Personal injury is a volume business pretending to be a relationship business. The successful firms run more like an insurance back office than a law office — intake centers, case managers, settlement schedulers, demand-letter mills. That structure is exactly what AI is best at unlocking, and the firms that are honest with themselves about it are pulling away from the ones still pitching billboard equity.
We have built or assisted on AI rollouts at four PI firms in the last year. Different sizes, different states, different case mixes. The patterns repeat.
Four bottlenecks AI clears today
1. Intake conversion
PI intake is the single most expensive funnel stage and the most under-instrumented. A typical firm pays $400 to $1,500 per qualified lead and converts 30% to 50% to a signed engagement. Most of the loss is operational — the lead called at 10pm and reached an after-hours service, the intake specialist did not have the right script for a soft tissue claim, the follow-up SMS went out 36 hours late. A voice and SMS agent that handles after-hours, qualifies on the standard intake script, and books the human attorney call lifts conversion 8 to 15 points in our experience. Cost-per-signed-case drops accordingly.
2. Medical record summarization
The single most labor-intensive part of building a demand package is the medical chronology. Hours of paralegal time per case, often more than once as new records arrive. Modern document models — properly fenced, with citations back to the source page — collapse this from days to hours. Two cautions. First, the human still has to verify; the model is fast but cites incorrectly often enough to be a risk. Second, the records often contain protected information that should never leave a controlled environment. Pick your stack accordingly.
3. Demand letter drafting
Demand letters follow a tight template that varies by jurisdiction and insurer. A model trained on the firm's prior demands, fed the case facts and the medical chronology, drafts a competent first version in minutes. The attorney edits, adds the legal theory, and signs. Cycle time per demand drops 40 to 60%. Settlement values, in our limited data, hold or improve — likely because the demand letter quality is more consistent.
4. Case status and client communication
PI clients want to know where their case stands. Most firms answer this with batch updates that are usually late and rarely specific. A retrieval-grounded assistant that pulls from the firm's case management system, drafts a status email in plain language, and queues it for attorney approval lifts client satisfaction without lifting attorney workload. The economic impact is in referrals and reviews, both of which are leading indicators that lawyers chronically under-measure.
Where the partners still have to be in the loop
- arrow_rightLiability theory and case strategy. The model can summarize precedent; it cannot decide which strategy to pursue.
- arrow_rightSettlement negotiation. The model can suggest a number; the negotiation itself is human work, every time.
- arrow_rightAny communication that affects fiduciary duty or attorney-client privilege. Drafted by AI is fine; sent without a human review is not.
- arrow_rightTrial preparation. AI-assisted research saves time. The witness prep, jury narrative, and closing argument are not AI work.
A typical 90-day rollout
- arrow_rightDays 1 to 30 — Wire the after-hours intake agent. Measure conversion lift against the prior baseline. Decide whether to expand to business hours.
- arrow_rightDays 31 to 60 — Add medical record summarization with human-in-the-loop verification. Track paralegal hours saved per case.
- arrow_rightDays 61 to 90 — Add demand letter drafting on a single insurer or case type. Measure cycle time and average settlement value.
The KPI that actually matters
Most firms watch billable hour metrics that do not really apply to contingency-fee work. The number that actually matters is fee per case-month — the average legal fee earned divided by the average months a case stays open. Every successful AI rollout we have run moves this number, sometimes dramatically. If you are tracking anything else, you will struggle to defend the program at year-end.
PI law is not going to be replaced by AI. The firms that treat it as a glorified word processor will get the gains they paid for and not much more. The ones that treat it as a structural rethink — intake to demand to settlement — are the ones whose growth is decoupling from their marketing spend. That is the moat.
