.svg){kind=logo}
On this page
Measure the work, not the signal.
See productive time and utilization by team, not just what looks active.
Key takeaways
- AI tax preparation pays off first in the prep before the return: collecting, sorting, and keying client documents.
- Positions, review, and the signature stay with a licensed preparer.
- Firms that skip measurement often automate the return itself, which isn't always where the repetitive effort sits.
- Measure time per step, waits, review loops, and rework before you choose what to hand over.
- Set the agent's rules before busy season, so it's ready when volume peaks.
According to 2024 Journal of Accounting Education research, more than 300,000 accountants left the profession between 2019 and 2022.
The Thomson Reuters Institute names thin talent pipelines and retiring senior staff among the top barriers to profitability at many tax firms.
Thomson Reuters also reports that a third of tax firms use generative AI in their work, with 14% using agentic AI.
Your team may already use an AI tax tool for research or drafting. Even so, the returns pile up in busy season, and source data still gets keyed in by hand.
The question is: which part of the tax workflow should you hand to an AI agent first?
Your own data can show you which steps to start with.
What is the first tax process to replace with an AI agent?
Start with the work that comes before the return. Collecting and sorting client documents and keying source data into the tax software are high-volume, repeatable steps an AI agent can handle. A licensed preparer still owns the judgment calls, the review, and the signature.
Measure how your team spends busy season before you choose what to automate. The most repetitive work often sits somewhere the firm doesn't expect.
Think back to the first week of March at your firm. Every client sends similar paperwork, and every return needs similar handling before a preparer can start. The intake moves data between the client portal, email, and file storage, and then to the tax software.
That's why AI for tax preparation works best at the front of the workflow. It's the part of the work that repeats for every client and follows clear rules. It also ends in a review step, so the preparer stays accountable for what goes out.
An agent that collects the documents, sorts them, and keys the numbers is agentic process automation at work on your intake. Your preparers still make the calls on positions, and a licensed preparer still signs off.
Tax planning, judgment-heavy positions, and client advisory conversations make poor first picks for automation. They depend on human judgment about context that a checklist can't easily capture. This is where your senior staff adds the most value.
Why most firms start AI tax preparation with the wrong task
AI adoption is staff using AI tools in their daily work: Human + AI.
AI implementation is handing an existing process to AI, with people reviewing and approving: AI + Human. An agent sorts the client documents and keys the data, while a preparer reviews the result.
In its 2026 report on professionals across legal, tax, accounting, and other fields, the Thomson Reuters Institute found that 40% of professionals say their organization uses generative AI, up from 22% the year before.
Many firms are still at the adoption phase. Tax professionals use generative AI for research and drafting, one case at a time. This is how AI for accounting firms usually begins.
The same report states that 57% of tax professionals call AI their top technology investment for 2026, up from 35% two years earlier. Yet Thomson Reuters adds that only 18% of professionals say their organizations measure return on investment in AI.
One trap these organizations fall into is assuming that the return itself is the biggest time sink in the firm. Without measured data, you end up automating the most visible tasks, which is usually the return by default.
The Institute's 2026 tax report notes that many firms may find it harder to identify additional processes to automate. An AI adoption framework starts from the measured process and picks the tool after.
What tax preparation looks like, step by step
Trace one individual or business return from the first request to e-file:
- Request the documents. The client gets a list by email or through the client portal.
- Chase the missing ones. Reminder emails go out for the same forms, again and again.
- Sort the W-2s, 1099s, and K-1s. Each file is opened, renamed, and saved in the right folder.
- Key the data into the tax software. Numbers are copy-typed from PDFs, field by field.
- Compare with last year. A side-by-side check, usually in a spreadsheet, for every client.
- Settle the research questions. The same lookups come up across clients.
- Work through review notes. Notes go back and forth, and rework follows.
- E-file after the preparer signs. Nothing leaves the firm until the licensed preparer has reviewed and signed.
- Update the client. Status questions arrive by email and get answered one by one.
Taking a position and ultimately signing the return still need a preparer's judgment. Collecting, sorting, keying, and comparing mostly need time and accuracy.
The hours don't sit in one place, and the time is hard to see from any one screen. Much of it sits in waiting on and chasing documents, well before anyone opens the return.
Typical example of a document chase: A client forwards one form and promises the rest next week. Someone notes it, sends a reminder, files the new form, and checks what is still missing. Repeat that loop for every client. An agent can send the reminders, file what arrives, and flag what is still missing, so your preparers handle only the client conversations.
If you split the same workflow by tool and by step, the places an agent fits start to show.
