Most AI commentary looks at the front office.
I’m more interested in the other side: operations, controls, exceptions, legacy systems, and the back-office workflows where the promise of AI meets the operating model.
The interesting AI question in finance is not whether agents can do the task.
It is whether the institution can explain, control, evidence, approve, monitor, and recover the task after the agent gets involved.
That is where the real operating model starts.
The useful version of AI in operations is not skipping the process.
Conversational workflow, duplicate checks, approval
Less form-filling. Fewer duplicate tickets. Better signal.
Team presented an excellent use case today, Claude connected to Confluence can write, review and keep all our procedures updated in a consistent format.
Simple things, but documentation has been low hanging fruit for auditors for years.
Some days I lose hope that my industry will ever really evolve and embrace the future that obviously seems to be barrelling towards us.
Am I being pessimistic or optimistic!?
Hit my weekly codex limit 2 days ahead of the reset this week for the first time. This is week 4 of the plus plan for me.
Feels like my WiFi has been cut off.
Claude is overkill for many corporate use cases. It’s also hard to put the guardrails in with it.
We recently tested a service ticket to scenario and not only did it find the relevant API without us asking, it then created the database insert and asked if it should approve it!
Using the dictation function on AI in the office is a cheat code. Just pretend you’re on a zoom call and you don’t feel like an idiot talking to yourself.
In what has been a traditionally slow moving in the control focused banking sector AI unlocks unprecedented speed in developing ideas.
My job is now to capture and document as many as I can so that they can be realised.
Talk to the Tech guys, they are the key.
Seeing lots of great desktop AI experiments this week.
Things like customer RORC analysis, RFI answer reuse, and margin-fail notification automation. Its great motivation for teams to get quick wins. What’s proving hard is turning them into controlled BAU production workflows.