VAT4U was built to move large volumes of travel expenses and accounts payable (AP) invoices through a VAT recovery process. The single biggest efficiency gain available to you as an Expert is not any individual feature - it is a change of working method: stop opening transactions one by one, and start driving them in batches, using Smart AI to enrich the data and a small set of statuses and reports to steer the flow.

This guide walks through that process step by step: running Smart AI in batch, using the specialised checks, the Human Review flag, reviewing and applying AI results, closing the loop with statuses, credit (AI Transactions) management, and the Processing Analytics Report.

Opening a small invoice costs more than the VAT inside it. Use Smart AI plus filters to push the bulk of transactions forward or out without ever opening them, and spend human time only on lines that are flagged, unusual, or with high amounts.

The traditional way of working is to open an invoice, read it, enter or correct the data, save, and move to the next one. Even with AI filling in the fields, this remains human data entry - just faster. It's acceptable for a handful of AP invoices. It does not scale for travel expenses, and it destroys profitability on small transactions.

The recommended pattern is a triage. After running Smart AI on a filtered population, split the result into three groups:

Everything else in this guide exists to make this triage fast, auditable, and repeatable.

Smart AI is not a single AI call. It combines OCR, VAT4U's own parsing and validation logic, AI extraction, and a second AI layer that reviews the quality of the first result.

On each transaction, it can:

Two design decisions matter for your process:

Never trigger AI from inside individual invoices when you have volume - a single extraction can take up to 30 seconds, and clicking, waiting, and moving to the next line wastes your day. Run it once on the whole expense set and let it process in the background.

In Manage Expenses, filter the transactions you want to enrich: company, country, period, refund status, missing data, and so on.

Lines without an attached document are automatically skipped.

Open the Smart AI menu and choose AI Data Enrichment. The other entries in this menu are specialized checks - see Specialized checks below.

Tick the option to compare AI results with your original data when that data is genuine (for example, imported from SRXP). Leave it unticked when the original data is placeholder data - for example, estimates where every line was imported as 10 EUR, Germany, January 1. Comparing against fake data produces noise.

Start the run. Each run is saved as a tagged AI batch, so you can retrieve and review that specific run later.

Track progress under Reports → Running Jobs, where you can see how many of the selected transactions have been processed.

When one-by-one is fine. Manually opening an invoice and clicking the AI button is acceptable for occasional AP invoices where you review everything anyway - but first process them by batch AI, so the AI process is already fast when you switch to Data entry. For travel and domestic volumes, always work in batch.

Besides AI Data Enrichment, the Smart AI menu offers targeted checks. Each one does a single thing, slightly better than the general enrichment does it. Use them as an extra pass on data you already have, or when a specific risk matters for a claim.

New checks can be built on request - if your market has a recurring risk that none of the checks covers, raise it with the VAT4U team.

The Human Review field is populated by a dedicated AI agent whose only job is to decide whether a line deserves your attention, and why. A line is flagged in two situations:

Management can impose review rules independent of AI confidence.

The setting exists at two levels:

The threshold is currently global per company, not per country.

Use it commercially. The review threshold is a discussion point with your clients. Some will trust AI fully; others will pay for systematic human review above an agreed amount. Agree on the parameters, configure them per claimant, and your process becomes both tailored and auditable.

After a batch run, open the Smart AI results preview for that run: click Show Preview Smart AI Results, then See Results.

Step 2: Read the per-field summary

Instead of reading 200 invoices, you read one summary. For each field, the platform reports:

From any difference count, you can drill down: the platform lists the AI value against the original value line by line, opens the underlying document on click, and can open the Manage Expenses table pre-filtered on the differing lines.

This is where you judge whether the AI or the source data is more trustworthy. A common finding: employees enter the order or transaction number where the invoice number belongs, or the declaration date from the expense tool differs from the real invoice date. In both cases, the AI value is usually the correct one.

Once you have decided, apply the AI results with a strategy per field:

Before executing, the platform summarises exactly how many lines will be overwritten per field. Overwriting is definitive for those fields - read that summary before confirming.

Lines that are already claimed are protected and will not be modified.

AI Transactions. Applying AI results consumes no credits. Transactions are spent when the AI runs, not when you use its output. Run once, and exploit the results as many times as you need.

The triage described earlier is executed with two batch actions in Manage Expenses.