Sending a request list is often the easy part. Documents then arrive in several batches. Some cover the wrong period; others need an explanation. The list itself also needs updating when the engagement changes.
We see useful opportunities for ChatGPT in this preparatory work. The starting point is to distinguish information already available from decisions the firm still needs to make.
Updating last year’s list
An existing list contains valuable knowledge. Someone still needs to compare it with changes agreed for the current engagement. One possible use is a proposed comparison linking existing requests to those changes, ready for the firm to review.
Working with tables and supplied files provides the technical basis. OpenAI describes these capabilities in its data analysis documentation. The existing list and engagement notes provide the reference for suggestions. The firm determines which requests belong in the current engagement.
Making follow-up questions clearer
An internal note such as “bank confirmation incomplete” may tell the client very little. A useful follow-up needs to explain what is missing and which document is required.
Once the firm has established the facts, a draft can help with the wording. We find this particularly relevant where similar questions are written repeatedly. The professional finding comes from the firm; the tool can support its explanation.
OpenAI’s Academy for finance teams explicitly includes PBC coordination. This includes overviews showing who provides which documents and which requests are still outstanding.
From individual drafts to ongoing support
Ongoing coordination requires access to the relevant systems. With suitable connections and defined permissions, an agent could take on further tasks, such as identifying new documents or updating progress.
An ICAEW practitioner report offers a look at how these applications are developed. Dudley Gould used ChatGPT to help build an experimental prototype for standardising ledger data. In this custom application, AI proposes mappings that the user confirms or corrects. We find the way it brings professional knowledge into a recurring task particularly interesting.
Where we would start in a firm
We would choose a recurring writing task with a clear basis, such as drafting questions from reviewed findings or proposing changes to an existing list. This puts the support where time is regularly spent on wording and list maintenance.
For work involving client documents, the firm determines which data the tool may process and who receives access. The Austrian Data Protection Authority provides an introduction to these questions.
If the main burden is reviewing incoming files, our article on Claude Cowork considers that part of the work.