralegal.
Mass tort screenings. A local contamination case surfaces. Two hundred residents call in over three weeks. Each needs an intake questionnaire, medical record retrieval, exposure timeline documentation, and preliminary causation screening. The firm that can process 200 screenings in 60 days gets the cases. The firm that cannot refers them to a mass tort mill and takes a fraction of the fees.
Complex lien resolution. A catastrophic injury case settles for $1.2 million. The client has Medicare conditional payments, a Medicaid lien, a workers' comp subrogation claim, and an ERISA plan asserting full reimbursement rights. Each lien requires separate negotiation. Missing one means personal liability for the attorney under the Medicare Secondary Payer Act. Most small firms hand these to a lien resolution company and pay 15 to 20 percent of the recovered reduction.
In every scenario, the firm loses revenue. Referral fees. Lien resolution fees. Cases that never come in the door because the firm's reputation is "they do the simple stuff."
The bottleneck is not intake. It is processing.
What private AI makes possible
Dhakma Core is not a research tool. It is a case processing engine that sits in your office like any other appliance. Pre-configured. Self-contained. No IT department required. Dhakma installs it, supports it, and keeps it running. Your team uses a secure internal web interface. Here is what it changes for a PI practice.
Medical record summarization
The problem: Your paralegal receives 800 pages of medical records from four treating physicians, a hospital admission, an ER visit, and two specialists. They read every page. They build a chronological treatment summary by hand. They extract ICD-10 codes, CPT codes, and billing totals. They flag gaps in treatment. They identify pre-existing conditions the defense will seize on. This takes 15 to 25 hours per case.
With Dhakma Core: The system ingests the complete medical record set and produces a structured treatment timeline. Every visit, diagnosis code, procedure, and provider is mapped chronologically. Billing totals are extracted and cross-referenced against the records. Gaps in treatment are flagged automatically. Pre-existing conditions are identified and separated from accident-related diagnoses. The paralegal reviews and validates the output instead of building it from scratch. Time: two to four hours.
What this unlocks: You can now take the med-mal adjacent PI case. You can now process the multi-defendant wreck with six treating physicians. The medical record analysis that made those cases uneconomical for a small firm is no longer the bottleneck.
Demand letter first drafts
The problem: Every demand letter follows a pattern. Statement of facts. Liability analysis. Medical treatment summary. Specials itemization. General damages argument. Per diem calculation or multiplier analysis. Your paralegal writes each one from scratch, copying and pasting from prior demands, adapting the language, and hoping the specials math is right. A comprehensive demand on a moderate injury case takes 8 to 12 hours.
With Dhakma Core: The system pulls from your firm's demand letter templates and prior successful demands. It generates a first draft incorporating the specific case facts, the structured medical summary it already produced, the itemized specials with running totals, and a damages argument calibrated to the venue and injury type. The attorney reviews, adjusts the tone and strategy, and sends. Draft production time: under an hour. Attorney review: two to three hours. Total: a fraction of the old timeline.
What this unlocks: Your senior attorneys stop proofreading specials spreadsheets and start doing what they are actually good at: crafting the core legal argument, reading the adjuster, and preparing for negotiation. A five-person firm can maintain 15 to 20 active demands simultaneously instead of 8 to 10.
Lien identification and tracking
The problem: Your paralegal opens a case file and starts tracking liens by hand. They call Medicare to check for conditional payments. They request an ERISA plan's summary plan description to determine reimbursement rights. They chase down workers' comp carriers. They build a spreadsheet. They update it manually. They miss the Medicaid lien because the client did not disclose it during intake. The attorney learns about it at settlement and faces personal liability.
With Dhakma Core: The system scans every document in the case file. EOBs, correspondence, intake forms, medical bills, insurance cards. It identifies every potential lien source and builds a tracking matrix with lien type, asserted amount, statutory basis, and key deadlines. When new documents are added to the file, the system updates the matrix automatically. Nothing hides in a stack of unsorted mail.
What this unlocks: Lien resolution stays in-house. The 15 to 20 percent you pay a lien resolution company stays in the firm's pocket. More importantly, no lien surprises at settlement.
Settlement valuation research
The problem: An attorney evaluating a settlement offer relies on experience, memory, and maybe a Verdict Search subscription. What did similar cases settle for in this county? What are the jury verdict trends for this injury type? What is this adjuster's pattern? The answers live in the attorney's head and in scattered databases.
With Dhakma Core: The system indexes your firm's complete case history. Every settlement, every verdict, every demand, every counter. When evaluating a new case, query by injury type, county, carrier, or adjuster. The system surfaces comparable outcomes from your own files. Over time, it builds a dataset that no subscription service can match because it reflects your practice, your venues, and your results.
What this unlocks: Junior attorneys make better settlement recommendations. The managing partner stops being the only person in the firm who knows what a case is worth.
The ethics of cloud AI in plaintiff work
This section is brief because the analysis is simple.
A PI case file contains medical records protected by HIPAA. Social Security numbers. Financial records. Employment history. Insurance policy details. Photographs of injuries. Mental health treatment records.
ABA Model Rule 1.6(c) requires reasonable efforts to prevent unauthorized disclosure of client information. Uploading a client's complete medical record set to a cloud AI vendor's servers does not satisfy that standard. It does not matter what the vendor's privacy policy says. The data leaves your control the moment it leaves your building.
In February 2026, Anthropic confirmed that state-linked Chinese AI laboratories extracted 16 million queries from Claude through 24,000 fraudulent accounts. The breach vector was the cloud connection.
Dhakma Core has no cloud connection. The air-gap switch is a physical disconnect. No software exploit bridges a hardware gap. Your client's medical records, SSNs, and financial data never leave the machine sitting in your office.
This is not a compliance argument. It is the only defensible position.
One case pays for it
Forget the per-hour savings calculations. Here is the math that matters.
Your firm refers out a complex multi-defendant auto case. The case settles for $400,000. Your referral fee is $133,000. The referring attorney's fee is $267,000. You did the intake. You signed the client. You handed $267,000 to another firm because you could not process the medical records fast enough.
With Dhakma Core, you keep that case. Your fee is $133,000 (one-third contingency). The system cost runs approximately $5,000 per month. One case you keep instead of referring out covers the entire annual cost and then some.
Now multiply that across every case above your current complexity threshold. Two cases per quarter that you keep instead of referring out. That is $200,000 to $500,000 in annual revenue your firm is currently giving away.
The question is not whether you can afford private AI. It is whether you can afford to keep referring out the cases that would transform your practice.
See it on a closed case file
We do not ask for commitment before you see results. Bring a closed case file. A completed demand package. A settled case with medical records, liens, and a final demand letter.
We run it through Dhakma Core on-site. You watch. Your paralegal compares the system's output against the work product they produced by hand. The medical record summary. The lien matrix. The demand letter draft.
If the output does not make your next staff meeting a different conversation, we take the hardware home.
Your firm's complexity threshold is not a ceiling. It is a constraint that private AI removes. The cases you refer out today are the cases you keep tomorrow.