What to do with the bots you already own
If you have an estate, this is the section you came for. Sort every bot into one of three fates, once a year, and do it on evidence rather than on how anybody feels about the technology.
Keep
It runs, it rarely breaks, the process it automates is stable, and the input is structured, so leave it alone. Rebuilding a working bot as an agent because agents are new is the most expensive mistake available this year, and it produces a less predictable version of something that already worked.
Most of a healthy estate belongs here.
Rebuild
The bot breaks often, or it handles only the easy half and dumps the rest in a queue, or somebody rewrites it every time a vendor changes a screen.
Before rebuilding, ask why it breaks. If it breaks because the underlying system has an API that nobody used, the fix is an integration, not an agent — and it will be cheaper and more reliable than either. Screen scraping was always a workaround for a missing interface, so if the interface now exists, take it.
If it breaks because the input is genuinely messy, that is the real rebuild case, and an agent at the front with the existing bot at the back is usually the shortest route. You keep the part that works.
Retire
The process changed, the system it drove has been replaced, or nobody has looked at its output in a year. Retire it, and check first whether anything downstream quietly depends on it, because there usually is, and it is usually a spreadsheet.
Every estate has retired bots still running. They cost licence, they cost maintenance attention, and they make the estate look busier than it is.
The maintenance number that decides it
One number sorts most estates: hours spent maintaining the bot in the last twelve months, against hours it saved in the same period.
You already have both: maintenance hours are in your tickets, and saved hours are in the business case somebody wrote to get the bot approved. Put them side by side for every bot and the keep, rebuild and retire piles form themselves, without anybody having to argue about whether agents are the future.
If that comparison has never been made, make it before you do anything else this year. It is a week of work and it will change your roadmap more than any trend article, including this one.
When the rebuild pile does involve changing platform, our RPA tools comparison sets out how the main options differ.
