AI can take the chores. Here’s how finance leaders decide where human judgment matters most.
Tesorio CEO Carlos Vega recently shared something a former Tesorio data engineer told him:
“AI is doing my art and I’m doing the chores.”
The comment stuck with him. As AI becomes capable of doing more of the work, companies have a new decision to make: what work should they give it?
For finance teams, two questions can help draw that line: Does a person need to do this? And where does having a person involved make the outcome better?
Collections gives us a concrete example.
What fills the AR inbox
We looked at the emails AR teams receive in Tesorio and sorted them by how overdue the invoice was. We found that 73% were about invoices that were current or under 30 days past due.
As Carlos wrote when he first shared the finding:
“That is most of the inbound a collector wakes up to, and most of it needs no judgment. Send the invoice copy, confirm the remittance, answer the portal question, log the promise to pay.”
Each one is quick. But the volume of quick tasks can add up to a full day. The answer usually already exists in the customer record, the ERP, or an earlier thread. The work is finding it, checking it, replying, and logging what happened.
AI agents can now take on more of that routine inbound without requiring a person on every email. Carlos described the potential this way:
“If AI agents take a good chunk of the 73% and work 24/7 with no rest, the same team has roughly 3x the bandwidth for customers that need the most attention in order to get paid.”
The 73% isn't a line between work for AI and work for people, though.
Aging is a clue, not the rule
Not every email about a current invoice should be automated, and not every overdue account needs a person.
Plenty of 31-to-60-day work is still routine. A buyer may be stretching terms on purpose, or an approval cycle may run long. But the accounts that stay outstanding, especially the large ones, can be telling you something else: the product didn't land the way it was sold, a dispute wasn't escalated, the PO number was wrong from the start, or the buyer is managing their own cash.
Those situations require getting on the phone, understanding the relationship, pulling in Sales or Customer Success, and deciding what to do.
AI can surface the context. Deciding what to do with it is the human part.
Knowledge and judgment are different jobs
Carlos has written about the distinction between knowledge and judgment as AI takes on more work:
“Knowledge is easily confused with power, but AI knows more than any of us do. Can AI appreciate? Can it relate? Can it read an emotion in a room and decide anyway?”
That distinction shows up in practical ways across order-to-cash. In cash application, matching the obvious payment is one job; investigating the exception is another. In reporting, retrieving an answer is one job; deciding what to do with it is another.
Human attention is finite. Every hour spent finding an invoice copy or working an obvious match is an hour not spent on the dispute holding up an important account.
Redesign, not reduce
When Carlos recently shared the collections finding with a finance leader, the leader started thinking out loud about how he would redraw his team. As Carlos described it, the conversation wasn't about reducing the team. It was about changing its KPIs, SLAs, and structure.
That suggests a few places finance leaders can start as agents take on more routine work.
Measure outcomes on the hard accounts, not activity. Emails sent and calls made sense when people handled all the volume. As agents take on routine inbound, teams can look instead at disputes resolved, how quickly important overdue accounts get meaningful attention, and whether those accounts are moving toward resolution.
Rethink service levels for routine requests. An invoice copy or remittance confirmation that used to wait a day can be answered right away. That's a better experience for the customer and one less reason for an easy invoice to slide past due.
Decide in advance what earns a person. A large balance past 30 days, a disputed line item, or a broken promise to pay can signal that it's time for someone to step in. Give those signals a named owner and the time to work them with Sales or Customer Success.
Redraw roles around judgment. As agents take on more routine work, collector roles can shift toward exceptions, strategic accounts, and the customer situations where their judgment changes the outcome.
For years, finance automation focused on making each step faster. AI gives finance teams a reason to ask a different question: which steps need a person at all?
Carlos put the challenge this way:
“The job I've given myself is to make sure the art stays with the person and the chores go to the machine, and to notice when it seems to flip. I don't have that figured out.”
The line will keep moving as AI improves. But some work still depends on things a system cannot reduce to a rule: understanding relationships, weighing competing priorities, and judging the broader business impact of a decision.
That is where people remain essential.
The future of finance may involve far less manual work. It should not involve less human judgment.
