Claude Cowork

Claude Cowork and incoming audit documents

Reviewing files, matching requests and preparing follow-up questions. Where an AI agent could support PBC work and what existing examples tell us.

Velatum · Updated

As documents arrive, the initial review begins. The audit team checks periods and entities and matches files to outstanding requests. Keeping track takes time, particularly when documents arrive in several batches.

An agent could help with this initial review. A useful output would link proposed matches back to the files and leave unresolved points visible.

Turning a collection of files into an overview

Claude Academy describes Cowork organising scattered documents for a SOC 2 audit, including an inventory and a mapping to requirements.

This suggests a related use for PBC work in Austrian firms. An existing request list could structure the initial review, with proposed matches and open points for each entry.

These proposals provide a starting point for professional review. The team assesses whether each document contains sufficient information and is suitable as audit evidence.

More precise follow-up requests

Armanino provides a relevant industry example. The US firm uses Claude in a custom application to make responses to incorrect document submissions clearer.

The underlying problem is familiar. Telling a client that a document is wrong can lead to another round of questions. Explaining what is missing and what is needed instead can make the next response more useful.

Armanino built its application using Claude through Amazon Bedrock and incorporated it into its own process. The example illustrates how a focused AI application can support communication about documents.

Access and permissions are part of the process

Cowork can read files and perform actions through authorised tools. Anthropic explains the types of access in its guidance on using Cowork.

The firm determines which steps the agent may carry out. Proposals in a separate overview are a useful starting point. Changes to original documents and sending follow-up questions can have their own approval points.

Where the support fits

We see potential in recurring initial review involving many documents. The aim is to make files easier to find and uncertainties easier to spot. An overview linking to originals gives the team a shared starting point for further work.

An existing client portal may already cover some of this work. Before adding another tool, it is worth identifying where manual effort remains.

Privacy and professional confidentiality are part of setting up access to client documents. The firm determines which collections are accessible and who may view the results. The Austrian Data Protection Authority provides an initial privacy overview.

Our ChatGPT article looks at work on the request list itself.

Sources and further reading

Try it together: our planned AI workshop for advisory firms →