Why Your Firm’s AI Agent Should Work in Finance

CEO Bill Bice sat down with Cheryl Wilson Griffin of Legal Technology Hub for The Back Office After Dark. Their shared view: if you want AI to show results this year, start with billing and collections, not the practice side. Most of the legal AI conversation is about drafting and research. Bill and Cheryl made the case that finance is where you can prove value fastest, because the results show up as cash.

1. Finance already has the data AI needs

Cheryl put it simply: “The back office represents some really interesting opportunities for firms of all sizes.”

The reason is structure. Your billing system already holds years of invoices, payment dates and collector notes. That’s exactly the kind of data an agent can work through in seconds, and that would take a person an afternoon.

You don’t need perfect data to start. Pick one problem, use the data you have, and measure what changes.

2. From a to-do list to an approve list

Bill demoed our A/R Agent on a scenario every collector knows. A client always pays around day 33. It’s now day 40 and nothing has come in. A standard aging schedule won’t send a reminder until day 60 or 90, and by then the conversation is harder.

The A/R Agent notices the payment is late against that client’s own pattern. It also catches stalled partial payments and skipped invoices, pulls in the relevant collector notes, and drafts a follow-up written for that client. The collector reads it, adjusts if needed, and approves.

In Bill’s words: “It turns this from a to-do list into an approve list.”

Collectors keep the relationships and the judgment calls. The agent does the digging.

3. Guardrails matter more than prompts

An agent that can see client data and act on it needs limits, and in a law firm the stakes are higher than almost anywhere. Bill and Cheryl’s checklist:

• Limit the data. Give the agent only what it needs for the task.

• Define the decisions. Be clear about what it can and can’t do.

• Check the numbers with code. Financial figures should be verified, not generated.

• Keep a human in the loop. Anything that reaches a client or involves a sensitive call needs approval.

• Log everything. An audit trail lets you review and improve what the agent did.

Here’s how that works in practice. When the A/R Agent writes a collections email, the AI writes the words, but the balance comes straight from your billing data through code. The model never does the math. One wrong number in a client email can sink trust in the whole project.

Reliable AI comes from the workflow you build around the model, not a cleverer prompt.

4. You don’t need another chatbot

If every vendor bolts a chatbot onto its product, you end up with five assistants giving five slightly different answers.

Bill argued for the opposite: make your firm’s data available to the AI tools you’ve already chosen. That’s what Model Context Protocol (MCP) does. It’s a standard way for AI tools to connect to other systems and data, so someone in finance can ask the firm’s AI assistant about A/R without opening yet another app. Cheryl’s caution: whatever you connect, keep one source of truth.

So the question for vendors is shifting. It’s less “Does it have AI?” and more “How does it plug into the AI we already use?”

5. Measure it in dollars, not agents

The number of agents you’ve deployed isn’t a success metric. Bill showed how a firm can look at AI usage side by side with financial results like collected realization. The questions worth asking:

• Are we collecting cash sooner?

• Is the team spending less time on repetitive work?

• Is collected realization going up?

• Is the value worth what we’re paying?

Even a small gain in realization adds up quickly for a law firm.

Don’t let your best automation live on one person’s laptop

Near the end, Cheryl called agents “knowledge assets.” People at firms are already building their own agents and automations, which is great until the person who built one leaves and nobody knows how it works. Treat agents like any other firm asset: document them, give them an owner, and manage them.

Their advice for getting started: pick a defined problem, test it somewhere safe, measure the results, then build the structure to scale.

Want the whole conversation? Watch The Back Office After Dark recording on-demand.