Ask in one sentence. Get the finished work back
Your own Claude, ChatGPT or Codex reads the consolidated group in Control and returns a checked list, a drafted report or a budget with a reason on every value. Every figure opens to its transactions.
- CFOWhat moved September against budget, and why?
- ClaudeThe two largest variances, and what is behind each.
- Operating costs. €14k worse than budget in Finland Oy. Marketing services: two of September’s 31 entries were not in the budget, a trade fair stand and an autumn campaign.
- Revenue. €10k better than budget in Finland Oy. Two implementation fees were invoiced at go-live in September. The budget had them in October.
It reads and proposes, and your AI client asks before it writes. A read-only connection can’t write at all, and a locked version refuses every write. What a client can see, and where the data goes, is on Control over MCP.
Twelve more jobs for AI agents in finance, sorted by the evidence
The first list we have run in customer demos, on a sample group. The second is written down step by step in the product docs, with what to check before you rely on the result.
Demoed to customers
A variance explained down to the vendor
Your AI client follows a group variance to the entity, the vendor and the transactions behind it, and leaves a note you keep.
A budget from what you say
You say the assumptions in plain words, and the client writes the budget values with the reason for each one.
This month’s board pack in last month’s format
Give Claude last month’s pack, and it fills the same pages with this month’s numbers from Control.
A dashboard described in plain words
Describe the dashboard you want, and Claude drafts it in Control for you to apply or discard.
The P&L layout from a screenshot
Hand over a screenshot of the old report, and Claude proposes the same structure as a layout for your team to review.
Stripe against the books
The client traces each Stripe payment, fee and refund to the journal entry that booked it.
In the product docs
Monthly close review
Before you close, the client lists unusual bookings, missing accruals and prepaid expenses for you to check.
Update a forecast
The client revises the coming months of a forecast from actuals, hiring plans and sales forecasts.
Import a budget
The client brings monthly values in from a workbook once you have reviewed the mappings and the amounts.
Quarterly budget against actual
The client compares the quarter with the budget and looks into the largest differences.
DATEV ledger upload
The client prepares a DATEV account-ledger export for import and checks the mappings first.
Duplicate supplier spending
The client finds overlapping services and recurring supplier charges across the group.
What finance teams ask before they hand work to an AI client
Ask us about your first questionWhich AI clients work with Control?
Claude, ChatGPT and Codex connect over MCP, and other MCP clients can connect the same way. Inside Control, the in-app chat works on the same numbers.
Can the AI change our books?
No. Control reads each ledger through its official API, read-only by default; write-back, if you turn it on, posts only drafts an Owner or Admin approved.
Where is the data processed?
Control’s own AI features process your data in the EU. When you connect your own AI client over MCP, that client’s provider processes what it reads, under your agreement with that provider.
Is an AI answer proof?
No. Treat it as a proposal until you have checked it. Every figure the client uses opens in Control to the accounts and transactions behind it, so the check is a click away.
Bring one question about your group
In 30 minutes we ask it live on a sample group and show you the finished answer.
The app works best on a desktop browser.
MCP access is included in every plan, from €249 a month. See pricing.