It is month-end at Keldmoor, a fictional industrial distributor. Finance has a stack of September invoices, a supply contract with an amendment, purchase orders, receiving reports and a morning scan from the mailroom. The questions are the usual ones: did we pay the right prices, did anything arrive twice, and how much is at stake?
Answering them usually means opening PDFs side by side, copying figures into a spreadsheet and hoping nothing slips. Our new video, Ask your invoices, shows the same month-end in DocReveal — from a question typed in plain words to a dashboard the whole team opens. It runs three and a half minutes, you can watch it at the top of this page, and it is made for finance directors, controllers and accounts payable teams rather than for data engineers. Here is what you will see, and why it matters.
The answers are already in the documents
The facts a finance team needs are rarely missing. They sit in contracts, amendments, invoices and scans — spread across files, formats and languages. The hard part is connecting them quickly, and in a way that someone else can verify.
That second part matters as much as speed. A number without a source is an opinion. When an answer ends up in a payment decision or in front of management, everyone should be able to see where it came from.
Ask: plain questions, cited answers
The video opens in RAG Manager. The finance team asks whether the supplier billed the correct prices after the contract amendment. The answer names the one invoice that is off — a lock nut billed at $1.15 instead of the amended $1.09, which makes the invoice $720.00 too high — and confirms that the other September invoices match.
Every figure carries a citation. One click on it shows the source — in this case the page of the amendment with the new price. Nobody has to take the answer on trust; the evidence is one click away.
The second question asks for all September findings in a table, with a chart by supplier. DocReveal draws on the results of the business rules checks, on the extracted invoice data and on the documents themselves, and answers with key figures, a table of five findings and a chart. Together they add up to $15,900.36 that should not be paid. The table can be exported to Excel, Word or Markdown.
The rule results come from the same checks we showed in Checked, not trusted. Instead of guessing whether an invoice is a duplicate, the chat can read the verdict.
Count: a query you can read
The largest finding is a duplicate: one invoice arrived by e-mail and again as a paper copy in the mailroom scan. For an amount that large, a controller wants proof, not prose.
So the video moves to Analytics Manager. The question “Did any invoice arrive more than once?” is translated into SQL, and the query is shown for review before it runs, together with a short explanation in plain words. Anyone who reads SQL can see exactly what will be counted. The result is a single row: the same invoice number, two copies, $27,250.00 in total.
These queries only read. They are limited to your own tenant’s data and to a capped result size, and every run is recorded with its ID, confidence and timing. Asking a question about your data can never change it.
Share: one dashboard, the same numbers for everyone
An answer in a chat helps the person who asked. A month-end close needs the same figures in front of everyone. In Dashboard Manager, a dashboard for the September close is created on the invoice dataset, starts as a draft and is built in the embedded editor by clicking — key figures, amounts invoiced by issue date and a table of invoices. No code and no separate tool.
One click publishes it to everyone in the tenant. On the Dashboards page, the whole team sees the same seven invoices, the same totals in dollars and in euros, and one bar towering over the rest on September 16 — the duplicate, visible at a glance. The data refreshes with every analytics run, so the dashboard keeps up as new documents are processed.
What to take away
- Ask for the source, not just the answer. If an AI answer cannot point to the page it came from, it is not ready for a payment decision.
- Prove the big numbers with a query. When an amount matters, confirm it with a count you can read, repeat and show to an auditor.
- Keep analysis read-only. Questions about your data should never be able to change it.
- Publish once, share one version. A published dashboard ends the debate about whose spreadsheet is right.
All companies and figures in the video are fictional demo data.
Ask in words. Get numbers you can check.