AI Agents in Financial Reporting: Where It Breaks
The report says the same thing every month and takes three days to build.
The loop, as it actually runs
- Period closes
- data pulled from each system
- schedules assembled
- variances explained
- report formatted
- distributed
Where it breaks
Not in the steps. In the joins between them:
- Pulling data is manual and error-prone
- Variance explanation is rewritten each month
- Formatting eats hours
- Distribution is a person emailing a PDF
What that actually costs
Three days to produce a report that says the same thing every month is not primarily a labour cost. It is a latency cost: decisions get made on numbers that are three days plus a close cycle old, which for most operating questions means acting on last month rather than this one. The formatting hours are annoying and the assembly hours are wasteful, but the expensive part is that nobody asks a question mid-month because the answer would take three days to produce, so the questions stop being asked.
What we would build first
Data pulling from each system goes first. It is the least interesting step and the one that determines everything else — variance explanation, formatting and distribution are all blocked on having the numbers in one place. Automating the pull also surfaces the reconciliation problems you have been absorbing by hand, which is uncomfortable and useful. Formatting is tempting to do first because it is visible; resist it, because a beautifully formatted report still takes three days if the data gathering is manual.
It runs inside QuickBooks Online, NetSuite, Excel, Google Sheets, Slack — the tools already in use.
How this one goes wrong
Drafted variance commentary is where this goes wrong, and it goes wrong fluently. An agent finds a plausible cause for a variance within the same period and writes a confident sentence — and the actual cause was a reclassification two months earlier. It reads well, it enters the board pack, and it is hard to catch precisely because the prose is good. The guard is that every commentary sentence names the transactions it draws from, and any variance the agent cannot tie to specific entries is written as an open question rather than an explanation.
The individual jobs inside this loop — what starts each one and where it has to hand back to a person — are covered under finance & bookkeeping.
Where it lands hardest: accounting & CPA firms. That page carries the sector’s own economics — the three roles that absorb the work, and the season that decides when a build can start.
In most companies this work sits with bookkeepers, accountants and management analysts. Each of those pages carries the wage arithmetic for that role, and says plainly where the numbers do not justify the spend.