Everything a plaintiff firm loses to admin, handled. Click any of these.
Legal is one of the industries Mayeshal runs in, and it is the one with the most finished work in it. Nineteen firms, direct relationships, in this vertical since 2019, and a vertical pack running in production today at a personal injury firm on Litify.
If you are a plaintiff personal injury firm, everything on this page is shipped and you can see it read one of your own matters in thirty minutes.
If you are a different kind of firm, the honest position is this. The baseline platform is not personal injury specific: the mirror of your case system, the document layer, search inside document contents, page citations, deadlines read from real calendar events and tasks, risk flags, the assistant, the audit log and the cost meter all work against any matter type. What is personal injury specific is the pack: the medical records work, the treatment timeline, the demand engine, the settlement verifier. Those would be built for your practice area with you as the first customer, and we would price and scope that conversation honestly rather than implying it already exists.
Firms we work with today run plaintiff personal injury, workers compensation, medical malpractice, mass tort, lemon law across thirteen or more state jurisdictions, litigation finance and general litigation. The platform work is the same. The pack is what differs.
79% of the legal industry now reports using AI in some form, rising to 93% at mid-sized firms. Almost none of that is governed at firm level. That gap is the whole product.
AI gets better the more of the matter it can see. That is the whole mechanism, and it is why buying five separate legal tools does not add up to one good one.
The medical indexer cannot see the negotiation history. The intake product does not know what the client said on the phone last week. The demand tool has never read the adjuster correspondence. Each answers confidently from a fraction of the file.
Status and date fields on a matter go stale the moment somebody forgets to update them, while the real schedule sits in calendar events and tasks that most integrations never enumerate.
Every additional tool is another copy of protected health information, another agreement to review, another security questionnaire, and another permission model rebuilt loosely or not at all.
When a bar authority or a client asks where an answer came from, five products means five partial logs and no one trail to follow.
Police reports pasted into a message, records scanned crooked, photographs of an intersection. A tool that only reads clean PDFs reads a fraction of what is actually in the matter.
Two tools produce two numbers, nobody knows which is right, and staff go back to doing it by hand. Most legal AI rollouts die here, months after the invoice cleared.
That last number is not a typo and it is the one worth staring at. Reading, indexing and summarising a hundred and twenty-seven gigabytes of medical records cost less than a deposition transcript. The $156.55 is what the build actually booked: $149.16 for the 37,852 summaries, about $0.0039 each averaged across the whole corpus, plus $7.39 for the vision pass over the photographs. On a sampled run a single summary was measured at $0.0046. Optical character recognition runs on our own hardware at no per-page fee, so the 331,486 pages added nothing. A 287-page binder extraction costs about thirty cents.
The remainder is worth naming too, because a vendor who only quotes the good number is not telling you much. Around 1,879 documents no longer exist at the source, in either the original or the rendered copy. Around 1,800 were never documents at all: video, audio, archives, application files. Five were zero bytes. That is the entire gap. Of everything the document system will actually serve, coverage is 100%.
It is graded against your firm's own closed matters, not a generic model, with a value band and a statute of limitations check. The grade is a recommendation and a person makes the call. What you get back is a queue sorted by what deserves attention first, instead of a list sorted by arrival.
They are opened, read page by page and summarised inside the hour, without anyone printing anything. The provider, the treatment window, the billed amounts and the ICD codes come out as structured lines with a page citation behind each one, and a coverage score that tells you honestly what it could not find.
The chronology shows a nine-week gap the carrier is going to argue about, flagged red, with the documented explanation from the file sitting inline next to it rather than in someone's memory.
The letter builds in your firm's own conventions, from your own clause library, with the expense table assembled from your own damage records and every provider matched to the records that support the line. The packet assembles itself with a table of contents mapping each provider to an exact page range, a divider page per provider, and the real record PDFs behind them.
The negotiation watchdog has already found every matter where the last counter-offer never got a counter-demand, sorted by how long it has been sitting, and it costs nothing to run because it never calls a model.
The pack composes onto the whole ground, the whole spine and all six jobs. Each capability below carries the objective it was configured against, what it actually does, and the operational need that put it in scope. Status: live, in production at one firm, pack version 1.5.0, signed.
Objective. Give every person on a file one screen that answers where things stand, assembled from the record, the documents and the calendar rather than from a status field.
The record snapshot puts parties, stage, money, dates and what is missing on one screen. Risk flags surface on the header rather than being found later. Deadlines are read from the calendar events and tasks in the source system, because the summary field on the record goes stale. Every claim opens the document and the page it came from.
Requirement: a Texas personal injury firm on Litify, where a case manager watched the system miss every real deadline on her active file while 116 calendar events were visible the whole time.
