Automate your step, then let the bottleneck show

A lot of work for an accounting team revolves around the month-end close. It’s not just a single task, but a chain of tasks across many stakeholders, teams, and tools. Source systems produce data, someone snapshots it, someone runs reports, accounting processes them, and then FP&A and executives review the results.

Over the past six months at an insurance company, my team built an agentic operation for parts of that chain. Software we’ve built in Malleable reads from our tools (Spreadsheets, Google Drive, Ramp), prepares journal entries, waits for a human to approve, and posts directly to NetSuite.

Our accounting workflows now run mostly on their own, and are very fast. But we still have dependencies. Premiums and claims reports still come from a person on another team running them by hand and dropping them in a Google Drive. And, after we’re done, other teams take what we’ve done and process them by hand.

When many people see a process like this, their instinct is to focus on the start. If we can’t automate the first step, why focus on the third step? It will still be manual, slow, and require cross-team coordination.

There is a rule I learned at a previous company: make your own part of a process as efficient as possible, even with dependencies on both ends.

Once you’ve done this, it’s very obvious to the company the potential of streamlining the other parts of the process. When the visibility increases, and the value becomes clear, it actually happens quicker than you’d think!

People say that AI is only as good as the data you give it - “garbage-in, garbage-out.” But I’m not a data engineer and can’t change our systems. What I can do: set the AI up to do what it’s capable of right now, and give it a way to pause and ask a human for help when it hits a dependency. For example, have someone drop a file in a drive and mark the next step as ready. As the rest of the company improves its process to be more agent-ready, our operations can gradually expand what happens autonomously.

Don’t wait for a company-wide plan. Make your step autonomous, see what works, and help your neighboring teams know what to prioritize to make it even better.