The problem is not the answer, it is the verification
Nobody in finance buys a conclusion without looking at its source. An M&A analyst reading the adjusted EBITDA of an issuer listed on Euronext Milan wants to know whether that figure comes from the half-year report, from which note, and with what reconciliations. A credit analyst assessing a covenant wants the exact paragraph of the facility agreement, not a plausible paraphrase. In this profession, the difference between a useful answer and a dangerous one is traceability.
General purpose AI tools answer from memory: they compress thousands of pages into a statistical representation and then generate fluent text. For finance this is unacceptable. A summary that looks correct but is not anchored to a precise page is not verifiable, and what cannot be verified does not belong in an IC memo, a credit file, or a due diligence report. AMARIL is built on this principle: every statement traces back to its source, with a reference to document, page, and paragraph.
IFRS and OIC: two grammars, one demand for precision
A European team works every day across two different accounting frameworks. Listed issuers preparing consolidated statements under IFRS disclose fair value, financial instruments classified by level, and goodwill impairment with its related tests. Companies applying the Italian OIC standards follow a historical cost logic, with their own depreciation criteria and provisions. Comparing two companies, or the same company over time after a transition between frameworks, requires knowing exactly which criterion was applied and where it is stated.
A model that answers from memory tends to flatten these differences: it presents a figure without saying whether it is an IFRS or OIC value, whether it includes or excludes non-recurring items, whether the line is already reclassified. AMARIL does the opposite. When it extracts gross operating margin, net financial position, or interest coverage, it reports the statement line and the note the figure comes from. The analyst sees the citation, opens the document, and checks in seconds. Review stays human, but it becomes fast instead of blind.
The same holds for documents that are not financial statements in the strict sense: management reports, filings submitted to Consob, prospectuses, term sheets, sections of a data room. The logic never changes. No statement without a source the analyst can open and read.
Source citation as risk control, not as a convenient feature
In a regulated environment, traceability is not cosmetic, it is part of the control. ESMA and Consob require consistency and accuracy in financial disclosure. An asset manager subject to SFDR must be able to show which data underpins a classification. A bank applying CRR/CRD and building a credit file must be able to reconstruct every input back to the original document. If a memo relies on an AI generated figure with no verifiable reference, the entire chain of accountability weakens.
Source citation solves two risks at once. The first is hallucination: a figure or a clause invented with a confident tone. By anchoring every output to the original text, AMARIL makes it immediately clear whether a data point actually exists in that form. The second is interpretive drift: the summary that, without lying, slowly shifts the meaning. With the paragraph alongside, the analyst separates what the document says from what the model suggests.
This also changes the pace of work. A comparable analysis across a set of DAX or FTSE MIB issuers, an earnings call summary, the cross reading of multiple filings: tasks that take days compress, but without giving up control, because every figure already arrives with its reference. This is not about automating trust, it is about automating verification.
European infrastructure: where the data lives matters
For a European finance team, the quality of the answers is inseparable from where and how the data is processed. An M&A data room, a loan portfolio, the confidential documents of a transaction cannot leave the perimeter of control. AMARIL runs on cloud infrastructure in the European Union, with end-to-end encryption, a zero retention policy, and full GDPR compliance. For institutions with stricter requirements, including under DORA on operational resilience, on-premise deployment is available, so that documents and prompts stay inside the client infrastructure.
The combination is precise: answers always cited, no hallucination passed off as data, and a treatment of documents that respects the European regulatory framework. These are the three pillars a team can genuinely rely on.
Takeaway
In finance the source is not a detail, it is the product. Reading IFRS or OIC statements, filings, and contracts with answers anchored to page and paragraph turns AI from a risk into a lever: faster at extracting, but always verifiable in what it claims. For those working in M&A, private equity, asset management, or credit, source citation is non negotiable, and that is exactly where AMARIL starts, not an optional extra.
