This article was originally published in Legaltech News, here.

A few weeks ago I wrote a piece here in Legaltech News arguing that law firms don’t have an AI problem, they have a deployment problem. That one was about the unglamorous groundwork: infrastructure, data cleanup, governance. This one is about the opportunity most firms are missing.

It isn’t with the attorneys.

Almost every AI conversation in legal right now is about attorney work. Drafting, review, research, contract analysis. Those tools are real, they’re getting better fast, and firms will need them to stay competitive. But look at what they do: they help attorneys work faster. In a business that sells time, faster doesn’t automatically show up as more revenue. Attorney-facing AI runs headfirst into the billable hour, and its value depends on attorneys changing how they work, which is a hard thing to do inside a law firm.

That isn’t a knock on attorneys. It’s just how the incentives sit. The people whose behavior has to change are the same people who generate the revenue, set the culture, and can opt out at any time without consequence. Unless a firm has a managing partner or a board with unusual appetite for pushing technology, adoption is voluntary. And voluntary adoption is how you end up with a platform the firm paid for two years ago that four people use.

Meanwhile, the part of the firm where process optimization has the biggest impact is sitting right there.

The Back Office Is Where the Firm Actually Runs

Intake. Conflicts. Matter setup. Reimbursements. Vendor payments. The work that has to happen before anyone can bill an hour, and the work that has to happen after.

Two things are true about that work at most firms. The first is that it runs on genuinely old technology. The accounting and practice management platforms in this industry were architected a long time ago, and it shows in the number of clicks it takes to do anything.

The second is more expensive: firms have built entire administrative processes around compensating for those interfaces. There are people whose job is substantially to be the connective tissue between systems that were never designed to talk to each other, and to chase attorneys who don’t want to fill out the form.

Fix that and you don’t save hours in some theoretical way that never shows up in a financial statement. You reduce cost, and it goes straight to the bottom line.

Start with Intake, Because Intake Touches Everything

When a firm asks where to start, we start with intake. Not because intake is the biggest pain, though it usually is, but because of what building it forces you to connect.

To run intake properly you need to reach the document management system, the accounting and billing platform, the conflicts data, and increasingly the CRM. Get intake right and you’ve built the foundation the rest of the back office runs on.

Here’s what it looks like in practice. Today an attorney gets asked to fill out a long form. It may be dynamic, it may be well designed — it doesn’t matter. They don’t want to fill out a form. So instead, they send an email the way they would to a colleague. The system reads the email, derives the context, populates the form, and hands it to the admin team to review before anything moves. The person who is good at this work still puts eyes on it, fills in what’s missing, and kicks off the conflicts check. The human stays in the loop. The friction moves off the person least willing to absorb it.

The pattern is the same for a reimbursement. An email with an attachment and an amount comes in, the documentation gets filed where it belongs, and the entries get made. Nobody learned a new system.

It’s a Layer, Not Another Platform

The instinct in this industry is to go find a platform that does this. For a mid-sized firm, I think that’s the wrong starting point, for a few reasons.

• What you actually need is a routing layer, not another system of record. It moves a request from the person making it to the system that has to act on it. It doesn’t need to be another database, which means it isn’t another store of client data to secure, govern, and audit.
• Traditional platforms come in and tell the firm to work a certain way because that’s how the software works. That assumption is what makes those implementations long and painful. You no longer have to accept it. It should fit the process the firm has today and improve it from there.
• It should run in the firm’s own environment, on the firm’s own data, and it should be able to change when the process changes.

This is roughly what the large consulting firms have been quietly building for their corporate clients for the last two years. There’s no reason a 200-attorney firm can’t have the same thing.

The Part That’s Hard

Two things make back-office AI difficult, and neither of them is the AI. The models are the easy part now and getting easier. What’s hard is everything they have to touch.

The first is integration. The same platforms that make the back office painful to work in are painful to connect to. Some have a real API. Some have an API that covers a third of what you need. Some have nothing worth using, and you need to build a database connection, a nightly file, or a vendor conversation that takes a year to resolve. This is where the budget on these projects actually goes. Not the AI, which is comparatively cheap and getting cheaper. Any firm scoping this work should establish, before anything else, what its core systems will permit, and expect a few engineering challenges. “We have an API” and “you can do what you need through the API” are different sentences, and vendors are not always quick to distinguish them.

The second is governance. You are building something that reads client email, touches conflicts data, and writes into the accounting system. That sits squarely inside the firm’s confidentiality obligations, and it sits on top of outside counsel guidelines that increasingly say specific things about AI. Some clients now require notice. Some require consent. Some prohibit certain uses outright.

This is why we build these systems inside the client’s environment rather than pointing the firm at someone else’s. The data stays in the firm’s tenant, under the firm’s controls, subject to the firm’s retention schedule and its audit logging. No third party is holding a copy of the intake queue. When a client asks, the answer is that nothing left the building. That is a materially stronger position than pointing at a vendor’s certifications. It takes more work to stand up than the alternative, and it is worth it. It’s the difference between a governance conversation that takes a meeting and one that takes a quarter.

The item, less dramatic and in my experience the more common cause of failure: somebody has to own it. This is not a project with an end date. Processes change, systems get upgraded, a vendor renames a field and something quietly stops working. The firm needs a named person responsible for that, internal or fractional, but named. Firms that treat go-live as the finish line end up with a layer that worked well for eight months and nobody who noticed when it stopped.

Why It Works

Go back to the deployment problem. The reason so much legal technology stalls is that it requires behavior change from people who don’t have to change. Back-office AI doesn’t. The attorney sends an email, which is what they were going to do anyway. The administrative team gets a better version of the process they already own. Nothing about the firm’s existing systems has to be ripped out, because the whole point is to sit on top of them and connect them.

None of that changes because the work is hard. It’s a trade, and it’s one I’d make every time. The difficulty in back-office AI is technical, and technical problems can be scoped, priced, and handed to people you hire. The difficulty in attorney-facing AI is behavioral, and there is nobody you can hire to make a partner change how they work.

It also does something less obvious that I think matters more over time. Mapping these processes requires the firm to articulate how the work actually gets done, and a lot of firms don’t know. The work lives in silos, in the heads of people who have been doing it for fifteen years. Getting it written down and running through one place gives firm leadership visibility it has never had.

The AI conversation in legal is going to stay focused on the attorneys, and it should. But if you’re looking for the place where AI can go into a law firm this year and produce a measurable result without a change management fight, it’s the part of the firm nobody is talking about.