AI for PI Lien Resolution: Services vs Software vs Agent

By Jude Lee · · Comparison

Attorney and paralegal reviewing medical bills and a settlement disbursement statement in a small law firm office

Where the settlement phase actually stalls

Ask a small PI firm where the file dies and almost nobody says “liability.” They say the post-settlement phase. The check is in trust, the client is calling weekly, and a paralegal is on hold with a benefits coordinator trying to confirm whether a balance of a few hundred dollars — the figure here is illustrative, not from any firm’s data — was already written off.

The work itself is mostly clerical reconciliation: pull every provider from the treatment chronology, match each to a bill and a current balance, identify which balances are actually asserted interests versus unpaid invoices, request reductions, log responses, and then produce a disbursement statement that ties out to the penny.

That is exactly the shape of work AI agents handle well — high-volume document extraction and repetitive correspondence — and exactly the shape of work where one transposed digit becomes a bar complaint. So the question isn’t “should we use AI here.” It’s which of three delivery models fits your volume.

The three realistic options

Outsourced lien resolution service

What you get: a vendor (ARCHER Systems, Ametros and similar firms operate in this space) takes assignment of the lien portfolio, handles Medicare conditional payment work, negotiates, and delivers a final lien package. Usually priced per lien or as a percentage.

Fits: firms with mass-tort or high-volume inventories, complex Medicare Set-Aside exposure, or no paralegal bandwidth at all.

Costs: fee comes out of the settlement, which your client notices. You also lose visibility — the file status lives in someone else’s system, and your client still calls you.

In-house AI agent over your own files

What you get: an AI assistant connected to your case-management and document systems that extracts providers and balances, drafts reduction letters, tracks responses, and prepares a draft disbursement statement for attorney review.

Fits: firms doing steady auto/premises volume where most liens are provider balances, health-plan subrogation, and the occasional Medicare interest.

Costs: build and maintenance effort, plus real governance work. It does not remove the need for a human who understands lien law.

The third option is the one most firms already own and underuse: your PI practice-management platform. CASEpeer, Litify, and Clio all track settlement and lien line items, and most practice-management vendors have shipped some AI summarization feature by now. If your lien volume is modest and your bottleneck is organization rather than extraction, configuring what you already pay for beats any new build. Say that out loud before anyone writes a statement of work.

What an agentic lien workflow actually looks like

The useful mental model: the agent assembles the packet, the lawyer signs it. Concretely, on a settled file:

  1. Settle the confidentiality question first

    Medical bills and records are protected health information, and lien files are full of it. Before a single PDF is uploaded, confirm three things: whether your AI provider will sign a business associate agreement (several enterprise AI offerings do, consumer tiers generally do not — check the vendor’s current documentation rather than a blog post); what the data-retention and model-training settings are on the specific account your staff will use; and whether your engagement agreement and HIPAA authorizations actually contemplate third-party processing of the client’s records. ABA Model Rule 1.6 and your state’s analogue, plus your state bar’s technology-competence guidance, are the governing frame — confirm the specifics with a qualified professional before you roll this out firm-wide.

  2. Extract the provider universe

    The agent reads the records and bills already in your document system and produces a provider list with dates of service, billed amounts, and payment sources. This is the same extraction backbone used in AI medical record review and chronology work — if you’ve built that, you’re most of the way here.

  3. Classify the interest

    Each line gets tagged: unpaid provider balance, health plan subrogation claim, statutory hospital lien, government payer, or already-adjusted. Classification rules are firm-specific, which is why this belongs in a skill — a packaged, versioned instruction set that makes the agent reason the same way on every file instead of improvising.

  4. Draft the outreach

    Balance verification letters, reduction requests, and plan document demands, each pre-filled from matter data. A human sends them.

  5. Reconcile responses

    Incoming letters and portal statements get parsed back against the ledger, with a flag whenever a stated balance differs from the firm’s figure.

