Somewhere in most businesses there is a person with a qualification on the wall and a spreadsheet open on one screen and a finance system on the other. They read a figure on the left. They type it on the right. They check it. They move to the next line.
They were hired because they understand how a business's money works: what a margin should look like, when a supplier's terms are unusual, why a customer's account is drifting. For a large part of the day, though, the work they actually do is transfer. Their knowledge is used mainly to notice when the transfer went wrong.
That is the moment this piece is about. Not the efficiency question, which Mellorca has covered elsewhere, but what it means for a skilled person when the job they do and the job they were hired for drift apart.
What the evidence says about this kind of work
The ILO's research on generative AI and jobs looked at which tasks within occupations are most exposed to automation. It found clerical work to be the most exposed group, with nearly a quarter of clerical tasks considered highly exposed and more than half at a medium level. For most other occupations, including professionals and managers, exposure was much lower.
The ILO's broader conclusion matters as much as the figures: most jobs are only partly exposed and are more likely to be complemented than replaced. The organisation also noted that the bigger issue may be job quality, including work intensity and autonomy, rather than job loss. That is a useful frame. The question is not only which tasks a machine could do, but what happens to the person's work once those tasks are handled differently.
Microsoft's Work Trend Index adds a view from workers themselves. In its survey, 64% of people said they struggled to have enough time and energy to do their job, and 76% said they would be comfortable using AI for administrative tasks. Those are global survey results from one vendor's research, not South African findings, but they suggest that many people would willingly give up routine administrative work if it were handled reliably.
How a skilled role becomes a transfer role
Consider an illustrative case: a Johannesburg property management company with about thirty staff. Its senior bookkeeper is a qualified accounting technician. Each month she downloads bank statements, matches deposits to tenant accounts in a spreadsheet, re-keys the results into the accounting package, and then re-keys supplier invoices that arrive as email attachments.
None of these steps was designed to be her job. Each arrived as a small workaround: the bank export did not match the accounting system's format, the property system and the accounting system were never connected, and suppliers sent invoices in whatever format they liked. Over time, the workarounds added up to most of her month. This is the pattern described in when copy-and-paste becomes part of the job description, and the cost side is set out in high-value employees doing low-value admin.
The root causes are usually ordinary:
- systems that do not exchange data, so a person becomes the connection between them;
- inputs that arrive in inconsistent formats, so they must be read and retyped;
- no agreed system of record, so the same figure is maintained in more than one place;
- a sense that the work is too small to fix, even though it repeats every day.
There is also a quieter cause. Because the bookkeeper is competent, the workarounds work. Errors are caught, the month closes, and nothing looks broken from the outside. A less skilled person might have struggled visibly, prompting someone to fix the underlying problem. Her skill, in other words, is part of what keeps the transfer work invisible to the people who could remove it. The better she is at absorbing the gap between systems, the less likely it is that anyone will close it.
The human tax: time and dignity
Time is the easy part to describe. Hours go on transcription that could go on analysis, on talking to tenants about arrears, or on spotting the account that is about to become a problem.
Dignity is harder to talk about and easier to overstate, so it is worth being careful. Some people find routine, accurate work satisfying, and there is nothing undignified about it in itself. The concern is narrower: when a skilled person's capability is consistently unused because the system requires them to act as a conduit, the role stops recognising what they can do. Their judgement becomes a quality-control step on their own typing. For some people, that can feel like being reduced. For others it may be tolerable or even comfortable. Mellorca cannot know which applies to any individual, and should not claim to.
What can be said is that the business is paying for judgement and receiving mostly transcription, and that the person has little say over the mismatch.
What could change
The intervention does not have to be ambitious to matter.
- Use the formats systems already offer. Bank feeds, import templates and supplier portals can remove much of the retyping without any custom development.
- Decide where each figure lives. A single system of record for tenant balances, for example, means it is maintained once.
- Connect the two systems that matter most. Often one integration, between the operational system and the accounting system, removes the largest share of transfer.
- Keep people on the exceptions. Automated matching handles the routine lines; the qualified person handles the ones that do not match, which is where their knowledge is actually needed.
That last point is the important one. As Mellorca has argued in how to reduce human error without removing human judgement, the aim is to move people from moving data to making decisions, not to remove them from the process.
Trade-offs
There are real risks in getting this wrong.
- Loss of feel for the numbers. Manual entry, tedious as it is, gives people direct contact with every transaction. When it goes, that familiarity has to be rebuilt through review and reporting, or it is lost.
- Exceptions can grow. If automation only handles clean cases, the person may be left with a smaller but harder and more stressful queue.
- Fear for the role. When transfer work is a large part of the job, removing it can feel like a threat rather than a release. That concern deserves an honest conversation about what the role becomes.
- New fragility. Automated feeds can fail silently. Someone must own and watch them.
Why it matters
The business outcome is straightforward: fewer errors, faster closes, and more capacity without more headcount. The human outcome is different and, arguably, more important to the person involved.
When systems carry the transfer, the qualified bookkeeper in the illustrative case can spend her month on the work that uses what she knows. She can look at why a tenant's arrears are rising instead of retyping the figure that shows it. The dividend is time, but it is also a closer match between who someone is at work and what they are allowed to do there. That match is easy to lose without anyone deciding to lose it. It is worth deciding to get it back.
Sources and further reading
- Generative AI likely to augment rather than destroy jobs (last verified 10 October 2026)
- https://www.microsoft.com/en-us/worklab/work-trend-index/will-ai-fix-work (last verified 10 October 2026)
Related Mellorca reading
- The Inbox That Follows You Home
- When Every Day Starts by Rebuilding Yesterday's Context
- Why Waiting for Information Feels Like Working