Published on 05/09/2026
Which office task should you automate first?
Monday morning brings three busy inboxes, receipts on a desk and a list of calls to return. In many small businesses, ideas for automation are plentiful. What is missing is the time to choose the right starting point. Trying to rebuild quoting, invoicing, filing and scheduling at once usually creates one more unfinished project.
Choose exactly one workflow instead. The best first candidate is rarely the most annoying isolated task. It is frequent, easy to describe, has known exceptions and produces an outcome that somebody can check.
Four questions for your shortlist
Collect three to five recurring tasks. Avoid broad labels such as "bookkeeping". Write down observable workflows, such as "file PDF attachments from the invoice inbox" or "transfer accepted quote details to the job list".
Score each workflow from 1 to 5:
| Criterion | 1 point | 3 points | 5 points |
|---|---|---|---|
| Frequency | monthly or less | several times a week | several times a day |
| Time spent | under 15 minutes a week | 1 to 2 hours a week | more than 4 hours a week |
| Cost of current mistakes | minor rework, quick to correct | recurring questions or rework | today's mistakes disrupt billing, delivery or customer contact |
| Process stability | changes often and is undocumented | core path is clear, with some exceptions | clear inputs, rules and owners |
Do not follow the total blindly. The mistake row measures the cost of today's manual errors, not permission for risky automated actions. A candidate is eligible only when a wrong action is easy to reverse or a person must approve every consequential action. This matrix starts a useful discussion; it does not measure return on investment. A workflow may score highly but still be unsuitable if nobody can define the correct result. A smaller, stable workflow can be a better pilot because your team can learn safely.
An explicitly hypothetical example
Consider a fictional trades business. Its owner reviews three activities:
| Candidate | Frequency | Time | Cost of current mistakes | Stability | Total |
|---|---|---|---|---|---|
| File incoming email receipts | 5 | 4 | 3 | 5 | 17 |
| Write individual quotes | 3 | 4 | 5 | 1 | 13 |
| Coordinate holiday requests | 2 | 2 | 4 | 2 | 10 |
Filing does not win because 17 points prove a financial return. It wins because the trigger, file type, destination and review are understandable. Individual quotes need professional judgment and vary by customer. A template might help later, but full automation would be a risky first move.
Observe the real desk work
Walk through a typical case with the person who does the job today. Ask:
- What tells you that the work should start?
- Which information must be present?
- Which decision follows a fixed rule?
- Which exception appears regularly?
- Where does the finished result go?
- Who notices an error today?
Record unofficial shortcuts too. If an employee calls a supplier when a document is unreadable, that call is part of the workflow. Automation designed only for ideal inputs pushes work into a hidden error pile.
Define a small result you can verify
Set a boundary before building. For example: "PDF attachments from one shared inbox are matched to a known supplier, given a consistent name and proposed for approval. Unclear or duplicate documents remain in a review list."
That is more useful than "automate receipts". It defines the accepted input, automatic action and human decision. Establish a short baseline: count suitable cases, handling time and resulting rework. Track the same figures after the pilot. Only that comparison can show whether the change frees capacity.
Freed time is not automatically a cash saving. If an employee spends two fewer hours typing and uses those hours to answer enquiries, the business gains capacity, but payroll does not immediately fall. Keeping that distinction clear prevents inflated promises.
Bound the pilot before building
Test the selected candidate with ten to twenty real cases from a manageable period. Do not choose only clean standard examples. Include a missing reference, an unusual file format and a genuine duplicate. For each case, write down the correct outcome and who can confirm it.
Then set the test rules: what counts as a correct filing proposal, which exception must reliably enter the review list and how will you detect duplicate processing? This makes the bounded pilot testable without rebuilding the entire office workflow.
Name an owner and a deputy as well. Both must be able to find the review list, correct a case and see whether the workflow is operating. A pilot without visible ownership quickly returns to manual work after its first exception.
Warning signs for a poor first candidate
Do not begin with a workflow whose inputs constantly change, whose ownership is unclear or whose correct outcome is disputed. Be cautious when a wrong action is difficult to reverse. Automatically sending, ordering or posting requires stronger approval controls than proposing a filing location.
Sometimes the honest result is: standardize first, automate second. A shared template, one shared inbox or a required field may help more than new software. After a few weeks of consistent use, score the workflow again.
Your first pilot should demonstrate what sound automation looks like: narrow scope, visible exceptions, a named owner and a measured comparison. To assess your candidates and select one dependable starting workflow, begin with the DevSeal process check.