It is software in which a language model plans and carries out a task in several steps, using tools such as search, spreadsheets or code. It acts, records what it did and can be tested.

Can an agentic system give tax advice?

Not on its own. It can gather facts, find the relevant rule and draft an analysis. A qualified person reviews the draft and takes the decision. That review is part of the design.

How is the quality of the output managed?

Through design. Calculations are handled in code, the model is used for reading and drafting, and outputs are compared with test cases. Human review remains part of the process.

Do we need a large data platform first?

Usually not. A first agentic system can work on a narrow task with the documents and ledgers you already have. Starting small also shows what is worth scaling.

02Technology

Agentic systems for finance and tax teams

Design, controls and handover of agentic systems for finance and tax functions, with people kept in the loop.

Written by Ajit Jain, founder of Averise. Last reviewed 13 November 2026.

An agentic system carries out a multi-step task with tools, under rules, and reports what it did.

In finance and tax, accuracy and review matter most. Averise helps teams design agentic systems around a clear task, sensible tests and human review, so the work stays understandable.

At a glanceFrom a task to a working agentic system

Tasks
Where an agentic system can help
Design
A clear role, tools and limits
Tests
Cases with expected answers
Review
People in the loop
Handover
A team that can maintain it

Where agentic systems help

Agentic systems suit work that is repetitive, rule-bound and reviewable: reading contracts, reconciling ledgers, drafting first versions and tracking dates.

We start with an inventory of tasks and rank them by value and by how easily the output can be reviewed.

Design principles

A few principles guide the way we design agentic systems for finance and tax work.

Calculation is handled by code, with the model used for reading and drafting
Statements about law are linked to their source
Sensitive data is protected before a model is used
Review is separated from drafting
People stay responsible for decisions

Build, test and hand over

An agentic system is typically handed over with its instructions, its permitted tools and a set of test cases, so a team can maintain it.

Documentation covers what the system is for, where its limits are and how it is reviewed.

AI advisory for leadership

For boards and finance leaders we provide a plain assessment: which uses of AI are ready, which need more time, and what controls a regulator or auditor would expect to see.

Questions clients ask

What is agentic AI?: It is software in which a language model plans and carries out a task in several steps, using tools such as search, spreadsheets or code. It acts, records what it did and can be tested.
Can an agentic system give tax advice?: Not on its own. It can gather facts, find the relevant rule and draft an analysis. A qualified person reviews the draft and takes the decision. That review is part of the design.
How is the quality of the output managed?: Through design. Calculations are handled in code, the model is used for reading and drafting, and outputs are compared with test cases. Human review remains part of the process.
Do we need a large data platform first?: Usually not. A first agentic system can work on a narrow task with the documents and ledgers you already have. Starting small also shows what is worth scaling.

Sources

This page is written in general terms and cites no legal instrument.

General information, not advice on your facts.

Tell us what you are working on.

Start a conversation