Scanning and routing is half a solution. Every billable thing your firm produces — the analysis, the reports, the work product — begins at intake, and none of it can be better than what intake handed it. ArcuityVizion's Intelligent Ingestion starts an authenticated, audit-proof data journey that turns documents into data, and data into billable hours — so your people spend the day using the record instead of hunting through it.
This is a medical-legal report as Arcuity reads it. Hover any highlighted passage to see what was found, how it was reached, and exactly where in the record it lives.
The applicant is a warehouse associate who reports a specific injury while lifting a 60-pound carton from a floor-level pallet. She felt an immediate onset of low back pain radiating into the left buttock, temporally consistent with the described mechanism of injury.BH1 She continued the shift and reported the injury to her supervisor the following morning.
MRI of the lumbar spine demonstrates a 4 mm broad-based disc protrusion at L5–S1 with mild bilateral neural foraminal narrowing. There is no evidence of nerve root impingement or frank radiculopathy on the current study.NEG Multilevel facet arthropathy and disc desiccation are noted, consistent with degenerative change.
In my opinion, 60% of the applicant's current lumbar disability is attributable to the industrial injury, and the remaining 40% is apportioned to pre-existing, non-industrial degenerative disc disease.§4663
The basis for this apportionment percentage is the patient's age and the radiographic degenerative findings.!R Permanent and stationary status was reached on the date noted below.
Applying the AMA Guides, the applicant warrants a DRE Lumbar Category II rating, yielding 8% whole person impairment prior to apportionment.AMA
The applicant was seen by a primary treating physician on eleven occasions. Records from a prior 2019 lumbar complaint are referenced but were not provided for review.GAP
The other half is why you bought it. Clean intake solves the first hour of a claim file and none of the rest: the mail is filed the same day, and the chronology still costs an associate a weekend. The prior-injury history goes unbuilt. The brief on everything unrelated to the injury does not get written, because it never does. The records leave for the evaluator without anyone confirming every discrete finding is in the package.
A document that has been named and filed is still a document. Until it becomes data — authenticated, addressable, and linked to every other document in the file — nobody can bill against it. That conversion is the whole job, and it either happens at intake or it does not happen at all.
Grouped by where each falls in the life of a claim. The first band is what most platforms sell as their entire product. We treat it as the gateway — because every one of the fourteen below it, and every hour you bill against them, depends on the file being turned into data properly the first time.
Everything arriving by mail, email, fax, portal or scanner. Consolidated files split into their constituent documents, each typed, matched to its matter, named to your convention, filed to the right folder and put in front of the right person. Every page is made Bates-addressable and content-hashed on the way in, so the chain from original byte to final citation is intact before any analysis runs.
Any platform can name a file. The part that matters is that this is where documents become data — authenticated and audit-proof from the first touch. Everything in the five bands below is only as defensible as this one.
Identify every custodian the record implicates, and generate the subpoena to get their file.
The records nobody remembered to request are the ones that decide apportionment.
Prepare and execute filing and service of the resulting documents.
Closes the loop between deciding to request something and it actually being served.
Assemble the evaluator's records package and the interrogatories that accompany it.
The evaluator answers the questions you send. A weak question set is a weak report you then live with.
Evaluate what is about to leave the firm for a complete history of every discrete finding — before it reaches the evaluator.
A pre-flight almost nobody runs. A package missing a discrete finding produces an evaluation you cannot use, and you find out months later.
Reduce medical records, legal filings and government forms to what bears on one or more specific medical-legal concepts tied to the injury.
Turns a 4,000-page file into the file for this question — without discarding the rest.
Surface all issues bearing on causation, rather than the ones a reviewer happened to notice.
Causation is where the case is won, and it is buried across hundreds of pages by definition.
Evaluate the mechanism of injury on its own terms and test it against the account in the medical record.
A claimed mechanism the physics does not support is a defense case.
Every injury in the record on one timeline, prepared for attorney review for subrogation or apportionment.
Both turn on what happened before, and both fail on an incomplete chronology.
Evaluate medical reports for where the rating or methodology strays from the Guides.
A rebuttal generator. Drift is the most reliable basis for challenging an impairment rating, and finding it costs an expert and a weekend.
Full-history screen across the record, including prescription drug history — not just the claim window.
Pre-existing condition evidence is exactly what apportionment requires, and the claim file never volunteers it.
An indexed brief of all documents unrelated to the claimed injury, with verbatim summarization and a table of contents.
Proving what the record does not contain is how you close off an alternate theory — and it is the most tedious job in the practice, which is why it gets skipped.
A draft outline built from what the record actually contains.
Preparation compressed from days to a review pass.
Deposition testimony compared against the entire timeline, producing a statement of every inconsistency — hyperlinked and supported with verbatim page quotes.
Testimony that contradicts the record is the highest-value thing in any file. Here it arrives already cited to the page.
Demand or response, drafted from the analyzed record.
The valuation argument arrives assembled and cited, not reconstructed from memory.
Prepared and ready to file.
Procedural work that consumes attorney time and generates no fee.
Reading the file is half of it. Arcuity turns what it finds into the next action — in your calendar, your inbox, your chat, your case file — so nothing waits on somebody noticing it.
Requested evaluations, hearings and trial dates are run against your actual week — conflicts flagged in red, open slots proposed in green — so a QME or AME gets scheduled against real availability instead of sitting in an inbox until someone reads it.