  6. Prepare the draft statement

    The agent assembles a disbursement statement draft with gross settlement, fee, costs, each lien line, and net to client — clearly labeled DRAFT, with every figure traceable to a source document.

  7. Attorney verification and signature

    The lawyer checks the math, the client signs, and the money moves through your trust accounting process — not through the agent.
An agent that drafts the disbursement statement is useful. An agent that moves money is a malpractice claim waiting for a cause number.

Where extraction breaks, and the spot-checks that catch it

Models pull provider names, service dates, and dollar figures out of clean PDFs reliably. Four things break them, and each deserves a standing human rule:

Why a handful of liens eat most of the timeline

On settlement files, the familiar pattern shows up with unusual clarity: most lien line items are small provider balances resolved with one letter, while a minority — government payers and self-funded ERISA plans asserting plan-document reimbursement rights — consume the bulk of the calendar days.

That’s your automation targeting rule. Point the agent at the long tail of routine balances so your paralegal’s attention is free for the two or three interests that actually need negotiation. Don’t try to make an agent handle a Medicare conditional payment dispute. Per the Centers for Medicare & Medicaid Services, conditional payment recovery runs through its Benefits Coordination & Recovery Center and the Medicare Secondary Payer Recovery Portal, with its own timelines and appeal process; confirm the current procedure directly with CMS rather than trusting any model’s recollection of it.

How firms are actually using AI on these files right now

Search “what AI systems do law firms use” and you’ll get tool lists. What we typically see in the small and mid-size firms we talk to, as of 2026, is three layers running at once: general assistants (Claude, ChatGPT, Copilot) for drafting and summarization; AI features embedded in the practice-management and litigation platforms the firm already licenses; and, in a minority of firms, custom agents wired into firm systems through MCP — the open Model Context Protocol that lets an assistant query your document management or case system under your permissions instead of through copy-paste. Treat that as an editorial observation, not a measured distribution; the ABA Legal Technology Survey Report and the Clio Legal Trends Report are where to look for actual adoption data.

The large legal-AI platforms — Harvey, CoCounsel, Legora — are aimed primarily at document review, research, and transactional work. Lien resolution is firm-specific plumbing, so it rarely shows up as a packaged product feature. That’s the usual signal that a custom MCP server over your own matter data is worth evaluating — not because custom is better in principle, but because no vendor is going to model your referral sources, your reduction templates, and your ledger conventions for you.

4 lien categories
Government payer, ERISA/private plan, statutory provider lien, unpaid balance — each needs its own handling rule
General practice categories; verify specifics against CMS guidance and plan documents
Rule 1.15
ABA Model Rule on safekeeping client property — the guardrail on any disbursement automation
ABA Model Rules of Professional Conduct

Modeling the payback without making up numbers

Don’t start from a vendor’s savings claim. Start from your own file log. Pull your last twenty settled matters and record, per file, the staff hours spent on lien work and the elapsed days from settlement to disbursement. Then estimate what share of those hours is pure extraction and correspondence versus genuine negotiation.

Your model is: (automatable hours per file × files per year × loaded hourly cost) + (value of faster disbursement, if your firm carries case costs) − (build and maintenance cost). Hours × rate is the only formula worth putting in a build decision — plug in your own figures, and if the answer is marginal, configure your existing software instead. Our automation ROI walkthrough covers the same arithmetic in more detail.

Two numbers matter more here than any benchmark: realization per settled file, and cycle time from settlement to disbursement. Slow lien work delays fee recognition and keeps advanced case costs tied up, which is the real cost of a paralegal on hold. Readers hunting for national lawyer-income figures are usually trying to answer this question sideways; your own realization and cycle time per settled file will tell you far more than any compensation distribution will.

Where is your firm losing billable hours?

Get a free automation audit: we map your intake-to-invoice workflow and show you exactly what's worth automating — before you spend a dollar.

Get a free automation audit

Related guides