Accounting teams are under pressure to close faster and with fewer errors, and much of the time goes to repetitive work: moving data between systems, reconciling it and checking it again.
AI can now take on that execution, which gives accountants more time for the analysis and judgment only they can provide.
Accountants will always remain accountable for the results of their analysis. It's their work week that is now shifting, from mechanical maintenance to the stuff that matters: analysis, exceptions and insight.
I'm the CEO of Latch, which builds software that captures how accounting and finance teams actually work, and we used it on our own month-end first. Here is what that process showed us about getting AI to handle day-to-day operations reliably, and four steps you can use to do the same with your own team.
Capability is no longer what holds back back-office automation. AI models can already do much of the repetitive work. What they lack is context: how the work actually gets done at your company, including the exceptions and edge cases.
Most AI automations fail on exactly that. Teams never capture the institutional knowledge that is so second nature to employees that nobody thinks to write it down. Those undocumented edge cases make an automation fail again and again, until the team loses trust in it and in the next initiative.
Back-office functions are the best place to start. Their tasks repeat on a schedule and follow set rules, so the same edge cases come up every time and can be documented once. People in these roles often spend more time updating and tracking data across disconnected systems than making decisions. With the right oversight, software can own that maintenance work.
We pointed Latch at our accounting team's month-end close, which is made up of many processes. In one of them, our accountant moved data out of Xero and into our financial model in Google Sheets. It took more than 200 clicks, it was manual and repetitive, and it happened on a regular basis.
Latch recorded our accountant working through it and asked clarifying questions as he went, which brought out the edge cases that documentation usually misses. Latch then turned that detailed record into a form a reliable, deterministic automation can be built from. The process went from more than 200 clicks to five.
Every step a person used to do by hand is a chance for a mistake we’ve eliminated. Capturing a process like this used to take hours of billable time with an analyst or consultant, which is why smaller workloads and edge cases never got automated. Now it takes one session.
An efficient back office is a skilled accounting team supported by a software-driven operating layer. Accountants spend the week on variance analysis, judgment calls, exceptions and insights, because software handles the mechanical maintenance.
Our close still runs under an accountant, who reviews what the automation produced, chases the variance that looks wrong, decides where the odd invoice belongs and flags the trends that need attention. Our accounting team gets those hours back for work that needs a person.
Your monthly or quarterly close is a good first process. If you want to try this in your own company, these four steps decide whether it works.
The advantage here is in achieving efficiency at a scale that really was only previously accessible to the largest firms. With the right set-up at start, a team can close faster, make fewer errors and most importantly, redirect those hours into what’s best for the business. Any company can begin with a single recurring process, and the people who get that headstart today are the ones who benefit most.