What does an AI consultant do for an accounting practice?
The work usually has three parts. Discovery maps where staff time goes, such as bill entry, document chasing, client queries and BAS preparation, and identifies which tasks have enough volume and clear enough rules to automate. Build connects tools to the practice's ledgers and inboxes and sets the review steps. Support keeps the workflow running as software, AI models and the client list change. Some consultants only advise, while others build and support. Know which one you are paying for before comparing proposals.
What should the consultant ask you first?
A consultant should ask about your work before talking about tools. Expect questions like the ones below in the first conversation, and treat their absence as a warning sign. The answers shape everything that follows, from which task to start with to how review is set up. If the first meeting is a product demonstration with no questions about your practice, ask for a discovery session before any proposal. Without your volumes and review steps, a consultant cannot say which task is worth automating first, or whether any is.
- Which ledgers you use (Xero, MYOB, QuickBooks or a mix) and how many client files.
- Monthly volumes of bills, bank transactions, client emails and calls.
- Where work waits: missing documents, queries or review queues.
- Who reviews and signs off, and what your engagement letters say.
- Whether you are a registered tax or BAS agent, and which obligations apply.
- What happens if the automation stops working on a busy day.
How will client data be handled?
Ask exactly where documents and extracted data go, which AI services process them, in which countries, and whether any provider can use the data to train its models. The OAIC's checklist for selecting AI products asks organisations to understand whether the system's terms of use give the developer access to information entered into it, and what the security risks are. For registered agents, the TPB's AI guidance says client permission is needed before client information goes to a third party, which can include entering it into AI tools depending on their configuration. The consultant should help you check what your engagement letters need to say.
Which obligations stay with the practice?
The TPB states that tax practitioners remain responsible for the tax agent services they provide when they use AI, and that AI output must be assessed with their own professional judgement before being relied on. Its guidance points to the Code's requirements on competence, reasonable care and confidentiality, and to the Determination's requirements to keep records of services and maintain a documented system of quality management. Where a consultant or its tools do part of the work, the TPB's outsourcing guidance says work by an unregistered party must be under the supervision and control of a registered practitioner. Ask the consultant to design the workflow so review and sign-off are recorded.
Who owns what gets built?
Get ownership in writing before work starts. A workflow is made of several parts: the automation platform account, the prompts and rules, the integration settings and the documentation. Ask whether the accounts sit in the practice's name or the consultant's, whether you can see and export the prompts and rules, and what happens to the workflow if you stop paying for support. If you cannot move the workflow to another provider, you depend on the consultant for every change.
How do you compare proposals?
Ask each consultant to describe a test on sample documents you are allowed to share, or fictional ones, and what they will measure. Compare scope, review steps, data handling, ownership, support response and exit terms side by side. Price comparisons only make sense once the scopes match. Treat any accuracy figure or time saving quoted before the consultant has seen your files as a sales estimate. The practice capacity calculator and the automation scorecard on this site can help you set your own baseline first.