Objective. Make every document in every file readable, searchable to the page, and summarised, at a cost the firm can audit.
44,956 documents catalogued, 331,486 pages indexed, 127 GB, 37,852 AI summaries at a measured $0.0046 each. A text layer where there is one, optical character recognition where there is not, and a vision model for photographs. Search inside contents returns the document and the page number, and the hit opens the document at that page. Documents are held in the tenant's own storage after first read, so a file already read stays openable while the source system is down.
Requirement: a Texas personal injury firm on Litify, whose own name for the problem was that the AI could not point at the page it got an answer from.
Objective. Produce the firm's own two demand documents in the firm's own conventions, assembled from the firm's own records, with the packet built from the real record PDFs.
Two structurally different letters, not one with a flag, derived from 763 of the firm's own past demands: the five section structure, the verbatim record sentence present in 696 of them, the expense table, the per provider divider pages, and a hard deadline on one type against a soft deadline on the other. The packet assembles the letter, a table of contents mapping provider to exact page range, the dividers and the real record PDFs from our own store. Exactly one part is written by a model, at temperature zero, so the same facts produce the same words. Everything else is arithmetic and the firm's own language.
Requirement: a Texas personal injury firm on Litify, whose existing letters carry no exhibit numbers and no Bates numbers, so inventing an exhibit scheme would have made our output look unlike everything they have ever sent.
Objective. Cover the rest of the pre litigation file.
Intake triage scored against the firm's own closed matters with a limitations check; medical records summaries with a page citation behind every line and a coverage score naming what could not be found; treatment timelines and gap detection; adjuster correspondence; a negotiation watchdog that finds every file where the last counter offer never got an answer and costs nothing to run because it makes no model call; a settlement verifier that recomputes the closing statement to the penny; disbursement tracking with its five queues; and weekly task hygiene.
Requirement: a Texas personal injury firm on Litify, ten capabilities specified by the client and verified against their production formulas before build.
Every intake graded against the firm's own closed matters, with a value band and a statute of limitations check. A person always decides.
ShippedExtraction with page citations and a coverage self-score, then a staged, approved write of damage rows back into your case system.
ShippedThe chronology with the gaps flagged, and the documented explanation for each gap shown inline.
ShippedThe letter, the expense table and the assembled binder, built from your own templates and your own records.
ShippedNightly, no model call, no token cost. Every file where the last counter-offer never got a counter-demand, ranked by days waiting.
ShippedDrafts in your firm's voice. Sending is always approval-gated, and the platform never dispatches on its own.
ShippedThe closing statement recomputed independently from the same formula chain, to the penny, with the offsets reconciled and every workaround audited.
ShippedThe checks-to-print queue with requested, received, printed, deposited and cleared states, and days-to-print on every row. No model call.
ShippedNative pulls plus one AI paragraph per case manager. Not a digest of everything, a digest of what changed.
ShippedDuplicates, tasks on closed matters, misroutes. Detected with no model call, fixed in bulk behind an approval.
ShippedA records tool reads the records. A demand tool reads the demand. Neither of them reads the email where the adjuster conceded liability, or the photograph of the intersection, or the police report that was pasted into a message instead of attached. Mayeshal reads all of it, because it is one system looking at one file.
If the document has a text layer, we use it. If it does not, optical character recognition runs on our own hardware, which is why the per-page cost is zero rather than a line on a bill. If the page is a photograph, a vision model describes what is visible and nothing more. The status of every page records how it was read, so a rasterised recovery is never mistaken for the original.
It describes what is in the picture. It does not assign fault, does not infer causation, does not rate injury severity and does not name a person from their face. Those are the lawyer's calls, and a confident wrong answer in a case file is worse than no answer.
Type a phrase and get page-level hits across the matter. Click a hit and the document opens at that page. In one test on a single matter, twelve page hits came back from 9,339 pages.
Once we have read a document we hold a copy in your tenant's own storage. Opening it takes about three hundredths of a second instead of one to two seconds, and more importantly it still opens when your case system is throttled or down.
We read 763 of the firm's own Stowers and underinsured motorist demands before writing a line of the generator. What we found decided the design.
There is not one exhibit letter, exhibit number or Bates stamp anywhere in any of those letters. The entire record citation apparatus is a single sentence that appears verbatim in 696 of them. Providers appear only as rows in a past medical expenses table with a dollar figure. Inside the assembled packet, each provider's records sit behind a one-page divider bearing only the provider name.
So the generator emits exactly that. No exhibit numbering, because inventing one would make the firm's output look unlike everything the firm has ever sent, and would need an attorney sign-off that the current practice does not.
