@dominicrume Right question. For finance the proof is boring and specific: what data the agent read, what it changed, who approved it, and whether someone can rerun it next quarter and get the same number. If the agent can't answer that, the auditor will ask you.
@BowTiedMarmott Same pattern here. Four or five people now carry what ten used to, and the close is where it shows: recs and accruals slide to day 8. If the close is what's stretching a small team, happy to compare notes on Assisted Close: https://t.co/dMQ5G1cGp3
@mardehaym The budget-per-team point is underrated. A surprise AI bill is often the first thing that costs a CFO's trust. In finance we'd add one early step: pick one recurring task, like a rec, and prove the output can be rerun and checked before adding more seats.
@DinoJVidiliCons Day 4 deadlines are where AI can help: first-pass flux from the GL detail. The analyst still decides which story is true, but starts from evidence. We have a free Claude skill that does this and writes the SQL behind every figure: https://t.co/OO3gvX73FQ
@ericbsaunders Daily close is the right direction. The hard part is usually the exceptions, not the entries: the unmatched deposit, the bill nobody sent. If staffing is what lets those pile up, happy to compare notes on Assisted Close: https://t.co/dMQ5G1cGp3
@gaze_observer 40% of the work is a big claim. For finance the question is which 40%. Preparation and matching, sure. Judgment calls and sign-offs still need someone accountable, so the mix of roles shifts more than headcount disappears.
@samwoods Classic bottleneck shift. Automating the top of the funnel just moved the constraint to partner time. Same thing happens in the close: speed up matching and the review queue becomes the bottleneck.
@sonuv_dev Smart use: AI as the tough reviewer, not the author. The test is whether it pushes back on the numbers themselves, not only the framing. We published free Claude skills that write the SQL behind every figure so that review can be checked: https://t.co/VmN6luV13f
@Cassxbt@airwallex This is the gap in most payment runs. Approval is tied to the invoice, not to the bank account. Any change to vendor bank details should reset the approval. If you want to compare notes on locking down that one step, DM us.
@newonpolsia Multi-entity finance is where a lot of close calendars slip, especially across currencies. Continuous recs help only if each tie-out can be rerun later. If staffing that close is the constraint, happy to compare notes on Assisted Close: https://t.co/dMQ5G1cGp3
@BMNRBullz Both can be true. Agents will get access to money, but only in scoped, logged, reversible steps. The banks that define those permissions first will set the standard everyone else follows.
@AccountingToday The lesson for firms is the same even when the story is an exam: if AI produced the work, you still have to show what it saw and who checked it. A score without that trail isn't evidence you can stand behind.
@peni963@meetmotes Agree. In finance, a useful agent is one whose limits you can explain to an auditor: which accounts it can touch, what thresholds trigger a human, and a log that shows it stayed inside them.
@Datweb3guy@openservai Right problem to start with. Before an agent pays, it should be able to show why: matched PO, receipt, unchanged vendor details, under threshold. If you want to compare notes on putting those checks around a payment run, DM us.
@JohnReedStark Footnotes are where the real story usually sits. Curious which disclosure changed most across the four years: segment and geography reporting, related parties, or revenue timing?
@donaldgorbachev Revenue recognition is where most of these stories end up. Was there a contract, did control transfer, and is the evidence the same thing the auditor tested? Hard to judge from the outside until those three line up.
@audracarpenter Going through 12 months of charges for unexpected fees and subscriptions is classic close work. The useful output is a list you can tie back to the bank or card file, not just a summary. That's the part we'd want any agent to leave behind.
@TanGravam A live pilot on month-end is a real risk. Replaying last quarter first, and scoring agreed, wrong, and unsure, is how we'd want any close agent tested. If you want to compare notes on testing one close step that way, DM us.
@DAIEvolutionHub In finance the junior work is also the training. Reconciling a balance sheet is how you learn to spot a bad number. If AI does that work, someone has to design how juniors still build judgment, because reviewing AI output is a senior skill.
@matt_haf Experimenting but not confident sounds right for treasury. Cash positioning and forecasting have clear answers to test against, which makes them better first projects than anything open-ended. Which use case came up most?