Fixed Costs and Thin Margins

Low-value claimant personal-injury work operates under an economic constraint which is easily obscured by discussion of legal principle. Much of the work is not difficult law. It is process: obtaining information, opening a file, collecting documents, arranging evidence, monitoring deadlines, drafting routine communications, updating the client, negotiating and closing the matter. Under fixed recoverable costs, the firm does not receive more from the opponent because it has performed that process inefficiently. Margin depends upon doing the necessary work accurately and with as little wasted labour as possible.

That makes the area an obvious candidate for automation. The useful question, however, is not whether AI can “run a personal injury claim”. That is too vague to be worth answering. A claim consists of a sequence of tasks, some mechanical and some requiring judgment. The sensible exercise is to separate them.

At intake, technology can collect a structured account, identify missing information, extract dates and create a provisional file record. It can ask for the registration number which the client forgot to provide, identify that no earnings evidence has been uploaded and flag that the accident date creates an approaching limitation issue. Much of this is administrative work and there is no particular professional virtue in requiring a human being to copy the same information between three systems.

The same is true of document handling. Medical records, photographs, receipts and correspondence can be sorted, named, indexed and searched. Dates can be extracted into a working chronology. Different documents can be compared and inconsistencies flagged. A properly designed system can be tireless about missing information in a way which human beings, after the hundredth routine file of the week, occasionally are not.

But the character of the task changes when judgment begins. A machine may identify that a medical record contains an earlier complaint involving the same part of the body. It does not follow that the earlier complaint breaks the chain of causation. It may identify that two accounts differ. It does not decide which account is credible. It may suggest a valuation range, but settlement advice involves more than selecting the middle figure produced by a model. The point at which the routine file ceases to be routine is precisely the point at which professional value becomes important.

The same distinction should govern drafting. Routine letters, information requests, first versions of schedules and other structured material are obvious candidates for automation. Court-facing documents, witness evidence and final advice require a different degree of control. The fact that a machine has produced a polished witness statement is not an answer to the question whether the statement accurately reflects the witness’s evidence. In litigation, elegance is not a substitute for truth, although lawyers have occasionally behaved as though it were.

The technological architecture matters less than the workflow. A firm does not need the most fashionable model. It needs systems which know what they are permitted to do, work with appropriate information, keep records of what has happened and stop at points where human approval is required. The most valuable “AI system” may therefore be less glamorous than a demonstration chatbot. It may simply be a well-designed process which moves routine files forward, identifies exceptions and puts the difficult cases in front of the right person early enough to do something about them.

That also explains why uncontrolled use of consumer tools is a poor substitute for institutional adoption. A low-value PI file contains precisely the material which lawyers ought not to distribute casually: medical evidence, financial information, identification documents and confidential instructions. The important question is not whether a particular chatbot is clever enough to summarise them. It is whether the organisation has approved the system for that purpose, understands what happens to the information and has proper controls over access and use.

The commercial opportunity is nevertheless substantial. Fixed-cost work rewards consistency. AI is particularly good at structured repetition. If a firm can automate intake, document processing, routine drafting, chasing and workflow management while ensuring that limitation, liability, causation, valuation, vulnerability and court-facing accuracy are escalated to human lawyers, it can reduce production cost without pretending that professional judgment has disappeared.

That is the sustainable model. Not a robot solicitor conducting a personal-injury practice with no adults in the room, but a highly automated production process interrupted deliberately at the points where somebody needs to think.

An earlier version of this article appeared in PI Focus in May 2026.

Original version published in PI Focus, May 2026 (PDF)