A process is reorganised before it is automated
24 October 2026, 09:39
Automating a process means having software carry out the steps that people perform today. The program carries them out as it finds them: if the path is full of pointless steps, it completes them faster, and the waiting between one and the next remains. That was the reasoning American Express gave on 23 October, when it presented its third-quarter results, to explain cutting about 5 per cent of its jobs worldwide: first layers of management are removed, then artificial intelligence is brought in. The chief financial officer, Christophe Le Caillec, put it this way to CNBC: «You don't want to automate processes that are not optimally structured». In other words: badly organised processes are not automated.
A company divided by functions measures activities, not the result
Many SMEs are organised by functions: sales, technical office, purchasing, production, shipping, administration. Each function has a manager and is judged on what it does: sales on the orders won, purchasing on the prices obtained from suppliers, production on the pieces finished. None of these measures says whether the goods reach the customer on the promised day and whether the invoice is paid.
A process is the sequence of steps that leads to a result for the customer, for example from order to delivery and to payment, and it cuts across every function. When nobody is accountable for all of it, time is lost between one department and the next. The order sits waiting for the technical office to approve the drawing, then for purchasing to buy the material, and finally for production to find it a slot in the machine schedule. Each function can do its own part well, and the delivery can still arrive late.
Organising the company by processes means giving one person responsibility for the result from start to finish, together with the information needed to decide: the salesperson knows what margin the customer leaves, the department head knows what the output earns. The chief executive of American Express, Stephen Squeri, said that removing layers of management serves to make decisions faster: with fewer middle managers, whoever has to choose waits for fewer approvals.
Software makes a disorganised process faster, not better
New management software or an artificial intelligence tool shortens the work inside each step: the program fills in a form with the data it already holds, and nobody copies it out by hand any more. On their own they do not shorten the wait between one step and the next, which depends above all on whoever has to decide. If an order passes through six offices and in three of them someone copies data or adds a signature, the program does those same things faster, and the three steps remain. After the purchase, moreover, those operations are written into the program: removing them means paying for another change.
The typical mistake is buying the tool to fix a delay before knowing at which step it arises. In many SMEs the problem is not missing technology but technology that is badly managed: different programs that do not exchange data, and people passing it by hand from one to the other.
The path of an order is reconstructed from the dates of each step
The orders of recent months are taken, and for each one the dates of the steps are recorded: receipt of the order, confirmation to the customer, technical approval, purchase of the material, delivery by the supplier, start and end of production, shipping, invoice, payment. With the planned date beside each entry, that list is the phase schedule. It gives two answers: how many orders left after the date promised to the customer, and at which step the late ones stood still the longest.
An example. A company that builds electrical control panels for industry has a turnover of €21 million a year. Its phase schedule shows that on average 30 days pass between the arrival of an order and shipping. Work is done on the panel in 8 of them; for the other 22 it stands still between one department and the next, waiting for a decision or an approval. A program that shortens the work inside each step does not touch those 22 days.
The dates show at which phase orders stand still the longest, and that is where to act before choosing any tool: deciding who takes the decision that holds them up, with what information, and who is accountable for the order until payment. The same dates serve controlling to judge a process on figures rather than impressions. Only then is it decided what to automate.
The question to ask before any purchase remains just one: in the company, who is accountable for an order from the day it arrives to the day the customer pays for it?
This article was prompted by a news item published in Bloomberg on 23 October 2026.