Intelligent document processing for law firms.
What it is, what it actually does to a file, and how to tell the products apart — including the question most pages in this category avoid, which is what it costs.
Twenty minutes. Nothing to install, nothing to send us.
What is intelligent document processing?
Intelligent document processing (IDP) is software that reads unstructured documents — scans, PDFs, emails, forms and photographs — and turns them into structured, checkable information: what kind of document each one is, the specific values inside it, and the record it belongs to. In a law firm it sits between the mail and the matter, doing the filing and keying that a person would otherwise do by hand before any legal work starts.
The phrase is newer than the idea. Firms have been scanning post and typing what they found into a case management system for thirty years; what changed is that the reading step stopped requiring a template. Older document automation worked by knowing exactly where on the page a value would be, which is fine for an invoice from one supplier and useless for records from four hundred different providers who each lay out a statement their own way.
Gartner published its first Magic Quadrant for intelligent document processing in September 2025 and counts more than a hundred vendors in the market. Almost all of them sell horizontally — insurance claims, logistics, accounts payable, lending — because the underlying problem is the same everywhere: work arrives as paper-shaped things and has to become fields before anything can be done with it automatically.
Law firms are a late and awkward market for it, for a reason worth naming. In accounts payable the document is the work: read the invoice, pay it, done. In a law firm the document is only ever evidence about the work, and the thing that matters is not any single document but whether the set is complete. That distinction runs through everything below, and it is where most of this category stops short.
The four things every IDP system does
Vendors describe this differently and the underlying sequence is the same everywhere. It is worth knowing because it is the most useful way to compare two products: ask which of the four they actually do, and which they hand back to you.
Classify
Decide what each document is before anything is pulled out of it — a police report, an MRI narrative, a therapy note, a lien letter, a declarations page. Nothing downstream is possible until this is right, and it is the step a scanner does not attempt.
Extract
Pull the fields that matter for that kind of document: dates of service, providers, amounts, policy numbers, claim numbers, parties, deadlines. Not the whole text — the handful of values a person would have keyed in, with the page they came from still attached.
Validate
Check the values against each other and against what is already on the matter. A bill whose date of service falls outside the treatment window, a provider spelled two ways, an amount that does not match the statement it came with — these are found here or they are found in a deposition.
Route
Put the document on the right matter, in the right place, and raise the thing a person has to decide. The output of an IDP system is not a file of text; it is work that has already been done and a short list of work that has not.
How is IDP different from OCR?
Optical character recognition turns an image of a page into text. Intelligent document processing decides what the document is, pulls out the particular values that matter for that kind of document, checks them against what is already on file, and puts the document where it belongs. OCR is one early step inside IDP — it produces characters, not meaning — which is why a searchable PDF still leaves every decision, and all of the keying, with a person.
This is the comparison worth getting right, because a great many products described as AI document tools are an OCR engine with a search box in front of it. They are not useless — a searchable archive beats an unsearchable one — but they change nothing about the work. Somebody still opens each file, works out what it is, decides which matter it belongs to, and types the dates and the amounts into the system where the firm actually keeps them.
The practical test takes one question: after this runs, is there less typing? If the answer is that the documents are now searchable, it is OCR. If the answer is that the bill is on the matter with its provider, date and amount already on the record and a note where they disagree with the statement, it is intelligent document processing.
- The characters on the page, as a block of text
- A searchable PDF, if that is what you asked for
- No idea what the document is
- No idea whether it belongs on this matter
- No idea whether anything in it is wrong
- What kind of document this is
- The specific values a person would have keyed in
- The matter it belongs to, and it is filed there
- The contradictions between it and what you already hold
- The next action, drafted and waiting on a person
What does IDP do inside a law firm?
Inside a law firm, intelligent document processing handles the intake side of every matter: it reads the records, bills, correspondence, insurance documents and forms as they arrive, files each one to the right matter without anyone sorting it, pulls the dates, providers, amounts and claim numbers out of them, and flags where two documents disagree. The work it removes is not legal work — it is the sorting, filing and keying that has to happen before legal work can start.
