Valuing a personal-injury file is not an isolated arithmetic operation: it is a chain that runs from the medical expert report to the schedule of losses, and software only has value if it holds the whole chain. Most tools stop at valuing one head or at searching for decisions. The selection criterion is therefore coverage: from reading the exhibits to justifying each amount.
The Dintilhac classification (nomenclature Dintilhac, the court-endorsed classification of the heads of personal-injury loss) has no binding force, but it structures practice and the schedule. Valuing means going through four steps, and each calls for different capabilities.
You first have to extract the parameters from the expert report: date of consolidation, permanent functional deficit (déficit fonctionnel permanent, "DFP") rating, degree of souffrances endurées (the pain and suffering endured up to consolidation, rated 1 to 7), periods of temporary functional deficit (déficit fonctionnel temporaire, "DFT"), third-party assistance needs. These elements are written in natural language, in a layout specific to each expert.
Then you have to break down by head, distinguishing temporary from permanent losses, and pecuniary from non-pecuniary. Then value each head, which draws, depending on the case, on a reference guide, a capitalisation table for lifetime heads, and comparable case law. Finally, summarise on a schedule that can be put before the other side, where each line can be justified.
A tool that only handles the third step leaves the user the two most time-consuming ones.
| Family | What it does well | Its limit |
|---|---|---|
| In-house spreadsheet | Controlled calculations, tailored to the firm's habits | Manual re-entry, no case law, errors hard to spot |
| Online calculator or simulator | Quickly applies a formula or scale to one head | Starts from hand-entered data, does not read the file, does not justify |
| Vertical valuation AI | Reads the exhibits, breaks down, values and ties each amount to its source | Assumes a quantified case-law base and a serious security framework |
Generalist assistants form a fourth category, better treated apart: they write well, do not calculate reliably and can produce a non-existent case-law reference. The subject is developed in our article on the reliability of ChatGPT for law.
Traceability. Each amount must be tied to the exhibit it comes from: page of the expert report, comparable decision, reference guide used. A valuation that cannot be traced cannot be checked and cannot be produced.
Access to quantified case law. A reference guide gives an indicative range. What makes the difference in the discussion with an insurer is being able to put up real decisions handed down at comparable rating, degree and age.
Handling lifetime heads. Third-party assistance, future loss of earnings, future costs: these heads are capitalised. Several capitalisation tables coexist, and the choice of table changes the result. The tool must state the one it uses and let you change it.
Third-party payers' recourse. Recourse is exercised head by head, on the sole indemnities repairing the losses covered, excluding losses of a personal nature (article 31 of law no. 85-677 of 5 July 1985). A tool that does not distinguish the head amount from the balance going to the victim produces a wrong schedule.
The volume of exhibits. A serious personal-injury file, with its expert reports, certificates and supporting documents, quickly exceeds several hundred documents. Generalist assistants cap the number of exhibits analysed in a single session, which forces you to split the file.
The security framework. This is the point most specific to this litigation, developed below.
A medical expert report, operating notes, an initial medical certificate: the personal-injury file is made up of health data. Its hosting on digital media is subject to certification, under the conditions set out in article L1111-8 of the Public Health Code.
Added to this is professional secrecy: the consultations a lawyer sends to their client, the correspondence with them and, more generally, the file are covered by it (article 66-5 of law no. 71-1130 of 31 December 1971). Before loading exhibits into a tool, three questions deserve a written answer: where is the data hosted, who has access to it, and is it reused to train a model?
Plato hosts data in France, in an HDS environment (sovereign hosting certified for health data), and client data is not used to train any model.
Software does not choose the strategy. It does not decide whether to challenge the rating retained by the expert, request a fresh assessment, or accept an offer. It does not sign the submissions. It does not promise an amount: it gives a method, comparables and a justification, which is already the longest part of the work.
That is the dividing line to keep in mind when a vendor talks about prediction. A serious tool says where its figures come from, not what the judge will decide.
Take a closed file, whose outcome is known, and submit it to the tool. Three checks are enough: are the parameters extracted from the expert report correct, is the schedule complete head by head, and is each line justified by a source you can consult? The real time saving is measured after the checking, not before.
Plato automates this chain for personal injury, on a base of more than 84,000 quantified decisions, and for employment law. To place a file against comparable decisions, the base can be searched at jp.plato.legal.
For the method itself, see our personal-injury valuation guide, the nomenclature Dintilhac head by head and how to find quantified case law.
Which software should you use to value the nomenclature Dintilhac heads?
One that covers the whole chain: reading the expert report, breaking down by head, valuing with a reference guide and comparable case law, then a justified schedule. A calculator that starts from hand-entered data handles only one step out of four.
Is a spreadsheet enough?
For the calculations, often yes. For the rest, no: it forces re-entry, brings no case law and makes errors hard to spot, especially on capitalised heads.
What is the difference between a calculator and a vertical AI?
The calculator applies a formula to entered data. The vertical AI starts from the file exhibits, extracts the parameters, values each head and ties the amounts to their source.
Can you use a generalist AI to value a personal-injury file?
It is not advised: no quantified case-law base, a risk of invented references, a cap on the number of documents analysed and no guarantee on the file's health data.
Where should the exhibits of a personal-injury file be hosted?
With a host meeting the requirements applicable to health data, hosting on digital media being subject to certification under the conditions of article L1111-8 of the Public Health Code. The lawyer's professional secrecy adds to this requirement.
Does software handle third-party payers' recourse?
It must, on pain of producing a wrong schedule. Recourse is exercised head by head, on the sole indemnities repairing the losses covered (article 31 of the law of 5 July 1985).
Which capitalisation table should the software use?
Several tables coexist and the choice changes the result on lifetime heads. The requirement is not that a tool impose a table, but that it state which one it applies and let you change it.
Can software guarantee the amount obtained?
No. It improves the method and the justification, not the outcome. No tool can promise an amount of compensation.