The Process Savings Matrix for tax preparation
Tax workflow automation starts from a simple comparison: Tool usage data vs. step-level process data.
Tool usage data tells you where the hours go: email, the client portal, the tax software. Step-level process data tells you what the hours are spent on once you open each step up.
For tax preparation automation, read the last two columns as a pair: how well an agent fits a step, and what the preparer still does.
High fit means the agent can do the work while the preparer spot checks or reviews.
Medium fit means the agent drafts or flags and the preparer decides.
Low fit means the agent only pre-checks.
Some examples:
- Key source data into the tax software: high. The agent reads each source document and enters the numbers, and your preparer reviews the result.
- Research tax questions: medium. The agent drafts a summary with sources, and the preparer verifies it and decides.
- Review and sign: low. The agent pre-checks for missing items, and the preparer reviews and signs.
Whether a step should be automated comes down to how often it repeats and how much rework it causes, and only measurement shows both.
Measure the workflow before you automate it
Most firms that map a workflow bring in consultants who sit behind staff for two to three weeks, writing each step down by hand, and charging handsomely for the service. But they still miss steps, and work off assumptions on how work runs minute to minute.
You don't need a consultant when you have a treasure trove of raw data that just needs to be captured and processed.
Step-level measurement captures how the work happens across apps, step by step, across the whole team. It shows which workflows to automate first, without waiting for someone to guess.
Task mining works at the desktop level, where your team already does the work. It's a part of process intelligence: which tasks repeat, which wait, and which loop back.
For accounting firm automation, measure five things across your tax workflow:
- Time per step. The time each step takes, from the first document request to the final client update.
- App switches. How many times work changes tools between the portal, email, shared folders, and the tax software.
- Wait time for client documents. The days a return sits until the last missing form arrives.
- Review loops. How many times a return goes back and forth between preparer and reviewer.
- Rework. Work your team does twice, such as a corrected entry or a second request for the same form.
Put a price on the choice before you make it: compare what an AI step costs with the value of the work it takes over.
Start with a Work Audit
Insightful is a work data platform in the Work Intelligence category, and its Process Intelligence maps a firm's tax workflow to rank which steps are ready for an AI agent.
You see where documents wait, where review notes loop back, and where your team keys the same data twice. The same data shows your team's current AI adoption, giving you real data on which tools should be scrapped and which can be utilized better.
Run it ahead of busy season, so the first agent is ready when volume peaks. An early audit leaves time for refinement, so you can have at least one critical business process fully mapped and ready for confident automation. Set the review thresholds, and test the agent on a few returns before the rush.
Your preparers keep every position, every review, and every signature. The audit only shows where an agent could take the repeatable work off their plates.
Avoid needless consultations, and measure from how your tax preparation actually happens instead.
Start with a Work Audit. Insightful collects 14 days of data from your tax workflow, then surfaces your first opportunities.
Frequently asked questions
Will AI replace tax preparers?
No. AI agents take over repeatable prep work, such as collecting documents and keying data. Your preparers keep the judgment, the review, and the signature. What changes is where their hours go.
Which part of tax preparation should I automate first?
Start with document collection, sorting, and data entry. They repeat for every client and follow clear rules, so an agent can take them on while the preparer reviews. Confirm the order with measured data from your own busy season, because the most repetitive step isn't always the one you expect.
Can AI prepare a tax return on its own?
It can draft much of it, but a licensed preparer must review and sign. Tax return automation covers the repeatable parts, such as sorting documents and keying source data. The preparer decides the positions and signs.
How much time can AI save a tax professional?
In Thomson Reuters research, professionals estimated AI would save them an average of five hours per week. That's an estimate from a survey of professionals across many fields. Your own figure depends on which steps you hand over, and measuring first tells you which.
What is the difference between AI adoption and AI implementation in a tax firm?
AI adoption means staff use AI tools for their own tasks, often working alone. AI implementation means handing an existing process to AI, with people reviewing and approving. In a tax firm, a preparer asking an assistant for a research summary is adoption. An agent sorting and keying the intake for review is implementation.
How do I measure my tax workflow before automating it?
Measure time per step, app switches, wait time for documents, review loops, and rework. Take the numbers from real work rather than interviews, so they reflect what happens in busy season. Start with one workflow and compare its steps against each other.
Is AI in tax accounting safe for client data?
It depends on the vendor's security standards, so ask for them before any client file leaves your control. Insightful is used by more than 5,000 organizations and is SOC 2 Type II certified and HIPAA and GDPR compliant. It conforms to ISO 27001 and offers cloud and on-premise deployment.
