Where AI projects go wrong
Most AI projects in operations fail for the same three reasons. Almost none of them are technical.
- Topic
- AI Enablement
- Kind
- From the archive
- Reading time
- 2 minutes
- Written by
- Telemyr
01 / 05
Most brands have tried something with AI by now. A pilot, a tool, an agency, a workshop.
Very few of them can point at a number that changed. The reasons are consistent and we see the same three most of the time.
02 / 05
It gets run as a technology project
IT owns it, picks the platform, and runs the pilot. The operations team is consulted at the start and shown the result at the end.
What comes out is technically sound and does not fit how the work is done. The planner still opens the spreadsheet. The controller still builds the file by hand. Nobody says anything because the project was never really theirs.
The people who own the process have to own the project. IT should be in it for the architecture and the security, and the decisions about what gets built belong to operations.
03 / 05
It starts too big
The instinct is to fix the biggest problem first. That usually means a twelve month programme, a large budget, and a steering committee.
Twelve months is long enough for the sponsor to change, the priorities to move, and the enthusiasm to run out. Most of these projects are quietly wound down before anything reaches a user.
Start with something that pays back in weeks. One report that a person builds by hand. One decision that is being made without the right numbers. It builds the foundation you need anyway, and it buys the credibility to do the bigger thing next.
04 / 05
Nobody's job changes
The tool gets built, launched, and demonstrated. Then the old process keeps running next to it, because nobody was told to stop doing the old thing.
Six months later two versions of the truth exist and people trust the one they have always used. The tool gets blamed. The tool was fine.
Adoption is a management decision. Someone has to remove the old report, change what happens in the meeting, and be clear about who owns the new number. That part is uncomfortable and it is the part that decides whether any of it was worth doing.
None of these three are about models or infrastructure. They are about who owns the work, how big the first step is, and whether anything actually changes when it lands.
05 / 05