A Stowers demand and an underinsured motorist demand are structurally different letters. Stowers carries the evidentiary banner, five Roman numeral sections, the Stowers line of authority and a hard five o'clock expiry. An underinsured motorist demand has no banner, no Roman numerals, a permission-to-settle paragraph, the insurance code and a soft deadline. They are generated as two separate documents because that is what they are.
Exactly one part of the letter is written by a model: the incident narrative, at temperature zero so the same facts produce the same words every time, grounded in the police report summaries already in the corpus. The expense table, the totals, the recipient resolution and the letterhead are arithmetic and lookup. If there is no key or no facts, it writes a placeholder and tells you, rather than inventing a paragraph.
Letter, then a table of contents mapping each provider to an exact PDF page range, then a divider page per provider, then the actual record PDFs from your own storage. The page ranges are computed during assembly, so they are always right. The largest packet we have assembled ran 1,508 pages across 23 contiguous ranges.
Every demand runs a contamination check on approval, on packet generation and again on send, and it fails closed. If a draft contains an identifier belonging to another client's file, it cannot be approved, cannot be packaged and cannot be sent. If the checker itself is unavailable, the send is refused rather than allowed. We tested it by deliberately seeding a draft with a former client's name and confirming it was refused at every gate, with the category and the source template named in the error.
For a period, the assistant answered deadline questions from a trial date field on the matter. A case manager watched it miss every real deadline on her active file. The field was stale, and the real calendar was sitting in the calendar events and tasks the whole time, including 116 calendar events on that one matter.
The reason it was invisible is worth telling, because it is the kind of thing that separates an integration from a demo. Calendar events and tasks each carry a lookup back to the matter that does not appear in the matter's own list of child relationships. An engineer enumerating child relationships carefully would still have missed it. Only a reverse sweep of all 807 objects in the org finds it.
So we stopped hand-picking objects. There is now a discovered relationship graph, 129 relationships, rebuilt nightly, and the deadline panel reads the real calendar. If a field disagrees with the calendar, the screen says so rather than picking one.
Coverage is pre litigation. This list is here because a vendor who will not tell you what is finished is telling you something, and you would find all of it in week two anyway.
It also does not replace your case management system. Your platform stays the system of record, keeps its permissions and keeps its data. Mayeshal reads across it and writes back only through an approval.
The pack connects the case management platform and its document system. Everything else on this list is the baseline platform doing what it does in every industry.
Read-only authorisation into the platform that holds your matters, mirrored into a database that belongs to your tenant alone with count parity verified against the source. It stays the system of record. Live in production on Litify today.
Mayeshal does not replace your case management system and it does not compete with it. Your platform stays the system of record, the permission model you already built stays in force, and nothing is written back into it without a person approving it.
Every attachment opened page by page, indexed, summarised and citable to the page, then held in your own storage so a file already read stays openable when the source system is slow.
The email thread where the adjuster conceded liability and the calendar events that carry the real deadlines, read alongside the matter rather than separately from it.
Agent and skill status, call lists, dispositions and interactive voice response paths, so a matter's phone history is part of the picture rather than a separate report.
Micronetbd builds and operates outbound dialer campaigns and contact centre integrations on NICE CXone, RingCentral, Dialpad, CallRail and Genesys alongside case management systems at plaintiff firms today: campaign construction, agent and skill configuration, disposition design, screen pop behaviour and reconciliation back into the case system. That build work is a Micronetbd implementation service rather than a Mayeshal feature, and we keep those two things separate on purpose.
Every write back into a source system is staged, whitelisted and approved by a person, and it enforces the calling user's own permissions in that system.
Delivered in your implementation
Voice agents for intake and records chasing. Approval-gated outbound dispatch of demands, adjuster letters and notices. A litigation suite covering discovery, complaints, deposition outlines, motions and mediation briefs. Legal research with citation verification. Production optical character recognition for scanned and handwritten records.
Implementation runs ninety days and these land inside it, so a firm starting now has them from go-live.
Provisioning, identity and single sign-on, connecting your case system, your document store and your mail and calendar, configuring the models, training, go-live. Four weeks on Growth, eight on Professional, twelve on Enterprise, depending on how many systems there are.
Week one: install, role mapping to your firm's own titles, and a field-mapping review against your actual configuration rather than a generic one. Week two: verifier and watchdog calibration on your real closed matters, an extraction trial on three live binders, approval rules, training, production authorisation, go-live.
Then ninety days of hyper-care, and we report weekly active users against the 70% bar at every checkpoint.
Thirty minutes, on a screen share, on a matter you pick. We connect nothing, install nothing and ask for no credentials to have that conversation. If it is not obviously useful in the first ten minutes, we will tell you so and you get your afternoon back.