AI use case · Finance

AI expense claim and receipt processing for UAE companies.

For finance teams that check every expense claim against a pile of receipts, a policy document and a spreadsheet. AI reads each receipt, checks it against your policy and the VAT rules you set, and sends only the exceptions to an approver.

Ahmed Khaled, FounderLast updated:

Discuss this use case

A short note is enough. Phone and email let us reach you.

About: AI & Intelligent Automation

Tell us about the workflow, who uses it and what is difficult today. Please leave out passwords, confidential records and sensitive personal information.

Add company and other details (optional)

After you send this, choose a time for a free 30-minute discovery call. If no time suits, we get back to you as soon as possible to arrange one.

What is AI expense claim processing?

AI expense claim processing reads the receipts and bills submitted with a claim into fields such as merchant, date, amount, VAT and TRN. It checks each line against your expense policy and VAT rules, flags duplicates, missing receipts and over-limit spending, and routes exceptions to an approver. Clean claims move on for approval and payment.

The problem

Claims are checked line by line against a policy few people remember.

Expense claims arrive as phone photos, faded till receipts, emailed hotel bills and fuel slips, in Arabic, English or both. A finance officer opens each one, compares it with the claim form and the policy, and decides whether it is allowed and whether its VAT can be claimed.

The VAT question is easy to get wrong. The VAT Executive Regulation lists cases where input tax cannot be recovered, such as entertainment for people who are not employees and cars available for personal use. A receipt without the supplier's TRN, or a total that does not match the claim, needs a decision, and different reviewers often decide differently.

  • Retyping receipts

    Merchant, date, amount and VAT are keyed in from images of varying quality.

  • Policy by memory

    Limits, categories and approval rules are applied from experience, not consistently.

  • VAT missed or wrongly claimed

    Recoverable VAT is missed on some receipts and claimed on others where it should not be.

How it works

One claim, from receipt photo to approved payment.

An illustrative workflow. The capture channel, policy rules and target system are agreed for each project.

  1. Capture

    Employees submit receipts through the agreed channel: a mobile upload, an email address or your existing expense tool.

  2. Receipt reading

    AI reads merchant, date, currency, amount, VAT and TRN where shown, keeping a link to each value on the image.

  3. Policy checks

    Each line is checked against your rules: category, spending limits, required approvals and attendee names for business meals.

  4. VAT checks

    Receipts are checked for the details you need to recover VAT, and categories your adviser marks as non-recoverable are flagged.

  5. Duplicates and gaps

    The same receipt claimed twice, a claim line with no receipt and totals that do not agree are flagged with the reason.

  6. Approval and posting

    Clean claims go to the approver and exceptions go to finance with the reason shown. Approved claims post to your accounting or payroll system.

Who does what

What the AI does · What a person checks.

What the AI does

  • Reads receipts and bills in Arabic and English, from photos or PDFs

  • Links each value to where it appears on the receipt

  • Checks claim lines against the policy rules you set

  • Flags receipts missing a TRN or a VAT detail you need

  • Finds duplicate receipts across claims and employees

What a person checks

  • Every flagged line and every low-confidence field

  • Justified exceptions to policy, such as an approved over-limit spend

  • The VAT treatment of any category your adviser has not ruled on

  • The final approval before reimbursement

First pilot

A first pilot on claims you have already processed.

We test on claims your team has already decided, so the results can be compared with real decisions. Scope, timeline and price are confirmed in a written proposal.

  • One department or cost centre

    Start with a group that claims often, such as sales or site teams.

  • Your policy as rules

    Your expense policy written as clear rules with finance, marking where judgement is still needed.

  • A sample of past claims

    Real claims and receipts, shared safely after the first call, including poor photos and Arabic receipts.

  • Line-level comparison

    We compare the workflow's flags with your reviewers' past decisions, line by line.

Data, hosting and Arabic

Where the data goes, and how Arabic is handled.

  • Hosting

    The application and stored data can be hosted on Azure UAE North, on private servers managed by Vloud, or in your own cloud account. The AI model is used through a provider agreed per project, and its processing region may be outside the UAE.

  • Access

    Employees see their own claims, approvers see their team's and finance sees the exceptions. Receipts can include personal details, so access is set by role.

  • Arabic

    Arabic and bilingual receipts, often printed small on thermal paper, are tested as their own sample.

  • Retention

    How long receipts and claim records are kept, and who can see them, is agreed with your finance owner and tax adviser.

Honest limits

Where it does not work yet.

  • Faded thermal receipts and handwritten bills may not be readable. They go to a person rather than being guessed.
  • Policy judgement stays with people. The workflow flags an over-limit meal; it does not decide whether it was justified.
  • It applies the VAT rules your adviser approves. It does not give tax advice or decide the treatment of new cases.
  • We do not quote accuracy or time savings before testing on your claims. Results are measured on a held-out sample first.
  • If your expense tool already reads receipts well, the better step may be adding policy and VAT checks to it, and we will say so.

What drives cost and timeline. The number of employees and claim types, receipt quality and languages, and how many policy and VAT rules are needed. Cost also depends on the capture channel and the accounting or payroll system the claims go to. We don't publish a price list; scope, timeline and price come in a written proposal after the discovery call.

Evidence

What we can show you.

Every example is labelled: delivered work, our own product, a demonstrator, or an illustration with synthetic data.

  • Synthetic data

    The illustrative invoice workflow

    A walkthrough with a synthetic invoice showing extraction, checks and the exception a person resolves. It uses an invoice, not an expense claim, and is not a live client connection.

  • Illustrative

    How to test AI document extraction

    The evaluation method we use on a pilot: development set, held-out set, field-level results and review rate reported separately.

Before you start

Questions buyers ask.

Can employees just take a photo?

Yes. Phone photos are the usual input. Clear, flat photos read best, so the capture screen can ask for a retake when an image is too blurred to read.

Does it check VAT on receipts?

It checks that the details you need are present, such as the supplier's TRN and the VAT amount, and flags categories your adviser marks as non-recoverable. The VAT Executive Regulation sets out the cases where input tax cannot be recovered.

What about receipts in Arabic?

Arabic and bilingual receipts are handled and tested as their own sample, because small print on thermal paper is often harder to read than the language itself.

Does it replace our expense tool?

Not necessarily. It can add receipt reading and policy checks in front of your current tool or accounting system, through a supported API or an agreed import route.

Who approves the claim?

Your approvers, as today. The workflow sends clean claims to them and exceptions to finance with the reason shown, and it records every decision.

Where is the data hosted?

The application and stored data can be hosted on Azure UAE North, on private servers managed by Vloud, or in your own cloud account. The AI model is used through a provider agreed per project, and its processing region may be outside the UAE.

Who owns the code?

You own the source code and IP of the custom work we build for you. Our own products stay ours, and any licence terms are agreed separately.

Next step

Test it on your own documents.

Tell us what the work looks like today. We reply with the questions we need answered before a pilot can be scoped.

Discuss this use case

A short note is enough. Phone and email let us reach you.

About: AI & Intelligent Automation

Tell us about the workflow, who uses it and what is difficult today. Please leave out passwords, confidential records and sensitive personal information.

Add company and other details (optional)

After you send this, choose a time for a free 30-minute discovery call. If no time suits, we get back to you as soon as possible to arrange one.