My current consulting gig is at a government "superdepartment" that was formed by mashing together a few dozen small agencies that were previously already combined into large agencies.

Their IT was as diverse as it gets, and then they were told to move into one building and "make it work".

It's as messy and as broken as you imagine it is.

I wonder if maybe AI could help with such an ungrateful job.

Yes because one division is sending the data in XML and a 3 letter county code and values in thousands of monetary unit and another system does things in a completely different way

No, because the field that in one system is called "cash" is called "settled" in another; system A uses BCD for money, system B uses 32 bit unsigned ints reckoned in millicents, and system C uses a cash object which can define original currency, calculated currency value, guaranteed conversion preference, contract reference number, and precision.

Understanding semantics is required, not just form.

Yes, you got it. But I still think AI can at least make it easier

If you mean deep learning type systems, you need to train them on vast training sets of similar data so they can learn very general patterns, similarities and connections. Like giving them 10,000 pictures of dogs, so they can learn to recognise pictures of dogs.

These banking systems are all massively complex one off systems, often with very different sets of functions or even definitions of the problem, and the implementation details are different in almost every way possible. They’re the exact opposite of the sorts of problems modern trained AI systems are good at dealing with.

well said

If you can write one in COBOL and it can finish processing the whole set of transactions in less than 24h you may have a chance. /s

IT guys are paid a lot not to be lazy.

Changing data types in an interface is a boring job (didnt Office Space joke at that?) but doing this via some broken AI will probably introduce even more spaghetti.

Reality is that those systems should be killed one by one and finally when there is one system, it should be upgraded to something better. What is very difficult.

20mil for the new system, 80mil + 5 years of implementing for the translation layers to communicate with the legacy systems.

> Reality is that those systems should be killed one by one and finally when there is one system

It take so long to replace each of the systems, by the time you get to the "one true solution" 40 years have passed and you are back where you started.

>I wonder if maybe AI could help with such an ungrateful job.

It's not necessarily ungrateful. It can even be interesting, but it all depends how everyone approaches it and the most important thing ever is: realistic timeliness and outcomes.

Yeah but it's hard not to feel it's ungrateful when you see the 5th "creative" way of doing something which is pretty much like running with a spare tire and thinking it's all good

> I have just seen that she has passed away last year at 35 years old from cancer.

You’ve misread the article, she was diagnosed with cancer on 2003 aged 35, and died this year aged 55.

You're absolutely right! I'll fix that up so! Thanks

One major acquisition that is missing from your list was the aquisition of Postbank in 2018.

This is part of the problem with banks with acquisitions and why they end up being so unwieldy. In the old days we used to create gateways to each by "joining" them in some short-term strategy to merge in the long-run. But as we know nothing in the long-run generally happens due to changing priorities and differing goals of teams.

Norisbank was bought 2006/2012.

https://en.m.wikipedia.org/wiki/Norisbank