Before a finance team hands a step of the month-end close to AI, two people on that team should be able to do that step the same way. A lot of mid-market finance teams can't, and it has little to do with how good the people are. The process lives in a few heads and a set of spreadsheets that each person keeps a little differently. Put AI on top of that and you get the same inconsistency, delivered faster.
This is worth your attention now because the sales pitch has moved on. The question used to be whether the tools could do the work at all. BCG's report with SAP on the CFO's AI agenda says the limit is no longer what the technology can do but whether finance can build the data, processes and skills to get a return from it. I agree, and for the close I would add one rule: don't automate a step until two people do it the same way.
Why does AI make a messy finance process worse?
BCG says it in one line: "automating a fragmented process scales fragmentation rather than eliminating it." Picture a close where three people code the same kind of expense three different ways and one person knows how to reconcile the difference. A tool trained on that history learns all three habits. The one person who could untangle them now has more volume to untangle.
My rule for any system is that you can scale a system that works, but you cannot scale a system you don't know works. If two people on your team would close the same month and land on different numbers, you don't yet know your close works. AI will not tell you which of them is right. It will pick a pattern and repeat it at speed.
Most of the return sits outside the model
BCG's research, cited in the same report, traces only about a tenth of AI success to the models and another fifth to the technology platform underneath them; the remaining 70% rests on the organization, its people and their skills. That is why a team with average tools and a clean process will usually get more from AI than a team with the best tools and a close nobody can explain. A tool aimed at a process nobody wrote down only gets you to the wrong number faster.
Three fixes to make before you automate a step of the close
BCG's starting list begins with getting the data in order and agreeing on one financial language, and it says to clean up processes alongside the first AI work rather than before it. Done one step at a time, that comes down to three fixes for a mid-market finance team this quarter:
- Write the close down as it runs today, step by step, with one person's name on each step. Where two people do the same step differently, pick one way and retire the other.
- Agree what your main numbers mean. One definition of revenue, gross margin and an active customer, owned by finance and used by sales and operations too.
- Run each written step by hand for two closes and check that it gives the same answer both times. Automate a step as soon as it holds, without waiting for the rest of the close.
None of that needs software, and all of it makes the software you buy later worth more. Trust but verify applies to your own process before it applies to any tool. Test the brick before you build on it.
When a step is already clean, start there
You do not have to wait until the whole function is tidy. BCG makes the point that momentum matters more than perfect data, and it notes that specialized vendors already report touchless rates above 90% for cash application, the work of matching customer payments to invoices. If a step already runs the same way every time, it is a fair place to put AI to work while you fix the rest. What you should not do is hand AI the steps that only one person understands and hope it sorts them out.
Ask two people the same question this week
Pick one step in your close and ask two people on the finance team, separately, to walk you through how they do it. If their answers match, that step is a candidate for AI. If they don't, you have found your first fix, and it cost you an hour.
Once AI is in, someone still has to check what it produces, which is why I wrote about finance teams spending more time checking AI than using it. And when you decide what the tool is for, score it on growth, not only on hours saved. Getting a finance team's processes ready for AI is part of how I work with mid-market companies. If your close depends on one or two people who know where everything is, I am glad to talk it through.
John Wyatt · John Wyatt's page · All posts · RSS