Finance
Budget reports written for the people who did the spending
Plan against actuals is easy to produce and hard to read. A worker writes the variance up in sentences a department head will actually act on.
What the worker does
An AI worker in Polaris produces budget reporting by reading the actuals export in Google Drive alongside Stripe revenue, comparing them to the plan, and writing up every variance above your threshold in plain sentences. It delivers a PDF as a comment on the task with the working shown line by line. Finance reviews the numbers and decides what the variances mean before anything circulates.
- Connections
- Google Drive · Stripe
- Delivered as
- PDF attached to a delivery comment
- Cadence
- Monthly, or on demand
The report that gets ignored is the one that only has numbers in it
A department head opens a variance table, sees their line is red by eleven thousand, and has no idea whether that is the contractor they hired in week two or an invoice that landed a month late. So they ask finance. Finance answers the same four questions every month.
The useful version of that report has a paragraph per material variance: what moved, the most likely reason based on what is actually in the transaction data, and what would need to be true for it to keep moving.
Writing four paragraphs is a twenty-minute job that finance does after the numbers are already right, at the end of a long week, which is exactly why it does not get done.
What the worker writes for each variance
One block per line that breaches your materiality threshold. Lines inside the threshold get a single summary sentence, not a paragraph each.
| Element | Source | Example of what it looks like |
|---|---|---|
| The number | Actuals export, plan file | Marketing is $11.4k over plan for the month |
| The composition | Transaction detail in the export | $9.2k of that is one agency invoice covering two months |
| The check against revenue | Stripe | Revenue for the same period came in 4% under plan |
| The open question | Written for the reader | Is the second month of that agency invoice already in next month's plan? |
| What it will not say | Deliberately absent | No recommendation to cut a budget. That is not the worker's call |
Making the report trustworthy
Show the working
The delivery includes the arithmetic behind each figure, so a reviewer can check a number in seconds rather than rebuilding it.
Cite the source file and its version
Docs in Polaris are versioned. A report that names the actuals export it read is a report you can reproduce next month.
Set materiality explicitly
Give a currency amount and a percentage in the acceptance criteria. Without one, the worker will either write about everything or about nothing.
Keep opinions out of the draft
Brief the worker to describe and to ask, not to recommend. Recommendations coming from a machine get either over-trusted or dismissed, and neither is useful.
Who does what
The worker prepares
- Reading the plan and the actuals
- Computing variance by line
- Pulling the transaction detail behind material lines
- Writing the narrative and generating the PDF
- Flagging any line where the source data looked incomplete
Finance decides
- Whether the actuals are final
- What a variance actually means
- What gets said to the board or the bank
- Whether a budget changes
- When the report is fit to circulate
Questions people ask
+Where does the plan come from?
Wherever you keep it. In practice that is a file in Google Drive or a doc in Polaris. The worker reads what you point it at and names the file and version in the delivery so the report can be reproduced.
+Can it forecast the rest of the year?
It can extend a trend and say so plainly, but a forecast is a set of assumptions somebody owns. Ask for the arithmetic and the stated assumptions in the delivery, then have a person decide whether the assumptions hold.
+Does the PDF come out of Polaris or do we make it?
The worker produces it. Deliveries arrive as comments on the task and can include generated files, so a finished PDF is attached to the thread where the review happens.
+How is this different from a BI dashboard?
A dashboard shows the variance. This writes the explanation next to it, in the same place the follow-up task will live. If you already have a dashboard that people read and act on, keep it, and use the worker for the narrative that nobody has time to write.
Related
Finance work with an AI worker on the roster
The chasing, sorting and assembling that fills a finance week, prepared by a worker you brief once and review every time.
A monthly close that stops drifting into week two
The close is a chase, not a calculation. A worker runs the chase: who owes what item, who has gone quiet, and what is blocking the two things that always block.
Board packs where the assembly is not your weekend
A board pack is 80% data you already have and 20% the story only you can tell. The 80% is what eats the week.
The Monday metrics email that somebody writes on Sunday night
The numbers pulled and the paragraph written, delivered as a file rather than a link.
Hire an AI financial analyst
You supply the figures; it writes the explanation of what moved, why it matters and which assumption the whole thing rests on.
Connect Google Drive to Polaris
Files a worker produces already arrive as attachments on the task. Files your team already keeps in Drive are the part still waiting.
When the deliverable is a file
A chat reply is not a deliverable if the thing you needed was a document somebody can open.
Acceptance criteria
Written before the work, checkable after it, and binary either way.