When a company wants to start using AI, the natural question is often: How can we automate this? The better first question is: Should this be done at all?
01
The report almost nobody read
Imagine a company where five employees submit figures to a colleague every week. The colleague compiles them in a spreadsheet, writes comments and sends a report to three managers. The workflow takes hours and has existed for years.
The obvious AI project is to automate the collection, write the comments and produce the report faster. Yet when the managers are asked how they use it, they reveal that they only look for two specific exceptions.
The best automation was not a faster report. It was removing the report and sending an alert when an exception actually occurred.
02
Old solutions outlive their problems
Processes begin for a reason. A previous system could not share data. A manager wanted more control. An error created another approval step. Technology and organisations later change, but the workflow remains.
If it is automated without being challenged, yesterday’s compromise becomes permanent.
03
Begin with the decision
Ask which decision the result is supposed to enable. Who uses it? What will that person do differently afterwards? If nobody can answer precisely, investigate the process before building anything.
Often the need can be solved closer to the source. A manager may not need a weekly report, but an alert when a sales figure becomes unusual. A customer may not need three internal forwards, but a direct answer from the person or agent holding the information.
04
AI should reduce work — not conceal it
AI can make heavy workflows impressively fast. That is useful when the work is necessary. But speed can also conceal that the organisation produces information nobody uses or moves data through more stages than required.
The good outcome is fewer unnecessary actions, a shorter path from information to decision and more time for work that requires judgement.