Accounting Policy Q&A Agent · Use case
How can we use an AI agent to answer questions on our accounting manual and group policies on our own infrastructure?
An accounting policy Q&A agent answers a colleague's question from the company's own accounting manual and group policies, cites the paragraph it used, and says when they are silent. The colleague reads the paragraph and applies it. The agent reads no ledger, decides no accounting treatment, posts nothing and changes nothing.
The answer is in the manual.
Finding the paragraph takes the time.
- ✗ 01The accounting manual and the group's policies run to hundreds of pages, change each year, and are applied by accountants in every entity and business unit.
- ✗ 02The same questions reach the technical accounting team by email at every close, and an answer from memory in one entity becomes a difference to explain in consolidation.
Where it stops for a person
Always. It informs, and a judgement on treatment goes to the technical accounting lead.Always. It informs a colleague, it does not decide a treatment: it applies no policy to a real transaction, proposes no journal and gives no view on what the accounting standards require beyond what your manual says. A question the manual does not settle is the technical accounting lead's, and the auditor's view is sought by a person. The manual, not the answer, is what the company relies on.
It answers with the paragraph cited.
The technical accounting lead judges treatment.
Step · 01
An accountant or a finance manager asks a question in plain words.
Step · 02
The agent searches the manual and the policies that colleague is allowed to read.
Step · 03
It answers from the text, with the document, the paragraph and the date of the version it used.
Step · 04
Where the manual is silent, out of date or disagrees with another policy, it says so and does not guess.
Step · 05
The colleague reads the paragraph and applies it, and takes any judgement on treatment to the technical accounting lead.
Reads from · connectors
The accounting manual, group policies and the technical team's issued guidance notes, indexed as a knowledge source with each reader's permissions kept. No ledger, no transaction and no system of record.
Acts through · MCP servers
Nothing. It has no tool that changes anything, and it writes to no system.
What you set up
- Index the current accounting manual, group policies and guidance notes as a knowledge source, and remove the versions no longer in force.
- Set which entities and teams may read which documents.
- Build the agent in Katonic Studio and test it on questions your technical accounting team has already answered.
- Publish it to one entity's finance team first.
Where it runs
The agent, its knowledge store and the model run on infrastructure you own or control. You choose the model and serve it yourself, so questions, documents and answers are not sent to a public AI service. The accounting manual, group policies and the questions asked stay inside the company.
How you get it today
This page is labelled Use case: the agent is built in Katonic Studio. You describe the job, connect what is named above, add test questions and release the agent once it passes them. When the Catalog holds a tested template for this job, this page will carry the Template label and show how to take it.
Does it decide how a transaction is accounted for?
No. It quotes the manual and cites the paragraph. The accountant applies it, and a judgement on treatment goes to the technical accounting lead.
What does it do when the manual does not cover the question?
It says so, and says which documents it searched. It does not guess, and the question goes to the technical accounting lead.
Does it read our ledger?
No. It reads documents only. It has no connection to a ledger or to any system of record.