Take an ordinary plaintiff matter. Over its life it will collect records from several providers, bills that mostly do not match those records, an insurer's correspondence, a police report, the client's own intake answers, photographs taken on a phone, and a steady drip of documents that arrive months after anybody asked for them. None of it arrives in order. Most of it arrives at a shared mailbox.
What a firm does with that today is a person, usually the newest one, opening each item, working out which matter it belongs to, filing it, and keying the parts somebody downstream will need. It is not difficult work. It is simply endless, and because it is endless it is the work that gets deferred when the week is busy — which is how a file ends up looking complete in the case system and turning out not to be.
What gets read, and what comes out of it
Medical records
Providers, dates of service, the first and last date of treatment, diagnoses, referrals made and whether they were taken up.
Medical bills and statements
Amounts billed, amounts paid, adjustments, the provider they belong to, and whether the bill has a record behind it.
Correspondence
Who wrote, about which matter, what they asked for, what they were promised, and the date a clock started running.
Insurance documents
Carrier, adjuster, claim and policy numbers, limits, coverage positions and the dates attached to each.
Forms and intake
The client's own account of what happened, the employers and providers they named, and the authorisations that have to exist before any of the above can be requested.
Scans and photographs
Whatever is legible on them, read as images rather than as text, so a phone photo of a bill counts as the bill.
The two practices this is sharpest in are the two we have written up in full: plaintiff personal injury, where the demand goes out on whatever the file happens to contain, and public adjusting, where the carrier prices the gap between what was claimed and what was evidenced.
Does IDP replace my case management system?
No. A case management system is the system of record — matters, deadlines, contacts, notes, billing. Intelligent document processing is a layer in front of it that reads what arrives and fills it in, so the two are complementary rather than competing. Any migration being proposed is a sign you are being sold a case management system with document features, which is a different purchase with a different cost and a different risk.
This question is asked early and it is asked defensively, which is reasonable: a firm that has already migrated once knows what a cutover weekend costs, and knows that the real expense is not the licence but the six months afterwards when half the team is still looking for things in the old place.
The distinction to hold onto is between a system of record and a system of intake. There can only be one of the first — one place that is authoritative about what a matter is and when it is due — and changing it is a project. The second is upstream of it, has no authority of its own, and can be added or removed without anything moving. If a document layer is worth having, it is worth having without a migration.
Kasii is the second kind. It reads Email, Client texts, A link your client uploads on and Shared drives. It also reads from and writes back to a short list of case systems — CASEpeer, Salesforce, Litify among them — and the list is short on purpose, because every name on it is one we can show working rather than describe. What a connection actually means is on the integrations page. The first day needs none of them.
What does IDP cost a law firm?
Nobody in this category publishes a price, and the reason is not coyness: the same software costs very different amounts at two firms of the same size, because the number is driven by document volume, by how much of it arrives as scans rather than as text, and by whether setup is a project or a week. Expect to be quoted per seat, per matter or per document, expect those three not to be comparable, and expect implementation to be priced separately. The figure worth putting beside any of them is what the sorting, filing and keying costs the firm today.
We do not publish a price either, and it would be a poor page that asked this question and then quietly answered a different one. So: what can be said honestly is how the pricing works and what moves it, which is most of what a buyer actually needs before a first conversation.
Three shapes are in common use. Per seat is the familiar one and the one that scales worst for this particular job, because the work is caused by documents rather than by people — a firm that hires nobody and doubles its case load pays the same and gets twice the problem. Per matter tracks the work reasonably well and is easy to forecast. Per document or per page tracks it most precisely and is the hardest to predict, because one set of hospital records can be the size of a month of everything else.
What actually moves the number
Volume, which is not headcount
Two twenty-person firms can differ several times over in how many documents reach them a month, because it is driven by practice area and case mix rather than by staff. Anyone quoting you from headcount alone has not asked the question that decides the price.
What the pages look like
Clean digital PDFs are the cheap end. Faxed records, phone photographs and scans of scans cost more to read everywhere, in every product in this category, because they have to be read as images rather than as text.
