Business
What the work you redo every day really costs
Repetitive manual work is a real cost no balance sheet shows: it hides in the salaries of capable people employed to re-key data. Before automating it has to be measured — how often, how many minutes, how many errors — and you have to decide what to deliberately leave manual. The automations that last are the ones that also know how to fail well.
In almost every company we visit there is a capable, well-paid person who spends the first hour of the day copying data from one place to another. Nobody counts it as a cost: it is "how it has always been done". Yet that hour, multiplied by working days and by the people doing it, is the largest expense you will never find in a balance sheet.
This article is not about technology. It is about how to measure that cost, how to decide what is worth automating and — above all — what is better left manual on purpose.
The cost that appears in no balance sheet
Repetitive work has three costs, and only the first one is obvious.
- Time. The hours someone spends doing something a machine would do identically. It can be counted, and that is the easy part.
- Mistakes. Every time data is typed a second time, there is a chance it is typed wrong. A wrong invoice number does not cost the minute it takes to fix: it costs the customer's phone call, the credit note, the trust.
- The things that don't get done. The biggest and least visible cost. The person spending an hour re-keying is not talking to a customer, not checking a supplier, not thinking about how it could be done better.
There is a fourth cost, visible only to whoever manages people: repetitive work wears out your good ones. Someone with real skill, employed to copy cells, eventually goes elsewhere.
If something is redone identically every day, it isn't work: it's a process waiting to be written down.The rule we use during analysis
Before buying anything: measure
The greatest waste in automation is not buying the wrong tool: it is automating something that wasn't worth it. Avoiding that takes half a day of measurement, not a project.
- List the activities that repeat — the ones someone does more than once a day, always the same way.
- Actually time them, for three days. Don't ask for an estimate: estimates from memory are routinely wrong by a factor of two, in both directions.
- Count the mistakes of the last three months on that activity. How many corrections, credit notes, phone calls.
- Multiply minutes × frequency × people × working days. The number you get is the annual budget you are already spending without knowing it.
- Sort by cost, not by annoyance. The most hated task is almost never the most expensive one.
At this point you have a list where the first two or three items are worth more than all the others combined. Those are the ones to automate first, and they are often simpler than they look: precisely because they are repetitive, they are also predictable.
What is worth automating, and what is not
An activity is a good candidate when it has three properties: clear rules (if this happens, do that), high volume (it happens often) and low ambiguity (two different people would do it the same way). If one is missing, the automation costs more than it returns.
Some things are better left manual on purpose, and not out of laziness:
- The steps that exist to stop and look. A human check before a large payment is not an inefficiency: it is insurance.
- Decisions with too many exceptions. If the rule has fifteen special cases, automating it means writing fifteen bugs in waiting.
- Whatever changes every few months. Automating an unstable process means rebuilding the automation at every change.
- The work that makes the difference with a customer. One reply written by hand is worth more than ten perfect automatic ones.
Automations that last also know how to fail
Automating is easy while everything works. The difference between an automation that holds for years and one thrown away after six months is entirely in what happens when something goes wrong: the supplier doesn't answer, the file arrives empty, the system on the other side is under maintenance.
A well-built flow knows where it was when it fell over, knows what to undo, and knows when to stop retrying and call a person. A badly built one starts from scratch, sends the same email three times or — worse — stops silently and nobody notices until month end.
It is also why we are wary of automations built in a hurry with tools anyone can wire up in an afternoon: they work beautifully in the demo and become a problem the first day something goes off script.
What about artificial intelligence?
AI has moved the boundary of what can be automated: today you can read an invoice in a format never seen before, classify a free-text request, prepare a draft reply. Five years ago those needed a person.
The point is where you draw the line between proposing and doing. In the automations we build, while the action is public or irreversible — sending a communication, issuing a document, deleting something — the proposal goes through a human approval, and it stays on record who approved and when. The value is not "AI does everything": it is taking the preparation work off a person and leaving them the decision, with the context already laid out.
What changes when it works
The first effect is the measurable one: the hours come back. But the two effects clients tell us about after six months are different ones.
The first is that data becomes trustworthy. When a piece of information is no longer re-keyed by hand, it stops existing in three different versions, and decisions get made by looking at the numbers instead of arguing about which ones are right.
The second is that the work becomes explainable. An automated process is a written process: a new hire learns it in a day instead of three months, and when someone goes on holiday nothing stops.
If you have in mind an activity that repeats every day and you are wondering what it costs you, the answer takes half a day of measurement. That is where we start, before proposing anything.