Insight
Numbra
As finance teams adopt tools such as Claude, Copilot and ChatGPT, the question is no longer whether AI can be useful in annual reporting.
It clearly can.
The more important question is where its role should end,and where human judgement must take over.
Finding information is not the same as validating it
An annual report draws on financial statements, accounting papers, board materials, risk assessments, sustainability data and previous disclosures.
But access to information is not the same as knowing which information is authoritative.
A model may find two versions of the same number. It cannot independently determine which one has been approved, whether the difference is intentional, or whether a subsequent adjustment has been reflected everywhere it should be.
The same applies to the text sections. AI can suggest language that sounds clear, balanced and convincing. It should not decide whether that language fairly represents the company’s performance, risks or financial position.
The ability to generate is not the same as the authority to approve.
AI should not become the process
A finance team could give a frontier model access to previous reports, spreadsheets and internal documentation.
A frontier model can already search, summarize, compare and draft across large volumes of reporting material. It can identify inconsistencies, surface potential gaps and materially accelerate parts of the reporting process.
But those capabilities do not, on their own, create a controlled reporting process.
A chat history is not a controlled reporting workflow. And an AI activity log does not demonstrate that a number was updated, reviewed and approved everywhere it appears.
The difficult part of annual reporting is rarely producing one paragraph.
It is managing hundreds of connected decisions across finance, management, investor relations, communications, legal, auditors and the board.
The question is no longer whether AI can produce useful output. It is how that output becomes part of a controlled reporting process.
Where AI creates value
Annual reporting should not be treated as a blank-page writing exercise.
Much of the relevant information already exists. AI can help retrieve that information, compare periods, prepare first drafts and direct reviewers towards areas that have changed.
But its output should remain connected to verified sources, controlled review and the people responsible for the final decision.
The objective is not 100% autonomous reporting.
It is to remove work that does not require human judgement – searching, copying, comparing and reconciling – so that people can spend more time on the work that does.
What Numbra is building
At Numbra, we see the AI model as one component of the solution, not the product itself.
Our role is not to ask companies to choose Numbra instead of ChatGPT, Claude or other general-purpose AI tools.
It is to combine more than a decade of annual reporting expertise with purpose-built technology, so that AI can be used within a structured and controlled reporting process.
Our first 2026 engagements will deliberately be high-touch and our immediate focus is practical: helping teams structure existing information, reduce manual work and maintain control over important reporting decisions.
Over time, more of the reporting process can be connected and supported by technology. But human review and approval will remain essential.
This article was developed with AI support and reviewed, edited and quality-assured by the Numbra team.
More on Our Approach
