Billing for AI-Assisted Work: Hourly vs Flat vs Subscription

By Jude Lee · September 26, 2026 · Workflow

The capability that starts the argument

An AI agent is different from a chatbot: it carries out a multi-step task and takes actions in your systems. In a small firm, that looks like an assistant such as Claude connected to your practice-management and document systems through MCP (the Model Context Protocol, an open standard for giving an AI governed access to specific data and tools). Ask it to prepare a first draft of a discovery response and it can pull the requests, locate prior responses on similar matters, apply your firm’s objection language as a packaged skill, and save the draft back to the matter — with a lawyer reviewing before anything goes out.

Be honest about what that does and doesn’t change. The agent removes retrieval, assembly, and first-draft typing. It does not remove judgment, client strategy, or the read-through that catches an invented citation. Practically, a task that took a couple of hours becomes a shorter drafting-plus-review cycle. The legal work didn’t disappear. The hours did.

The failure modes are specific and worth naming, because they’re what you’re billing review time to catch. An agent pulling “our standard objections” from a prior matter can just as easily pull them from a response that was later amended after a meet-and-confer — the language looks like yours, reads as authoritative, and is the position you already abandoned. A chronology agent working across a document set can silently omit an undated exhibit rather than flagging it, and silent omissions are far harder to spot than wrong entries. Two review steps catch most of this: require the agent to cite the source matter and document ID for every borrowed passage (so the reviewer can see which prior response it copied), and require it to output an explicit exclusions list — every document it couldn’t date, parse, or classify — instead of quietly dropping them.

AI doesn’t reduce what your work is worth. It removes the excuse for pricing it by the hour.

What the ethics guidance actually says about billing AI time

Start with the primary source rather than vendor marketing. ABA Formal Opinion 512, issued in July 2024, addresses generative AI across competence (Rule 1.1), confidentiality (Rule 1.6), client communication (Rule 1.4), supervision (Rules 5.1 and 5.3), and fees (Rule 1.5). On fees, the thrust is straightforward: charges must be reasonable, and a lawyer billing hourly should bill the time actually spent — not the time the task used to take. The opinion also addresses when AI costs can be passed through as expenses and when time spent learning a tool is properly billable.

Rule 1.6 gets the least attention in pricing conversations and deserves more of it, because wiring an agent into matter files means client confidences leave your four walls. Before you connect anything, know which data the agent can read, whether the provider trains on your inputs, and what retention terms apply — zero-retention, private-cloud, and self-hosted arrangements exist at the enterprise tier for several major assistants, but confirm them in the vendor’s own documentation and contract rather than a reseller’s summary. Then decide what your client is told: ABA 512 discusses when disclosure to the client is required, some engagement letters and outside-counsel guidelines demand consent before third-party tools touch matter data, and your state’s opinion may go further than the ABA’s.

State guidance sits on top of all of this. The State Bar of California’s Practical Guidance for the Use of Generative AI in the Practice of Law and several state bar ethics opinions cover the same ground with local variations, and courts have their own standing orders on AI disclosure in filings. Read the opinion your jurisdiction actually adopted, and confirm anything fee- or disclosure-related with bar counsel or a qualified legal professional before you change engagement letters.

Hourly versus flat fee once agents are in the workflow

Hourly, with honest time capture

Simplest to implement and defensible: you bill what you spend. Fits unpredictable litigation, court appearances, and anything where scope genuinely moves. The catch is structural — every efficiency gain lands as a smaller invoice, so the firm pays for its own automation. It only works as a growth strategy if the recovered hours get refilled with other billable work or new matters. If your calendar isn’t full, hourly plus AI is a revenue cut.

Flat fee per matter type

Prices the deliverable, so efficiency accrues to the firm. Fits repeatable matter types — estate plans, uncontested filings, residential closings, immigration petitions, demand packages. The catch is that Rule 1.5 reasonableness still applies, and flat fees punish you brutally on scope creep. You need a written scope, a named trigger for moving off the fee, and enough completed matters to know your true cost per matter before you set a price.

