First contact with a client about missing items usually happens late. A preparer opens the file, starts the work, and only then discovers the prior year had something this year does not, or a document never came in. By the time anyone catches it, hours are gone and the client is hearing about it weeks into the engagement instead of on day one.
A pre-prep assessment workflow that reads each document as it arrives, files it to the correct client folder, and flags anything it cannot match for review. It then compares the assembled file against the prior year return and identifies what is missing before any staff time is spent. The output is a draft request to the client for the missing items, ready for the partner or manager to review and send.
First contact moves to day one. The preparer opens a complete file instead of chasing gaps.
We have spent two years using these tools on real returns in active practice. They earn their place on narrow, well-defined work. Prior year comparison, document matching, first-pass review, intake. What they do not do is judgment, accountability, or client communication. That stays with the professional, and that is the point. The technology handles the production layer. You keep the work that carries the fee and the risk.
AI is reliable for
- Reading and classifying documents
- Extracting figures from source records
- Organizing and routing files
- Tracking what has come in and what is missing
- Drafting routine client communication
- Cross-checking entries against source
Judgment stays with the practitioner
- Deciding whether a position is defensible
- Weighing a client's specific facts
- Choosing how to handle a close call
- Determining whether a form applies
- Final review and the signature
- Anything a regulator would question you on
Before any productivity question, there is a gating concern every firm should resolve first: client data. Firms hold some of the most sensitive financial information their clients possess, and any use of AI has to begin with knowing where that data goes and who can see it. A firm should be able to answer these before adopting any AI tool:
- Does client data leave the firm's controlled environment?
- Is it used to train external models?
- Who, inside the vendor, can access it?
- Is it retained, and for how long?
- Does the arrangement satisfy the firm's confidentiality obligations and its clients' expectations?
The workflows we build run inside the firm's own Microsoft 365 environment, so client data stays within the firm's controlled tenant rather than being handed to an outside platform. Data is not used to train external models, and access stays governed by the firm's existing Microsoft security and permissions, the same controls already protecting the firm's other client data.
The point is not that AI is unsafe. It is that AI must be deployed in an architecture where the firm keeps control of its data, which a practitioner who has worked this through can build deliberately rather than hoping a vendor has handled it.
Each one targets a specific point in the return lifecycle, and each is configured around how your firm actually works, not around a standard workflow no partner actually runs.
This is not a platform you buy and adopt. It is built to fit your existing process, by working CPAs who understand the work from the inside. The configuration is the project. The judgment stays at the signature line.