When someone asks me what the most common mistake banks make with AI is, the answer tends to surprise them. It is not bad models, missing data, or slow IT. The mistake is far more prosaic, and it happens long before the first line of code gets written.

In this respect AI is no different from any other technology. The moment you buy a tool just to have it, you walk straight into the hammer and nail problem. When you are holding a hammer, everything starts to look like a nail. You go looking for places to put AI, instead of solving a real need. And that is exactly where projects die.

What gets overlooked in all of this is something simple: inside every bank, and frankly every large company, there is fierce competition for the resources to actually deliver.

Initiatives win, not ideas

There are always fewer people, less budget, and less capacity than there are ideas for using them. So the initiatives that win are the ones promising the biggest business impact, in the shortest time, with the highest certainty. And those three axes trade against each other: the lower the certainty, the higher the promised benefit has to be. Not just to get support in the first place, but to keep it over time.

Here is the trap with AI. On the surface it looks like everything will be fast. A demo in an afternoon, a prototype in a week. But that apparent speed collides with the reality of the organisation. And the bigger the company, the harder that collision hits.

A big company has a lot to lose, so it becomes risk averse. Out of that grow the committees, the multiple environments, the security units, the processes, and the gates you have to pass before anything ships. They are not obstacles born of bad intent. They are the bank's immune system, protecting what it already has. They just run slowly.

There is no such thing as low hanging fruit. The small pilot you will just try, ship, and love is the one most likely to die on the way.

A year to production, and a defence every quarter

The consequence is uncomfortable. Even when a use case looks technically trivial, getting it into production realistically takes a year or more. And across that whole time you have to win support again and again, quarter after quarter. Budget and capacity get reallocated, priorities shift, people leave. A project without a hard anchor in the business becomes a write-off candidate in every one of those rounds.

If what you are solving is a nice to have, you are betting that the initial excitement holds long enough. It usually does not. The extra budget and capacity never get freed up, the project quietly stalls, and a year later nobody remembers why it started. Nobody officially kills it. It just slowly fades out.

A few things I recommend walking through before deploying AI in a bank:

So my advice is simple: pick projects that genuinely have impact. Solve what contributes directly to the bank's business metrics. Anything else gets dropped the moment business pressure hits. And it always does.

If you are weighing up which AI use case to start with in your bank so that it survives the road to production, get in touch. I am happy to look at where your case has a hard anchor and where it has only the initial excitement.