Skip to content
omarsoliman.dev
AI for FinanceAITechnology

Three layers of AI in tax — and the data problem underneath

Omar Soliman
Omar Soliman — Indirect tax + AI · Cairo
June 14, 2026 · 2 min read
An editorial illustration of three ascending tiers — gears and routine marks at the base, stacks of ledgers in the middle, and a single figure pausing to decide at the top beneath a star — the task, knowledge, and judgment layers of AI in tax.

“AI will replace tax teams.” We’ve been hearing it for two years — and the claim isn’t the problem. The missing distinction behind it is.

Most discussions conflate three fundamentally different layers of AI in tax. Separate them, and the conversation gets a lot more useful.

The three layers

Level 1 — Task

Automates repetitive, rules-based work: data extraction, VAT return preparation, invoice matching, reconciliations. Already operational today. It reduces effort, not headcount.

Level 2 — Knowledge

Interprets legislation, answers technical queries, reviews contracts for tax exposure, drafts positions. Evolving fast — it compresses hours of junior work into minutes.

Level 3 — Judgment

Supports decisions made in ambiguity: structuring, audit strategy, risk positioning, stakeholder alignment. Not replaceable. Still fundamentally human.

The gap few teams are addressing: data readiness

AI’s effectiveness is capped by the quality of the data beneath it. In many organizations, tax-relevant data is fragmented across systems, inconsistently structured, and not easily query-able. The usual suspects:

  • No standardized taxonomies
  • No single source of truth
  • Heavy reliance on manual extraction and spreadsheet adjustments

The right order of operations

The sequence is strategic, not technical:

  1. Establish data readiness — structure, standardize, and govern tax data so it becomes usable and query-able.
  2. Deploy Task AI — automate execution across clean, connected datasets.
  3. Scale Knowledge AI — cut repetitive advisory and internal query cycles.
  4. Preserve Judgment value — focus human expertise where it creates differentiation.

Teams rushing to “implement AI” on top of fragmented ERP landscapes and unstructured data are building on unstable foundations. The competitive advantage today isn’t the AI tool itself — it’s the organization’s ability to make its data usable for AI.

The real question

It’s no longer “Will AI replace tax teams?” It’s this:

So, honestly: what’s the primary constraint in your environment today — AI adoption, or the data beneath it?

My weekly brief

More writing
All posts
Editorial clay-and-paper illustration in a warm cream, terracotta, ink and olive palette: a stack of messy handwritten paper invoices on the left connects by curved ink wires and plug connectors through a row of four rounded terracotta-clay automation nodes — an n8n-style workflow — into a tidy, neatly-ruled VAT-ready ledger on the right, where one tax row is highlighted in olive and a small gold pile of coins sits beside the AED currency mark. Messy invoices in, a structured VAT-ready ledger out.
Jul 1, 2026

Invoices land in your inbox, a VAT-ready ledger builds itself: a no-code n8n pipeline

AITechnologyMENAVATn8nautomationno-code
Warm pixel-art editorial illustration in the clay-and-paper palette: a friendly blocky terracotta robot 'tax agent' stands at the centre holding a tax return form and a calculator, with thin ink connector lines linking it to a small network of smaller robot agents arranged like an org chart, plus pixel icons of invoices, a ledger, verification check-marks and a gold eight-pointed star — a network of autonomous AI agents running a tax department.
Jun 21, 2026

The Zero-Human Tax Company: How Multi-Agent AI Will Transform Tax Operations

AITax AutomationMulti-Agent AISLMVATCorporate TaxFinance Transformation