What counts as a unit
Per seat, per matter, per document and per page are all in use, and they are not comparable. A per-page rate looks small next to a per-seat one until you count the pages in one set of hospital records.
Setup, and whether it is a project
Implementation fees, connector work and the weeks before anything is live are where the quoted number and the first-year number diverge. Ask what the first ninety days cost, not what the monthly is.
The comparison most firms reach for first is against other software, and it is the less useful one. The number that decides whether any of this is worth buying is what the firm spends today on the same work: the hours that go into sorting and keying, and the cost of the files that sit because nobody noticed something was missing. The second of those is invisible in every accounting system and is usually the larger.
What we will say about our own pricing is where it sits and why it is not on this page. Kasii is built for boutique and lower-mid-market firms rather than for the enterprise end of this category, and quotes are per firm because the volume question above has to be asked before a number means anything. It takes about twenty minutes and you do not have to send us anything to have it.
Reading what arrived is half.
The half that closes files is knowing what never did.
Every product on the page above reads the documents you have. None of them can tell you about the one that was never sent — and that is the document the file is actually waiting on. A matter does not stall because a record was misfiled. It stalls because a request was never made, and nobody could see that from a list of what is in the folder.
Kasii checks each open file against what that kind of work requires, names the gap with a person against it, and drafts the request. That is the part of this we would still argue for if somebody else did all the reading.
What this looks like in your practice.
The category is the same everywhere. What a complete file means is not — it is a different list of documents, deadlines and providers in every kind of work, which is why these are separate pages rather than one with a dropdown.
Personal injury
Records, bills, treatment gaps and liens, checked on every open matter so the demand goes out on a complete file.
See it for personal injury →Public adjusters
Scope, photos, estimates and policy language in one file, with the carrier's clock running where you can see it.
See it for public adjusting →Your practice
Your documents, your checklist, your deadlines, worked out with your team and set up the same way.
Tell us about your work →Questions people ask about IDP.
Is intelligent document processing the same as AI document review?
They overlap and they are not the same. Document review is a litigation task — deciding which documents in a production are responsive or privileged. Intelligent document processing is an intake task: reading every document that reaches the firm, whatever it is, and turning it into filed, structured, checkable information. A firm can need both, and the products that do them well are usually not the same product.
Does IDP work on handwritten notes and faxed records?
Partly, and it is the right question to ask a vendor. Printed text, including poor scans, is read reliably. Handwriting is read with much lower confidence and should be treated as a prompt for a person rather than as an answer. The useful test is not whether a system reads handwriting, but whether it tells you when it is unsure instead of guessing quietly.
How accurate is it, and what happens when it is wrong?
No system in this category is right every time, and any vendor quoting a single accuracy percentage is quoting it from their own test set rather than from your mail. What matters more is the shape of the failure: whether a low-confidence value is flagged for a person, whether the page it came from is one click away, and whether a correction sticks. Ask to see a wrong answer in the demo.
Is it safe to put privileged client documents through this?
That is a question to answer in writing, not on a website. What a firm should ask for is: where the documents are stored and in which country, who at the vendor can read them, whether they are used to train any model, what happens to them when the firm leaves, and what the vendor will sign. A vendor that cannot answer those in a document has answered them.
How long does it take to get running?
The honest range is days to months, and the variable is not the software — it is whether anything has to be integrated before the first document can be read. A system that has to be wired into your case management system first is a project. A system that can start on a mailbox or a shared drive you already have can be working on real matters the same week, with the integration following later.
Do we need to change how we work?
Not at the start, and be suspicious of a tool that requires it. Documents already arrive the way they arrive; an intake system that only works once everyone remembers to file things in a new place has moved the problem rather than solved it. The change worth making comes later, once the file is reliably complete, and it is a change to what your team spends the day on rather than to where they click.
Bring a matter that went wrong.
Twenty minutes on how documents reach your firm today, where files stall, and what a complete file would have to contain in your practice — including what this would cost you, which is a conversation rather than a page.
Nothing to install, nothing to connect, and none of your files leave your firm.