What AI Can And
Cannot Do In Accounting
Samantha Singer, who completed a specialization in AI for finance, on where AI already does real work in accounting, and the judgment it still cannot replace.
- Samantha Singer already uses AI for bank reconciliation, accounts payable, and building complex Excel models and macros she could not otherwise build without a coding background.
- She expects AP work to become largely obsolete, but not AR, since collecting on unpaid invoices still needs a person making calls.
- Her hard rule: if you cannot explain what a number is made of beyond "AI gave it to me," the answer is not finished.
- Removing entry-level accounting roles risks removing the path people use to rise into senior finance roles in the first place.
Samantha Singer was already using AI to build financial models before she completed a formal specialization in AI for finance. What the specialization changed, she said, was her sense of scale: feed it twenty financial statements from different companies, and it can generate benchmarks, flag what competitors are doing better, and read through annual reports in a fraction of the time a person would need.
Her view on where that leaves accounting as a profession is specific, not sweeping.
Watch the full conversation with Samantha Singer, or keep reading below.
Where Singer already uses AI day to day
Bank reconciliation, which she called the harder use case, and accounts payable are both places she uses AI heavily today. The bigger shift for her personally has been in building financial models: interactive spreadsheets, macros, and charts she could not have built without a coding background. "It's given me an ability to really elevate my output in what I provide to clients, in terms of how data looks, how interactive your spreadsheets can be. These are things that usually take years of practice."
Why she expects AP to disappear before AR does
Asked which accounting tasks AI will fully take over, Singer's answer was specific: basic journal entries and AP work, she believes, will largely become obsolete. AR is different, because someone still has to make the calls that get an overdue invoice paid, a task she does not see AI handling well without a substantially more advanced system than what exists today.
Her rule: if you cannot explain the number, redo it
Singer's biggest concern is not that AI produces wrong answers. It is that people stop being able to explain the right ones. "You still have to understand what's behind the number, and that's something that's very important. My concern with AI is that people in finance are going to forget that element. AI can spit out answers for you and reports for you, but if somebody comes to you and asks what's behind how you got to this number, you can't just say AI gave it to me."
In her words
"If you're going to give me AI and you're going to tell me an answer, and you can't back up what that number is comprised of, then I'm going to tell you to leave my office. Go back and redo it."
Prompting is a skill, and skipping judgment costs time
Singer connected output quality directly to how well someone prompts. "It's all about how you feed it. If you know how to prompt AI properly, it can help you a lot. If you're prompting it with not the greatest questions or tasks, it's like anything else, garbage in, garbage out." She also warned against handing a problem straight to AI without applying any judgment first, describing what happens when the first answer does not work: a person prompts again, and again, chasing a fix instead of stepping back to do the research that would have gotten there faster.
The real risk is the entry-level pipeline, not senior jobs
Singer does not think accounting knowledge stops mattering. Senior roles, controller, director of finance, treasury, remain necessary because someone has to understand the mechanics of the business and the numbers behind it. Her concern sits one level down: eliminating clerk and data-entry roles removes the path people have historically used to rise into those senior roles in the first place.
"Sometimes an AP clerk will start as an AP clerk and have a great eye for things, and have that drive to want to improve and get better." Remove that entry point, and the harder question becomes structural: how does someone reach the middle of a finance career if they never got to start at the bottom.
I don't think AI is going to replace the human component to finance. I think it's going to replace a lot of the lower-level roles. There'll always have to be high-level finance people who really understand the workings of the business.
Samantha Singer, Agent CFO PodcastThe rule Singer describes is the rule ChatFin is built around.
Singer's standard, that a person has to be able to stand behind what a number is made of, is the same principle ChatFin runs on. Every agent action carries the data it read and the reasoning behind its draft, so the exception a person reviews comes with its explanation attached, not just an answer to accept or reject.
See how ChatFin shows its workAI that shows its work, not just an answer.
ChatFin runs the repetitive accounting workflows, reconciliation, AP matching, variance review, on your existing systems, and attaches the data and reasoning behind every draft so your team reviews the logic, not a black box.