Agentic Billing for Finance Teams

A due date in Excel told our finance team when to bill an account. It couldn’t tell them whether the customer had already paid. To issue an invoice, someone had to open the spreadsheet, check the date, then work through emails or look up the account in a legacy system. Only after piecing together what the customer still owed could they make out the bill.

At Sulta Tech, we measured the elapsed time from the start of that process to the point we issued the bill. After we moved custom billing off Excel and automated the workflow, that interval fell by more than 80%.

What had to change before a bill could go out

Building the invoice was the last step. Even a perfect invoice template would have left someone in finance with the same questions. Who is due to be charged? What have they paid already? How much should appear on this bill?

We designed the new workflow around those questions. The billing system works from the account’s billing and payment data, determines what is due and issues routine bills automatically. We retired Excel from the process entirely. A routine charge no longer waits for someone to open a workbook and reconstruct the balance.

Where finance still exercises judgement

Each account controls its own agent. Our internal agents interact with the billing system through a tool interface modelled on the Model Context Protocol (MCP). That gives them a way to carry out billing work and to ask for human guidance when a case needs judgement.

Finance can see exceptions and weigh in when an assistant asks for guidance. The account owner can ask a billing question, review the resulting invoice in the same conversation and check the record afterwards. Every billing action is tied to a registered assistant and an accountable owner. The record shows who acted, when they acted, what changed and whether the action succeeded.

How much billing work can we complete in the same time?

Our clock started when billing work began and stopped when the bill was issued. That span includes the time that used to pass between checking the schedule, finding the payment information and preparing the invoice. Measuring invoice generation alone would have missed much of the work we changed.

We also ran capacity tests over a month to see how much billing work the redesigned workflow could handle. In those tests, it completed 15 times as much billing work in the same amount of time as the previous process. That is what we mean by 15× throughput.

The distinction matters when several accounts come due. Under the old process, each routine bill added another set of checks to someone’s queue. Now routine bills can go straight through to issue, while finance gives its attention to the cases that need a decision. The 80% figure tells us how much sooner a bill went out. The 15× figure tells us how much work the system could handle under test.

For a team redesigning its own billing process, a good place to start is the last invoice it issued. Trace it back to the first check someone had to make. That is where our billing run really began, and it is where we started measuring.