Every month, the same few days disappeared. Tally the hours. Pay our people. Invoice our clients. None of it is hard, and all of it has to be right, which is exactly why back office automation is worth more to a small professional services firm than it looks from the outside.

We are a diverse professional services firm of around a dozen people. Everyone has a different role, a different skill set, and a different time allocation each month. Checking that, then checking it again for accuracy, is not a five minute job. It is a careful one, and careful takes time.

The month that repeats

The cycle repeats every month, which is the first clue that it was a back office automation problem and not a staffing one.

Hours came in. We validated them against each person’s allocation. Then expenses, because we reimburse our people and we pay for specialized software. One month that meant five hundred dollars of Google Gemini credits for a video project, which was a good use of five hundred dollars and also one more line that had to find its way onto the right invoice.

The record keeping lived in email. People sent things in, someone collected them, someone added them up. Then it all had to be set up and itemized in QuickBooks before anyone could actually get paid.

Then the other half. Invoicing clients, where every client has their own format, and a travel invoice looks nothing like a normal one. How many days were traveled. Who pays for what. Those details are not administrative trivia to a client, they are the difference between an invoice that builds trust and one that starts a conversation nobody wanted to have.

We had help from a bookkeeper. It still took a few days out of every month, and those were days not spent on something new for a client.

So what did we build

We are builders, so we built our own.

Not all at once, and not from a requirements document. Each month we took one more piece of the cycle and automated it, in the order the pain showed up:

Hours entry and approvals, because everything else depends on the hours being right
Expenses, with receipts dragged straight in instead of forwarded through email
Client travel invoices, with their own format and their own rules
Contract change orders
Contract renewals
Cash flow projections

We call it Billie. It came out of our real workflows, the exact ones, not a generalized version of them. Two days of work a month is now about an hour.

That order matters more than the list does. Nobody sat down and specified Billie. We built the piece that hurt most that month, used it, and let the next month tell us what to build next.

How does a firm of a dozen people build specialized workflow apps like these on the side? We built the harness first. Billie runs on our own internal platform of Cocoon AI apps and agents that share authentication, deployment, and the guardrails that make them safe to point at real company data. That is what turns a month of chipping away into something that compounds. We grow our own organic software here, which is also what we do for clients.

Why does software built for one firm beat software built for everyone?

Because back office automation built for everyone is usually very hard to configure for just how you work.

A professional services automation platform has to serve a firm that bills hourly and a firm that bills on milestones, one that travels constantly and one that never leaves the office, one with two clients and one with two hundred. Every one of those choices becomes a setting, and a way to map your work to the software. The software should map to you. That is the part that changed.

Our travel invoices have a shape because our clients asked for that shape. Our approval step exists because of who approves what here. Building around those specifics is not a luxury any more. This is the same economics we wrote about in custom software in weeks, applied to our own back office rather than a client’s product.

What should stay manual?

The judgment. All of it.

Billie is not fully automated, and that was a choice rather than a limitation. We like being in the reviewer and approver chair. When we run projections, we are looking for things a machine has no business deciding: where we are charging a client less than the work deserves or whether we can give a raise for work well done.

Those are the moments the whole month is actually for. Automating the arithmetic is what makes room for them. Automating the decisions would defeat the point.

This is the part most back office automation advice gets backwards. The goal is not to remove people from the process. It is to remove the two days of collating that stand between people and the decisions only they can make.

Where to start if you are looking at your own month

Start with the piece you dread, not the piece that takes longest.

Dread is the better signal, because it usually marks the step where a mistake is expensive and the checking is manual.

Then build one piece of back office automation and use it for a month before building the next. We did not know cash flow projections were worth building until hours and expenses were already clean, and we would have specified the wrong thing if we had tried to design it all up front.

If you want a sense of how this looks in businesses very different from ours, the a16z episode on AI for America’s small businesses has good, plain examples, including a dentist’s office with the same problem we had.

Frequently Asked Questions

The four questions we get asked most when we tell this story to another firm.

Should we build or buy back office automation?

Buy first if a tool genuinely fits how you already work.

Build when the fit is the problem, when you keep finding yourself working around the software instead of with it. The signal is not cost. It is how many of your real rules the tool makes you abandon.

How long does something like Billie take to build?

Ours was never a project with a start and an end.

It was one piece a month, each small enough to finish and use before the next one started. That is a slower story than a six week build, and it is also why the thing that came out fits.

Does this replace a bookkeeper?

No, it changed what we hand over to ours.

The collating and itemizing that used to go out is done before anyone looks at it, so the help we get is spent on the parts that need a professional eye.

What is the first piece worth automating?

Whatever everything else depends on being correct. For us that was hours entry and approvals. Automating invoices on top of hours nobody trusts just produces wrong invoices faster.

The point of the power tools

Billie is a power tool for our business.

It is one of several we have built for ourselves, and building them is how we learned most of what we bring to clients.

The interesting part is not that we got back office automation working for our billing. It is that a firm of a dozen people can now afford software shaped exactly like the way it works. That was not true five years ago.

If your month has a shape like ours, we should talk. And if you want to see what we do with this for clients rather than for ourselves, that is our services.

We Built Our Own Back Office Automation