Every deadline, request, objection window and obligation in the file, extracted and assigned — not summarized, assigned.
When something needs a second set of eyes, it starts the conversation with the relevant people, with the document and the finding already attached.
Where the record references something it never produced — prior treatment, an earlier award, unauthorised care, an unnamed provider, missing wage or benefit history — the request is written for you: what is missing, why it matters, and the stamped page that raised it. Anything that cannot be cited to a page is held back for review rather than asked without support.
Every provider, employer, carrier and agency the record implicates, compiled into a service list with the subpoenas ready to go.
The same body part in a 2019 record, a PR-2 and a deposition answer are linked to one another, and every version of the evidence base is tracked — so a contradiction three documents apart is visible instead of theoretical.
Where a utilization review modification or denial is inconsistent with MTUS as applied to that claimant's documented course, the grounds are drafted with the passage that shows it, the IMR eligibility noted and the 30-day appeal deadline calendared.
Generic document AI is trained on everybody's files and belongs to the vendor. Every correction your team makes quietly improves a model your competitors are also using. Meridian is designed to invert that — giving each firm its own adapter, shaped by your documents, your naming conventions, your work product, and your senior people's judgment about what actually matters in a file.
Being straight about status: Meridian is the next layer, in active development and not yet serving client adapters. We are telling you about it now because the design decisions behind it are the ones worth judging a vendor on — and because nobody else in this market is even promising them.
Its own database schema, its own storage, its own model weights, per client. No other firm's corrections reach your adapter, and nothing your people teach it leaks into anyone else's.
Each adapter is a numbered artifact — trained on a schedule, evaluated against a held-out standard, and promoted only if it beats the version it replaces. Name the version behind any output, and run last year's file against last year's adapter to get last year's answer.
It improves what the system proposes. What the system decides stays with deterministic code and the qualified human who signs — a rule that holds today and does not relax when Meridian arrives.
That last card is the whole point, and it is in force right now. A platform whose classifier retrains continuously on everyone's corrections cannot reproduce its own past output or tell you which model reached a conclusion. An adapter that is versioned, isolated and barred from deciding gives you the opposite: a system that learns from you without ever drifting away from you.
Document-AI platforms built for legal intake are good at what they do, and most firms should have one. The differences below are not about accuracy at filing. They are about whether anything produced from the file can be defended after it leaves the building.
| Typical document-AI platform | Arcuity | |
|---|---|---|
| Scope | Inbound handling — capture, classify, match, file, route | Inbound handling, plus the outbound work product, plus the action it triggers |
| What you receive | A correctly named document in the correct folder | A subpoena, an evaluator package, a chronology, an inconsistency statement, a demand |
| What it does next | Notifies a person that a document arrived | Proposes the date against your calendar, drafts the cited adjuster inquiry, opens the Teams thread, builds the custodian list |
| Deadlines | Surfaces a date it found on the page | Extracts every obligation, checks availability, and proposes the calendar entry |
| Across the file | Each document handled on its own | Correlations and Claim DNA link issues across every document, so a contradiction three files apart is visible |
| Medical review | Classifies a record as a medical record | Drafts the UR appeal grounds where a determination is inconsistent with MTUS, with the passage that shows it |
| Whose model learns | One shared model retrained on every customer's corrections — the asset that improves belongs to the vendor | Meridian gives each firm its own isolated adapter, shaped by your files and your people's judgment (in development) |
| How it learns | Continuously and silently, with no version you can name | Numbered adapters, evaluated against a held-out standard and promoted only on improvement (in development) |
| How answers are reached | Classifiers that retrain on user corrections, so behaviour shifts over time | Deterministic algorithms decide. Learning improves the proposal; it never gets a vote on the finding |
| Reproducibility | Re-running last year's document may not reproduce last year's answer | Same file in, same result out — engine and contract version recorded on every artifact |
| Provenance | A confidence score | Every statement resolves to a page, a Bates number, coordinates and a content hash |
| Completeness | Ranked lists, top results surfaced | Nothing is dropped. We rank, group and section — never truncate |
| Unknowns | A blank field | UNKNOWN is a value, and it is distinct from "nobody checked" |
| Finality | Output is output | Nothing is marked final until every completeness and legal check passes. Otherwise it says what is missing, on its own face |
| Image handling | One processed copy of the page | Two lineages — the evidence original is never the enhanced copy, and every transform is logged |
Fix intake alone and the backlog does not disappear — it relocates. The mail is filed same-day and the analysis queue grows instead. Covering the whole run takes the drag out of the whole run, and for a firm that bills by the hour there are three places it lands.
The hours are already being spent; they are simply unbillable, and carriers will not pay for sorting and sifting. Every hour moved from administration to case work is an hour that can be invoiced.
Your best paralegal stops doing clerical work. The firm scales caseload without scaling headcount, and the people you most want to keep spend their day on work worth keeping them for.
The Munger brief. The completeness check. The inconsistency statement. The Guides review. Not deferred — never attempted, because nobody had the hours. A savings pitch has a ceiling. This does not.
A subpoena, an inconsistency statement or a demand is only worth having if it survives what comes back. These are architectural rules, not settings — each written against a specific way this work fails.
Give us a single claim file and we will return the work product — the chronology, the causation issues, the completeness check — with every line cited to a stamped page. Then decide.
