Agentic AI and indirect tax: a practical view for tax directors

Tax directors frequently operate at the intersection of compliance pressures and business expectations.

On one side, tax authorities are moving towards e‑invoicing, e‑reporting, and faster access to more detailed transactional data. Errors surface earlier, tolerance for manual fixes is shrinking, and tax audit questions arrive sooner and with more precision. In these circumstances, being ‘audit ready’ becomes a real-time everyday requirement.

On the other side, the business expects tax to be responsive and able to explain the potential impacts of supply chain changes, new market entry, pricing decisions, and structural shifts on demand, and without slowing things down.

Many tax directors are trying to meet both these demands with fragmented systems, inconsistent data, and teams already stretched thin. It’s in this context that AI, specifically agentic AI, has the potential to make a practical difference.

Indirect tax teams have used automation for years. Rules‑based engines calculate tax. Reporting tools support filings. More recently, generative AI has helped with document extraction, anomaly detection, research, and drafting responses.

Agentic AI goes a step further. It uses agents, or ‘digital workers’, combined with advanced reasoning models to streamline data analysis and support aspects of decision‑making and implementation. Instead of helping with individual tasks, agentic systems can:

• Work across multiple tax and finance systems;

• Follow defined objectives;

• Execute sequences of work end‑to‑end; and

• Surface results for human review where judgement is required.

This is not about replacing the judgement or accountability of tax professionals. To the contrary. It’s about reducing the volume of manual coordination, rework, and follow‑up that consumes a tax team’s time today. It’s about empowering tax professionals with tools powerful enough to elevate their game and move upwards in the tax value chain. At the same time, tax professionals must remain responsible for decisions taken, and make sure the agentic AI operates within established governance frameworks.

Many indirect tax issues don’t start in tax, they start upstream. Missing fields, inconsistent configurations, timing differences, and poor system handoffs all show up later as reconciliations, adjustments, or tax audit questions.

A consistent lesson Deloitte has heard for years is reinforced with AI adoption: outcomes depend on data quality. If the data is unreliable, automation and AI struggle to deliver lasting value. Data improvement can involve upskilling teams in basic data management and ensuring everyone understands how clean, well-structured data underpins successful AI projects.

AI can also support data cleansing and diagnostics by automating the identification of gaps and inconsistencies. Agentic AI can be positioned closer to where data enters the process; for example, within master data set-up, accounts receivable/accounts payable flows, or tax determination steps, to extract, enrich, and validate data as transactions occur. Instead of identifying issues at month‑end or return time, AI agents can help identify them from the outset, reducing the likelihood that they escalate downstream.

For a tax director, the practical impact can include:

• Fewer late adjustments;

• Fewer surprises during close;

• More confidence in the data supporting signing off; and

• Less time spent by the tax team performing low-value data remedy, cleansing, and reconciliation tasks.

VAT, goods and services tax, and sales tax filings remain a core responsibility of most tax directors, even as authorities move towards more real‑time models. These processes are structured and repetitive, creating potential for further automation.

Agentic AI can work across ERP systems, tax engines, and reporting tools to support gathering data, assist with reconciliations, populate returns, and facilitate submission while keeping review and approval with the tax team.

From a tax director’s perspective, the potential value is not just speed. It’s also:

• Fewer manual handoffs;

• More consistent application of rules; and

• Clearer visibility into where judgement is actually being applied.

Agentic AI can also improve indirect tax accruals by addressing data gaps and providing a view of exposures, potentially reducing the need for conservative over‑accruals driven by uncertainty.

For many tax directors, the issue is not a lack of ideas but limited time to execute.

When routine activities such as data cleanup, document extraction, and variance chasing are automated, tax teams gain capacity to focus on work that requires experience and judgement: advising the business, preparing for audits, and anticipating risk.

This is where the idea of a ‘digital workforce’ becomes tangible. AI agents support repeatable tasks consistently, while people focus more on interpretation, escalation, and decision support.

Over time, this allows tax to engage earlier in decisions rather than being asked to validate them after the fact.

Tax audit support often pulls experienced people away from day‑to‑day work, especially when documentation must be reconstructed under time pressure.

Agentic AI can support audit readiness by logging actions, data sources, and rationale as work happens. This can create a clearer audit trail and helps reduce reliance on institutional memory when questions arise.

AI agents can also help identify patterns that may indicate audit risk before authorities raise formal inquiries, allowing issues to be addressed earlier.

When documentation is requested, agents can retrieve, assess, analyse, and summarise invoices, contracts, or customs records, helping tax teams respond more quickly and consistently.

Agentic AI does not remove responsibility from the tax director. Poor implementation can create real exposure, including filing errors and penalties.

Deloitte's Tax Transformation Trends 2025 research found 77% of tax leaders surveyed require at least 90% accuracy before trusting AI. Human oversight and professional judgement remain essential, particularly for classification, calculation, and filing decisions. Agentic AI works best when it assists with execution and coordination, while people retain control over judgement.

Equally important for successful AI use is readiness to implement and govern agentic AI safely in live tax processes. This includes:

• Data architecture;

• Legacy system constraints; and

• Governance frameworks.

All of which affects what is achievable in practice.

Teams also need enough familiarity with AI to challenge its outputs, understand limitations, and identify where further automation makes sense.

Agentic AI is already being applied in practical, incremental ways across indirect tax. It doesn’t require a full redesign of the tax function, but it does benefit from deliberate choices about where to start.

For tax directors, the most common early benefits are:

• Fewer data issues late in the process;

• Smoother filing cycles;

• Improved confidence in accruals;

• Better tax audit readiness; and

• More time for advisory work.

As indirect tax moves to more continuous digital models, strong data and targeted use of AI will be important to keep pace without putting unsustainable pressure on teams.