A finance team is told that the new system will automatically approve and pay supplier invoices that match their purchase orders. It works. Payments go out on time, month-end is quieter and nobody complains. A few months later a long-standing bookkeeper mentions, almost in passing, that she misses deciding which suppliers to pay first when cash is tight. The system pays in the order the invoices arrive. She used to choose, and she chose well.
This piece looks at the choices that automation can quietly remove, why people notice their loss later than the time saved, and how to design automation so that it takes away work people wanted to be rid of, without taking away decisions they wanted to keep. It is a hypothesis drawn from general research on autonomy and automation, not a finding about any specific South African employer.
The difference between a chore and a choice
Most automation projects begin by listing tasks and asking which ones can be done by a system. That is a sensible start, but it tends to treat every step in a process as the same kind of thing. In practice a step can be one of two very different things:
- A chore: repetitive, rule-based work that nobody would miss, such as keying in an invoice number.
- A choice: a moment where a person applies judgement, such as deciding which supplier to hold back, which customer to give a little extra time, or how to word a difficult reminder.
From the outside, both look like tasks with inputs and outputs. From the inside, they feel quite different. A chore removed feels like relief. A choice removed feels like being shrunk, even if nobody says so.
Self-determination theory, developed by Edward Deci and Richard Ryan, proposes that autonomy, the sense of acting from one's own choice, is one of the basic psychological needs that support motivation and wellbeing. The theory does not say that every decision must be kept. It does suggest that stripping out the decisions people care about can reduce their engagement, even when the outcomes are acceptable.
What the research on automation adds
There is a second line of research that points in the same direction. Studies of what is called automation bias describe the tendency of people to over-rely on automated suggestions, accepting them even when other evidence points the other way. When a system makes the decision and the person is left to monitor, attention drifts. People become supervisors of a process they no longer understand well enough to challenge.
That has a human cost and a business cost. The person loses the use of a skill. The business loses a second pair of eyes, because the person who would have noticed that paying the electricity supplier last was unwise is no longer in the loop.
The UK Health and Safety Executive's management standards for work-related stress treat control as a core condition of a healthy workplace: how much say people have in how they do their work. Automation that is designed without asking people what they want to keep can reduce that control, even as it reduces effort.
The human tax: autonomy
The cost is autonomy, and it is easy to miss because it rarely shows up in a time-and-motion study. The hours saved are measurable. The loss of discretion is not. People tend to mention it indirectly, in comments such as "the system just does it now" or "I don't really see what happens any more".
The article on when a tool rollout takes choices away describes the same pattern from the perspective of a new tool being introduced. Another related piece, when the system decides what you can do today, looks at what happens when the sequence of work is determined by software rather than by the person doing it.
An illustrative case
Consider an illustrative example, not a real case: a Pretoria property management company with a six-person accounts team. The company introduced automated rental reminders and arrears letters. Previously, a credit controller named in this example as Thandi read each arrears account, phoned the tenants she knew to be going through a hard time, and delayed letters for those with a record of paying late but in full.
The automated system sent the same reminders on the same days to all tenants. Collection times improved slightly. Thandi, who had built up goodwill and a reputation for fair dealing, began to receive angry calls from tenants who had received a formal letter a day after a death in the family. Her manager saw a better collection rate. She saw a fair process replaced by a blunt one and had no way to change it.
Nothing in the system was faulty. What was missing was a decision about which choices should stay with a person.
Designing automation that leaves room for choice
Asking people what they want to keep is the most reliable test. A short conversation before any project starts will usually surface it. Beyond that, a few design habits help:
- Separate the rule from the exception. Automate the standard case and route unusual ones to a person. A tenant with a hardship note, a supplier with a dispute, or a customer flagged by the account manager goes to a human queue.
- Let people override, and record why. An override button that needs a one-line reason respects judgement and gives the business a record. If overrides are frequent, the rule probably needs revising.
- Show what the system did, and why. People who can see the logic can challenge it. The guidance on how to design automations that people can understand covers this in detail.
- Keep a human at the points that matter. Designing human approval checkpoints for AI workflows explains where review is worth the delay. The broader principle applies to simpler automation as well.
- Review after go-live. Ask the people who used to do the work what has been lost, not only what has been saved. How to reduce human error without removing human judgment offers a useful frame for that review.
The business outcome
Automation that preserves meaningful choices is more likely to be used as intended, and less likely to be worked around. People who feel their judgement is respected tend to flag problems earlier, such as a rule that is producing odd results, because they still feel responsible for the outcome. In a country where customer goodwill matters and many relationships are long-standing, the ability to handle the exception sensibly is often worth more than the marginal efficiency of handling every case identically.
There is also a protection against the quiet failure of a system nobody is watching. If a person still makes the decisions that matter, they are more likely to notice when the results look wrong.
The human dividend
The dividend is autonomy: keeping the parts of the work that call on judgement and relationships, while being relieved of the parts that do not. For a credit controller, that might mean spending her day on the ten accounts that need a phone call and none on the two hundred that do not. It will not suit every role, and some people will be glad to see a decision go. What matters is that the choice about what to keep is made with them, not for them.
The trade-offs
- Less consistency. Where people retain discretion, outcomes will vary. Some customers will be treated more leniently than others, and that needs to be fair and explainable.
- Slower processing. Exception queues and overrides take longer than straight-through processing.
- More design effort. Working out which steps are chores and which are choices needs time with the people doing the work.
- Judgement can be wrong. People make poor decisions as well as good ones, and automation sometimes does better.
- Some choices will have to go. Where a decision carries legal or financial risk, removing individual discretion may be the right outcome, as long as it is done openly.
Why it matters
Automation is usually sold on the hours it saves. Its quieter effect is on which decisions remain with people. Businesses that decide that deliberately, and ask the people involved, tend to keep both the efficiency and the judgement that made the work good in the first place.
Sources and further reading
- Self-determination theory - Wikipedia (last verified 10 October 2026)
- Automation bias - Wikipedia (last verified 10 October 2026)
- Management Standards - Control - HSE (last verified 10 October 2026)
- Human-in-the-loop - Wikipedia (last verified 10 October 2026)
Related Mellorca reading
- The Freedom That Comes From Knowing Where Things Stand
- What Changes When People Can See Their Own Numbers
- Why Approval Chains Make Capable People Wait