VAT-05 · FlowCount VAT

VAT return preparation

A draft VAT return and its exception pack, built from transaction data.

مسودة إقرار ضريبة القيمة المضافة من بيانات المعاملات

Design partners wantedFor: VAT manager · Tax & accounting practices

7 to 8 h

of manual preparation per client per return today

Basis: practitioner estimate, manual preparation today

The problem

Preparing a return still means exporting the GL and registers, pasting them into last quarter's working file, repairing the formulas and mapping each line to a return box by hand. Practitioners put it at seven to eight hours per client per return before anyone reviews it, and larger clients post up to a million GL lines a month.

“Seven to eight hours per client per return, most of it rebuilding the same spreadsheet.”

What FlowCount does

Map every transaction to a VAT return box; classify standard-rated, zero-rated, exempt and out-of-scope supplies; separate recoverable from non-recoverable input VAT; and compare the draft with prior returns, with the material movements explained.

AI and rules used
Rules for box mapping; LLM classification of descriptions, with abstention
Builds on
VAT return ↔ GL ↔ e-invoice reconciliation

How it works

Who does each step

  1. 01Rules

    Map to return boxes

    Tax code, account, document type and invoice attributes first. The description is a secondary signal.

  2. 02AI

    Classify the rest

    Propose 15%, zero-rated, exempt or out of scope for lines the rules cannot place, with the words relied on. Low confidence is flagged, never forced.

  3. 03Rules

    Test and compare

    Invoice fields, duplicates, missing VAT, credit notes, reverse charge, imports and manual VAT entries; movement against up to five prior returns.

  4. 04Reviewer

    Approve the draft

    A person approves uncertain treatments, material adjustments and the final return. FlowCount never files.

What you send

  • GL (up to 1M lines a month)
  • Sales and purchase registers (up to 1M lines a month each)
  • Up to five prior VAT returns
  • Tax code mapping from the ERP (up to 50 codes)

What you receive

  • Draft VAT return by box
  • Exception report
  • Reconciliation to the GL
  • Reviewer notes and movement commentary

Controls built in

  • Raw sign kept beside the normalised economic sign
  • Low-confidence classifications flagged, not concluded
  • Human approval for uncertain treatment, material adjustments and filing

KPIs we report

  • Preparer hours per return
  • Lines placed by rules vs proposed by AI
  • Review notes per return

Sample input → expected output

What the result looks like

Basis: practitioner estimate, manual preparation today

How a line finds its return box (illustrative)
SignalExampleReturn box
Tax code S, Saudi customerLaptops sold to a Jeddah retailerStandard rated sales
Tax code Z + export declarationTablets exported FOB JeddahExports
Tax code E, residential leaseStaff accommodation rentalExempt sales
Document type: customsSmartphones imported, VAT paid at the borderImports subject to VAT paid at customs
Non-resident supplier, serviceCloud infrastructure from an overseas providerImports subject to VAT accounted for through reverse charge
Credit note linked to an invoiceDefective units returnedSame box as the original invoice, negative
No tax code, manual journalYear-end VAT reclassificationINDETERMINATE → T2 proposal → review

Portfolio scorecard

Where it sits on the roadmap

Value

4.50 / 5

Feasibility

3.90 / 5

Priority index

5.85 · rank 12 of 27

Quadrant

Quick win

Delivery

Wave 1 (0-3 m)

T-shirt size M: 4 to 8 weeks with 2 engineers + 1 tax SME + 0.25 QA, indicative build 120k to 275k. New data source or document type + rules + review-UI changes. How we score and sequence →

A draft VAT return and its exception pack, built from transaction data.

Design partners shape the rulebook and review console, and get first access on pilot terms.