The subscription and standing-retainer option

A monthly subscription or standing retainer for ongoing advisory work fits small-business clients who want access more than deliverables — handbook reviews, vendor contract questions, routine employment issues. It benefits most from agentic support, because an agent that can answer matter questions against your own files turns a high-touch relationship into a manageable one.

It’s also the model most easily abused, so write the guardrails before the first invoice. Cap scope explicitly: which matter types are included, how many hours or requests per month, and what is always out of scope (litigation, transactions above a dollar threshold, anything adverse to another client). Decide in advance what happens when a client over-consumes — overage billed at your hourly rate, a tier upgrade at a review point, or a hard stop until the next cycle — and put that in the agreement rather than negotiating it while annoyed. Define the exit too: monthly term, notice period on both sides, what happens to unused capacity, and how files transfer. Without those three pieces, a subscription becomes unlimited hourly work at a flat discount. Many firms end up with a mix: hourly for litigation, flat for repeatable filings, subscription for a handful of business clients.

Modeling the money without inventing a number

Don’t take anyone’s published savings figure — including anything you read in a vendor deck or a practitioner forum. Model your own:

H − h

Recovered hours per matter: your baseline hours minus AI-assisted hours (measure both on 5 real matters)

(H − h) × N × R × U

Annual upside: recovered hours × matters per year × your rate × the share you actually refill with billable work

Build + license + review time

Cost side — include the lawyer minutes spent checking every AI output, which never go to zero

The variable people skip is U — utilization of the recovered time. Hours you free up and then spend on email are worth nothing. Under a flat fee, the math flips: the gain shows up as margin per matter rather than as extra billable hours, which is why efficiency and hourly billing fight each other. Our law firm automation ROI calculator walks through the same arithmetic in more detail, and automating time tracking and billing covers the capture side that everything here depends on.

Where the extra capacity turns into actual revenue

Revenue is rate × billable hours × realization × collection, and there are only so many hours in a week. AI doesn’t add hours; it changes which of those levers you can move. Three realistic paths for a small firm:

Raise throughput on flat-fee matter types. More matters per lawyer at a stable price, with cost per matter falling.
Convert more of the leads you already get. Intake speed is usually the binding constraint, not demand. Agentic intake — qualify, conflict-check, open the matter — is covered in our intake automation playbook.
Shift work up the value chain. Recovered hours spent on strategy, referral relationships, or higher-rate matters, not on document assembly.

Which slice of the work to automate first

The 80/20 framing lawyers use — a small share of clients and matter types producing most of the revenue — is a useful filter, but run it twice. First on revenue: which two or three matter types actually pay the bills? Second on drag: which recurring tasks eat the most non-billable time? Automate where those two lists overlap. A skill for the deliverable you produce fifty times a year beats a clever agent for the thing you do twice.

What firms are actually running today

Last reviewed: early 2026 — this layer of the market moves fast, so treat product names as examples rather than a current shortlist. Small and mid-size firm stacks tend to combine some of: AI features built into practice management (the current generation includes things like Clio Duo and Smokeball’s assistant); point solutions for a specific job such as contract review, deposition summaries, or citation checking; a general assistant like Claude, ChatGPT, or Copilot for drafting and analysis; and — least common, and the most work to stand up — a custom agent connected to firm systems via MCP. The built-in features are the cheapest place to start and are often enough. Custom builds earn their keep when the workflow is specific to your firm and high-volume; the tradeoffs are laid out in vendor agents vs a custom build.

Pick one repeatable matter type

Highest volume, most predictable scope. Not your hardest litigation.
Instrument the baseline

Track true hours and true cost per matter for a month, including paralegal and admin time.
Add the agent where retrieval and assembly live

Draft generation, document abstraction, chronology building — with provenance citations, an exclusions list, and mandatory attorney review before anything leaves the firm.
Re-measure, then reprice

Set the flat fee against demonstrated cost plus a margin for variance.
Reinvest the hours deliberately

Name what the recovered time is for — intake response, business development, higher-rate work — or it evaporates.

If your matters are genuinely unpredictable and your calendar is already full, stay hourly, capture time accurately, and treat AI as a quality and turnaround improvement rather than a pricing event. That’s a legitimate answer, and for plenty of firms it’s the right one.

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