The teardown
Everything we’d automate in your firm.
Every system on this page runs in production today at a Social Security disability firm we operate, a firm that doubled its revenue in about seven months on top of this build. Here is what each piece would do in your office, what a human still touches, and the math on what it carries. Most agencies would call this proprietary. We’d rather you see exactly what you’d be buying.
Why we're publishing this
Nothing here is secret in a way that protects us. Anyone can have the idea of automating mail. The value sits in the few hundred small decisions buried under each system: what happens when a letter matches no case, when the same client exists under two spellings, when a scheduled job dies at 2am and nothing looks wrong. You can’t copy those off a page.
So the page costs us nothing, and it saves you a discovery call full of vague answers. If your firm has someone who can build all of this from what’s written here, you should let them.
The map
Grouped the way you’d meet them in your own office: intake, mail, client communication, and the plumbing underneath all of it. Every one of these is live in production right now.
Intake
Getting a client in the door is probably your biggest pile of retyping. It stops existing.
Lead intake
A new lead submits your intake form, and the data lands standardized: the document folder tree builds itself, the contact appears in your CRM. Nobody retypes a name. Where this runs today, bringing a client on dropped from about 45 minutes of staff time to under 3.
Matter creation
When a lead signs, the system checks whether that person already exists (people come in twice more often than you'd think, once as Bob and once as Robert), matches or creates the contact in Clio, opens the matter, and hands it to whichever of your teams is carrying the lightest load. Your staff see a ready case appear instead of a form to key in.
Signed documents
Every completed signature envelope files itself to the right folder, logs on your tracking sheets, and announces itself to your intake team. A person only touches the ones that don't match anything.
The intake watchdog
A finished signup should leave three traces: the qualification form, the document packet, and the signature. The watchdog matches all three per person, and when one is still missing an hour later, it emails your team naming the person and the missing step. Stalled signups stop being something you find by accident weeks later.
SSA mail is relentless, and almost all of it is mechanical. This is the system we'd build first at any disability firm, yours included.
Mail triage
Every piece of your scanned mail gets read, classified into one of 17 letter types, filed to the right folder, matched to the case, and logged, with a note or task added on the matter where one is needed. This alone gave one paralegal back about ten hours a week.
Client mail routing
Filed mail then moves into each client's own folder, matched by name even when names refuse to cooperate: maiden names, hyphens, misspellings. Anything the system can't match confidently goes to a person instead of being guessed at.
Client communication
A worried client who hasn't heard anything calls your office, and every one of those calls pulls a paralegal off real work. These systems exist so your firm speaks first.
Case milestone updates
When a case reaches a stage a client is actually waiting to hear about, a hearing requested, a decision in, an appeal filed, the client gets a text and an email without anyone in your office remembering to send one. This is the system that kills the "did anything happen with my case?" call.
Exam confirmations
When a client confirms their consultative exam by text, the attendance letter is generated, dropped into the right team's folder, and the team gets notified. A five-minute chore, times every appointment on your calendar.
Dire-need outreach
Clients whose situation may qualify for SSA's dire-need expedite get asked about it directly, by text. A reply opens a task for your staff with the expedite letter ready to send. Staff handle everything after the reply; the system's only job was remembering to ask.
Surveys and check-ins
Thirty days into an active case, clients get a two-way text survey. At six months, a check-in. Replies land on the case, where your team actually looks.
Low-rating alerts
Any rating of 3 or below texts whoever you choose, the same day, and a human picks up the phone. The machine's whole job here is making sure an unhappy client is never discovered three weeks later in a spreadsheet.
The plumbing
The least interesting systems on this page, and half of everything above depends on them.
The mirror
A read-only copy of your whole case management system, refreshed daily, that every other system reads from instead of hammering the live API. When something needs an answer about a case in milliseconds, this is where the answer comes from.
The deduplicator
It finds person records that are probably the same human being and produces a review for sign-off. Detection is automated. Merging client records is not something we let software decide on its own, and you shouldn't let anyone's software decide it either.
The math, so you can argue with it
A partner taking this to the other partners needs a number they can defend, so ours is built from the bottom. The systems run about 38,900 times a month, and each kind of run is timed against what the task takes a person. That comes out to roughly 1,962 hours of work a month, about 12 full-time employees’ worth of output.
