Trying to decide on what to automate? Looking at what is repetitive and low judgment gets you close. Determining whether a mistake is noticed or the process is already consistent gets you the rest of the way.
Key takeaways
- Determining which are repetitive and low judgment tasks points you in the right direction to start, but not exactly to the right task.
- Before automating, ask yourself whether you’d notice if something went wrong, and whether the process is consistent enough to automate yet.
- These two questions apply to every future automation decision, not just the first one, and each decision that follows gets easier.
Most advice about automating a tax practice says the same two things: start with something repetitive and start with something low judgment. That’s true, and it’s already common in many organisations.
For example, if you pick a task that’s performed frequently enough that fixing it pays off within a ‘season’, not a year, that’s a smart move. And picking a task that calls for little or routine judgement, even if an experienced person has always done it, is also a good idea — for instance, simply copying figures from a source document involves practically no judgment.
Thomson Reuters’ 2026 State of Tax Professionals Report backs this up directly. More than half of respondents cite a lack of time and resources, and 47 per cent cite cost of implementation, as the top barriers to automating more of their work. If time and money are both scarce, the tasks you pick first need to earn their keep fast. That’s the easy part – but here’s what usually gets skipped.
While tackling repetitive and low-judgment tasks first may be smart, this choice only gets you into the right neighborhood – it doesn’t tell you which house to walk into. There are two more questions you should ask that matter just as much:
- Would you actually notice if something went wrong once the task was automated? and
- Is the process behind performing the task consistent enough to automate?
Would you actually notice if anything went wrong?
This first question separates a safe automation done early from a risky one, and it has nothing to do with how repetitive or judgment-light a task may be. For example, take two tasks that both look reasonable under the standard advice. One might be simply copying from a source document to another. In this task, if the tool misreads a digit, you catch it in seconds, because you’re looking directly at the source document.
The second task might be drafting a client-facing explanation of a tax position. If the tool gets a small detail wrong performing this task, nobody may notice until a client asks a follow-up question, or worse, until it’s already been sent out to all clients.
Both tasks are repetitive, and both involve modest judgment; however, the first fails quietly, while the second fails loudly.
Most firms treat each automation decision as a fresh puzzle, which is part of why the effort to automate stalls after the first attempt.
This isn’t hypothetical. Thomson Reuters Future of Professionals Report 2026 found that more than one-third of professionals say they already use AI tools that their firm hasn’t formally approved. Any of that so-called ‘shadow AI’ use that touches client work leaves a gap, in which there’s no record of what the tool was given or whether anyone checked its output. That’s exactly the kind of gap a quiet failure slips through. Pick your first task to automate one that checking it takes seconds, not a second look weeks later.
Is the process behind the task you’re automating actually consistent?
Here’s a test worth running first. Ask three people on your team, separately, how they currently manage the same task. If you get three different answers – this is not a task to automate just yet. In fact, what you’re looking at is three different tasks pretending to be one.
This problem shows up constantly in many practices and across different functions. One person formats workpapers a certain way because that’s how they were taught it. Another does it a different way because a different partner trained them. And a third improvised because nobody trained them how to do it at all. Automating any one version just locks in whichever method happened to get picked.
The State of Tax Professionals Report found that only about half of tax professionals describe their firm’s core processes as consistently standardised. This is not a criticism, in a small firm, standardisation often loses to whatever gets the work out the door fastest, especially when it’s busy. However, this also means the honest first step, before automating, is sometimes just deciding on one method of doing a task and writing it down.
Say, for example, your firm sends the same document request checklist to every new individual tax client. Right now, one person copies last year’s email and edits it by hand, another starts from an outdated template and updates it each time, and a third writes it fresh each time.
If you run it through both questions before it touches any software, you might make better automation decisions.
First, on error visibility, this task passes easily. If the automated version misses an item, you’ll see it the first time a client asks about it. Cheap to catch, cheap to fix.
On consistency, it fails, at least today. Three versions means there isn’t one process to automate yet. Pick the version you actually want, write it down, and use it by hand for a few weeks before automating it.
Use these two simple questions next time
Once the checklist example works, you’re left with a method, and the next potential automated task will take less thought to evaluate. Say the next idea is having a tool draft the cover email that goes out with a client’s finished return. Would you notice if it went wrong? Probably not right away. A slightly off tone or missing detail is easy to miss before it’s sent, and the client may say nothing even if they do notice. That’s a quieter failure than the checklist, so the fix is a mandatory review step before anything goes out, at least for the first few weeks. And that’s better than a rule you add later after something slips.
Is the process consistent? Likely not, since what goes in that email tends to vary by client and by whoever’s sending it. The fix is the same one as before: decide on the one version of that email you actually want going out, write it down, and have everyone use it by hand for a few weeks before you hand it to a tool.
And there you have it. Two potential automated tasks evaluated and resolved in the time it used to take to think through one.
Most firms treat each automation decision as a fresh puzzle, which is part of why the effort stalls after the first attempt. Asking the same two questions each time turns it into a habit instead.
