Finance industry · M&A and transaction services

AI for M&A and due diligence: just ask the question

Full data room, crooked scanned PDF, 300-page annual report, Excel workbooks, the target's ERP export or general ledger file: ATG reads it all.

Your transaction services teams ask the question, and the answer cites the document it comes from.

By Jean-Christophe BudinUpdated on

Your sources

All your sources, even the worst ones

Scanned PDFs

Skewed, stained, badly scanned: OCR turns them back into text, tables included.

Financial reports

Annual reports, prospectuses, consolidated accounts: tables and notes are read along with the text.

Excel

Multi-tab workbooks and financial models, queried directly on their figures.

ERP exports

General ledgers, trial balances, extracts from your ERP, cross-checked with your other sources.

FEC

The French standard audit file of accounting entries, queried entry by entry.

You upload or connect the files. Then you just ask the question.

Proof

Accuracy, measured on a public benchmark

We run ATG on FinanceBench, released by Patronus AI: 150 analyst questions over 368 filings of listed companies, all in one knowledge base.

FinanceBench

ATG

Correct answers

92%

Average time

11 s
Mistral AI

Correct answers

86%

Average time

71 s
See the protocol and the comparison

Use cases

M&A has never been this simple

Financial due diligence

Data room, target accounts, accounting exports: every question gets its answer from the documents, and ATG flags what is missing or contradictory.

Adjustments and normalised EBITDA

Spot non-recurring costs, net debt items or off-balance-sheet commitments in the notes, board minutes and contracts. Each item comes with its source; the analyst decides on the adjustment.

Buyer questions

During the process, find the answer to a buyer's question in the data room, along with the supporting document.

Accounts and ledger review

Query the ledger file and trial balances: unusual entries, balances to justify, reconciliations across files.

Financial analysis

Find a figure, recompute a ratio, compare two fiscal years or two companies, with the source document cited.

Target and comparables

Cross-check the target's accounts with public data: filed accounts, officers, legal notices, share prices of listed peers.

Excel and accounting data

Your Excel files and accounting exports, queried like a database

Trial balances, general ledgers, ledger files, ERP exports, the target's Excel workbooks: ATG loads them into a dedicated database and documents what each column means. Your teams ask their question in plain language, ATG turns it into a SQL query and answers with the figure, a table or a chart.

In due diligence, this is what lets you recompute monthly revenue, seasonality or a net debt item without rebuilding a pivot table.

Exact calculations

Totals, variances, breakdowns by account or by month: the calculation runs on every row, not on an excerpt read by the AI.

Several files, one database

Trial balances for several years, the ledger file and a billing export are loaded together, so a single question can cross-reference them.

Figures and documents in one answer

The figure comes from the database, the explanation from the management report or the contract: ATG combines both and cites its sources.

Read-only

The queries the AI generates only read: the loaded data is never modified.

Industry requirements

What a firm or financial institution checks before deploying AI

Traceability

Every answer cites the documents used: an analyst checks before reusing a figure.

Choice of AI providers

The EU-only providers policy restricts processing to European providers. We publish the effect of that choice on accuracy and latency.

European hosting

The platform and data storage are hosted by European providers, in European data centres.

Published performance

Accuracy and response time measured on a public benchmark, configuration by configuration, compared with results published by others.

Our approach to sovereign AI

Frequently asked questions

Can AI perform a financial due diligence?

It does not sign off the conclusions, but it speeds up all the reading and searching: ATG goes through the data room, finds the figures and clauses, flags what is missing or contradictory, and cites every source. The transaction services team keeps the professional judgement and validates each adjustment.

Which documents can ATG analyse?

PDFs, including scanned ones (OCR), financial reports with their tables, Word, PowerPoint, Excel workbooks, ERP exports and FEC ledger files. You upload the files or connect your sources (SharePoint, Google Drive…), then ask your questions.

Can ATG query our Excel files, trial balances and ledger files?

Yes. Structured files (trial balances, general ledgers, ledger files, ERP exports, Excel workbooks) are loaded into a dedicated database. ATG turns each question into a read-only SQL query, computes on every row and can cross-reference the result with the documents in the data room.

How is this different from a virtual data room (VDR)?

The data room stores and shares the deal documents, with its access rights. ATG does not replace it: it reads the documents it holds, or their export, and answers the team's questions citing the relevant document.

Can ATG's answers be checked?

Yes. Every answer cites the documents used, so an analyst can check the source before reusing a figure.

What accuracy can we expect on financial documents?

On FinanceBench, the public benchmark of analyst questions over filings of listed companies, we publish ATG's accuracy and response times for every configuration. The best test remains your own documents: we can measure it during a demo.

Can we use European AI providers only?

Yes. The EU-only providers policy restricts processing to European providers, and the platform is hosted in Europe. Your institution's own requirements are handled before deployment, with your risk and compliance teams.

Test ATG on your own documents

A demo on a data room or a set of accounts of your choice, with your teams' questions.