Notebook / Choosing what to automate

The cost of a repetitive task before you automate it.

An eight-minute task can take several working days each month. To decide whether automation is worthwhile, count corrections, supervision and the work your team will still do afterwards.

By Kenea · Published and reviewed: 9 September 2026.

Define the work you are counting

Choose a specific unit: preparing an order for review, registering a request or compiling a weekly report. Mark its beginning and end. If one person stops measuring when the data is copied and another stops when errors are resolved, their figures are not comparable.

During a representative week, record volume, active handling time, corrections and waiting. Separate effort from elapsed time: an approval waiting for two days does not necessarily consume two working days. Extend the observation if month-end work or seasonal peaks change the pattern.

GOV.UK's public guidance calculates cost per transaction from the cost of providing a service and the number of completed transactions. It is a useful reference for defining the measurement boundary; its public-sector accounting rules are not an obligation for your business. Source: GOV.UK, cost per transaction.

Build a calculation someone else can check

The arithmetic difference would be 21 hours a month. At an assumed internal cost of €30 an hour, that represents €630 of staff time. It is not a cash saving: the team may use those hours to handle a backlog, while some costs remain unchanged.

It is not a complete profitability calculation either. Implementation, training, licences, service usage and maintenance are still missing. Two minutes of review is an assumption for the pilot to test. At four minutes, the proposed workload rises to 23 hours and the difference falls to 13.

Compare opportunities as well as hours

A higher-volume task may depend on an application you cannot access with permission. A smaller task may have clear rules and a trial that is easy to reverse. Compare candidates for the first pilot using four questions:

• Do we have a baseline and someone who understands the exceptions?
• Can we access the necessary data with permission?
• Is there a correct output we can verify?
• Can we limit the change and return to the previous workflow?

A high score elsewhere does not remove a blocker. Resolve missing permissions first. If the process changes every week, stabilising it may be the better first step. The guide to choosing a first automation explores these conditions further.

Turn the estimate into a test

Trial one type of request and compare an equivalent sample with the current method. Count preparation, review, correction and follow-up. Record incomplete outcomes and work passed to other people: leaving one inbox does not mean the task is finished.

Before starting, write down what would justify continuing and what would make you stop. A lost request or unexplained duplicate, for example, would require a design review. Time and quality thresholds should be agreed for that process, its volume and the consequences of failure.

The final decision could be to automate, improve a template or keep the current procedure. A useful estimate makes its assumptions visible and helps you choose a trial that can challenge them.

Start with the free visual process assessment for initial guidance without registration. Reviewing actual data, dependencies and scope is a separate professional assessment with Kenea.

Explore my process

Source and editorial approach

Source consulted on 9 September 2026: GOV.UK, Measuring cost per transaction. The example, comparison and questions are Kenea's editorial proposals, not commercial results or a promised return.