The audit of expense claims is undergoing a quiet but radical transformation: whereas manual sampling was previously the norm, artificial intelligence now enables a comprehensive 100 per cent check of all claims before reimbursement. The aim is threefold: to reduce errors, detect fraud, and focus human effort where it truly adds value.

How it works

Extraction & understanding (OCR + NLP): the AI reads expense receipts regardless of their format: amount, date, VAT, merchant, hotel line items. It translates where necessary, then automatically matches the content against the declared type of expense and the applicable policy rules (spending limits, per diem rates, required supporting documents).

Cross-checks & verifications: The system detects duplicates (such as the same receipt submitted multiple times or the same transaction entered twice), checks the consistency of travel dates and locations, and verifies VAT compliance. Some solutions go a step further by cross-referencing data with external sources (public or private databases) to verify the existence and nature of a trader.

Risk scoring: Each report is given a score. ‘Green’ files are approved without issue. ‘Amber’ or ‘red’ files are referred for manual review, accompanied by specific reasons: per diem and meal allowances claimed on the same day, missing invoice, unusual supplier, incomplete hotel itemisation. The focus of the checks is thus concentrated where it is justified.

Distinguishing between error and fraud: The system tracks incidents per employee over time, making it possible to distinguish between an isolated accidental error and repeated deliberate behaviour. This level of detail is essential for dealing with cases in a proportionate manner.

Who benefits and how 

• Travelers / Submitters:
  • Fewer back-and-forth exchanges, fewer blanket rejections. Targeted prompts (“add the invoice,” “correct hotel itemization”) replace unexplained refusals, resulting in a smoother, more intuitive user experience.
• Approvers:
  • Focus shifts exclusively to qualified exceptions. Reports are faster to review, and decisions are better documented.
• Finance / Audit / Compliance:
  • 100% report coverage, with measurable reductions in errors and undue reimbursements.
  • According to a 2026 European study conducted by N2JSOFT among nearly 250 CFOs, 70% of finance leaders consider manual verification processes “entirely inadequate or excessively time-consuming.” Three out of four decision-makers now view AI as a “major asset” for fraud detection and improving the productivity of finance teams.
• Travel / Procurement
  • Stronger alignment with travel policy (hotel rate caps, per diem vs. meal reimbursement), and enhanced evidentiary value (VAT compliance, itemization) for supplier reconciliations and audits.

Figures published by publishers (for guidance only): up to −66% fewer errors, 10 times more discrepancies detected before payment, and up to 90 per cent reduction in audit time. 

Practical examples

• Hotel abroad: The AI reads an Italian hotel receipt, breaks down the items (room, breakfast, taxes), compares the amounts against policy limits and flags a missing VAT charge. The person entering the data receives a targeted request (to add the itemised invoice or correct the breakdown), without their report being rejected outright.

• Duplicate entries and per diem: A meal is recorded whilst a per diem is active for the same day. The AI links the two pieces of information and issues an alert: the user keeps one of the two, and the approver approves the entry without lengthy back-and-forth.

• Suspicious merchant / reused receipts
  An ‘entertainment’ expense is linked to a merchant that is inconsistent with the declared context, or to a receipt that has already been submitted in a previous claim. The system flags the case and places the report under manual review, without blocking the processing of the other claims in the batch.

What the market has to offer: from native modules to dedicated audit layers

Faced with increasing compliance requirements and the rise in document fraud, the leading expense management software providers have each developed their own AI-powered audit module. SAP Concur Detect by Oversight, Cegid Notilus, Rydoo Smart Audit and Expensya (Medius) now incorporate anomaly detection, scoring and policy compliance features directly into their platforms.

These native modules have a clear advantage: they utilise data as it is entered in real time and fit seamlessly into the user’s existing workflow. Their limitation is equally clear: they only cover expenditure entered into their own tool, and their scope of audit remains dependent on the software provider’s functional choices.

It is precisely this positioning that EVA by TEVASOFT complements. Whereas expense management tools perform audits from within, EVA operates as a dedicated and independent AI audit layer, designed to integrate with the T&E solutions already in place within the organisation (SAP Concur, N2F, Lucca and, more generally, market leaders) without altering existing processes or requiring migration. Entirely specialised in expense claim auditing, fraud detection and e-invoicing compliance (Factur-X, UBL, PDP 2026), EVA provides an additional level of control that native modules, by their very nature, cannot offer: a cross-functional, multi-tool view of the company’s entire expenditure portfolio.

Fraud detection is now a major challenge, and there are many tools available to address this need. 

The next challenge to tackle – an additional challenge that has been emerging since early 2025 – is the use of generative AI to forge supporting documents. The use of generative AI to forge supporting documents. In September 2025, 14 per cent of fraudulent documents submitted are reported to have been created by AI, compared with less than 1 per cent a year earlier. The race is on. This figure — 14 per cent in September 2025, virtually zero a year earlier — speaks volumes about the speed at which the risk is evolving. Will the tools manage to keep pace?