Almost everything called AI transformation is in fact a new layer over an old process. The process stays exactly as it was, with its nine steps, four approvals and three spreadsheets, and an assistant is added that drafts one of those steps faster. You gain something. You gain very little.
The steps that exist only out of habit
Every process older than five years accumulates steps that no longer serve any purpose. A check added after an incident that happened once in 2017. A report somebody once asked for and nobody has read since. A second signature that exists because the old system left no trace of the first.
Automate that and you enshrine it. Now the useless step is faster, cheaper and much harder to remove, because it has a tool behind it and somebody defending it.
Start from the end
Redesigning is something else. You start from the result you want — a quote sent, a case closed, an order shipped — and ask how you would get there if you started today, knowing what technology now allows. Not how you would improve the current path, but what path you would build from nothing.
The question that works best for us in a room is a simple one: if the machine could do seventy per cent of this, what would the process look like? It forces people to talk about the work rather than the tools. And two or three steps almost always simply disappear.
How it is done in practice
- Fix the result and the moment it counts as finished. No ambiguity.
- Break the path into small, concrete steps without inheriting the current order.
- For each step decide who does it: person, machine or both.
- Mark the handovers. That is where time is lost and where errors get in.
- Define what gets recorded at each step, because that data is what will make the process improve on its own over time.
- Define what happens when something fails, and who picks it up.
An ordinary example
Quoting in an industrial company. The original process: sales receives the request, asks engineering for technical data, engineering answers in three days, sales builds the price in a spreadsheet, a manager approves it, somebody copies it into a template and it goes out. Eleven days on average, half of them waiting.
Redesigned: the system pulls the technical data out of the request, compares it against the history of similar jobs, proposes a configuration and a starting price, engineering only steps in when the part is genuinely new, and the manager approves only above a certain value or below a certain margin. Sales reviews and sends.
The interesting part is not the tool. It is that two waits and one approval disappeared, and that engineering went from answering forty queries a month to answering six — the ones that actually needed their judgement.
If your redesigned process has the same steps as before, you have not redesigned anything.
The cost of doing it properly
Redesigning costs more conversation up front and considerably less money afterwards. A serious redesign of one critical process takes four to eight weeks, pilot with real users included. That is less than the time it takes for the invoice to arrive for a platform nobody ends up using.