Put a rate on it and you get the dollar figure. At a $22 an hour loaded admin rate, it’s about $200K a year. Price those hours at what the roles actually cost and it lands closer to $520K. If you think our per-task timings are generous, our deliberately conservative model, which only credits time we can defend line by line, still lands around 760 hours a month.
A partner can argue the rate, and we hand over the rate precisely so they can. The run counts come out of production logs, and we walk through them on the audit call.
What stays human
None of this decides anything legal. Nothing approves or denies, nothing files with the agency, nothing discusses the merits of a case with a client. When a client is unhappy, a person calls them. When a letter is ambiguous, a person reads it. When two records might be the same human being, a person confirms it before anything merges. The systems exist so those people find out fast, with everything they need already in front of them.
What breaks
Here’s the part that never makes it into a pitch: this stuff decays. State agencies reword letters. A county starts stamping documents differently. A form grows a new field and an integration quietly starts dropping it. A system that ran clean for months will eventually meet a piece of mail it has never seen.
That’s why everything deadline-sensitive in this build pauses on what it doesn’t recognize instead of guessing, and why the whole thing is monitored from outside itself. The enemy in back-office automation is silent failure, not error messages. It’s also why this is an ongoing relationship rather than an install. Whoever builds yours, make sure someone is still watching it in month six.
Security and privilege
The confidentiality question gets its own page. The short version: client data is isolated per client and encrypted, none of it is used to train AI models, privilege is handled in writing, and the setup maps to ABA Formal Opinion 512 and Texas Opinion 705. The details are at /security.
Could you build this yourself?
Genuinely, maybe. The stack isn’t exotic: webhooks out of the tools the firm already uses, the practice management system’s API, a small database, scheduled jobs, and a language model for the reading and classifying. A capable ops person working with a developer can stand up a first version of the mail sorter in a few weeks.
What we’d warn you about, from running this in production: authentication tokens expire in chains and take five systems down at once. Name matching is far harder than it looks, because real humans arrive with maiden names, nicknames, typos, and the occasional twin. Scheduled jobs die silently, and unless something watches the watchers, nobody notices for days. And the second version of every system is the real one, built after production teaches you what the first version missed.
None of that is a reason not to try. It is the actual job, though. The sorting was never the hard part.
What to ask anyone selling you this, including us
- Ask to see production run logs, not a demo. A demo proves it worked once.
- Ask what happens when the system hits something it doesn't recognize. The right answer involves a person. "It handles everything" is the wrong answer.
- Ask who owns the code and what happens to your systems if you leave.
- Ask where client data goes and whether any of it is used to train an AI model.
- Ask what broke last month. Anyone running real systems has an answer. Anyone who says nothing breaks isn't running real systems.
- Ask how they would know if a system quietly stopped running. If the answer amounts to "the client tells us," walk.
Common questions
Does law firm automation actually work?
At the disability firm running this build, the systems carry roughly 1,962 hours of work a month, about 12 full-time employees' worth, and the firm doubled its revenue in about seven months on the same back office, tracking toward 4x. The numbers come from production logs, and the free audit runs the same math on your operation.
What does law firm automation cost?
It depends on what you automate and what that work is worth to your firm, which is what the free audit figures out. We price against the outcome and put the target number in the agreement before anything gets built. The math on this page, hours carried and what those hours cost at a real labor rate, is the same math we run on your operation.
Will automation replace my staff?
The firm in this teardown didn't cut anyone. The same team now carries a much larger caseload because the retyping, sorting, chasing, and reminding run on their own. Most of the hours the systems carry were being done late, at the margins, or not at all before, which is why nobody had to lose a job for the firm to gain the capacity.
Why custom code instead of Zapier or Make?
Zapier and Make are fine for low-stakes connections between clean tools. Law firm work stops being clean fast: scanned mail nothing can read, one client entered under two spellings, a deadline buried in an oddly worded letter. Custom code handles those cases properly and routes what it can't handle to a person, and in a law firm the exceptions are exactly where the risk lives.
Want this mapped onto your firm?
Answer eleven quick questions and get the automation map for your own operation: where the hours are going, what to build in what order. Free, on screen, about three minutes. Or skip straight to the free audit and we’ll do it from your real volumes.
Get your firm’s map