By Leda GlyptisAuthor, seasoned tech executive and banker in recovery
It’s been four years since ChatGPT came into our lives.
Eleven years since Open AI (who brought us Chat GPT) was incorporated.
70 years since the official academic discipline of AI came into being (though to say that the term’s meaning and resonance has changed in that time, would be the ultimate understatement) and 71 years (and one month, to be precise) since the term Artificial Intelligence first appeared, in a proposal submitted by a group of computer scientists. For a summer school programme, of all things.
The dates don’t matter other than to highlight that the thing that consumes everyone’s thoughts and budget discussions right now is neither new nor a sudden apparition. It evolved painstakingly over many decades and its gestation is nowhere near finished. That means there’s more to come both at the frontier of innovation, cyber-readiness and capability expansion. But also there’s more to come in the much less exciting and yet as vital deployment space. There’s more to come because how we use things is as important as what they are capable of.
This is our cue, by the way. And we keep missing it.
We have a part to play here. Not a glamorous or boundary-busting part. But a part nonetheless.
And it bears remembering this as the industry is caught in soul searching around the fact that our own AI engagements are into their third year with little to show for ourselves other than burgeoning bills, messy estates and a heady sense of possibility. Because, although most organisations have little to show by way of transformational outcomes, individuals feel empowered. Nobody is vibe-coding their way to greatness and yet the workforce feels better about work than we have done in decades. And that’s not nothing.
But it will amount to nothing if that is all we have to show for the collective billions that have been spent so far.
By most industry benchmarking calculations we have about a year to start showing traction for AI initiatives (not least because many organisations have pre-empted the savings and have started ‘right-sizing’ their teams in anticipation of efficiencies that are yet to be identified, let alone realised). A year. In which to show traction, in which to embed a brand new way of working and understanding work and showing measurable benefits from it all.
What can be achieved in a year?
In human life span terms, it’s time enough.
In corporate terms it’s the blink of an eye.
So. What is to be done?
I have two words for you: ISO 20022.
And you will ask me, why that?
And I will answer… why not?
It is something we have made an industry-wide commitment to and acquired an industry-wide challenge triggered by the global migration to said ISO 20022 standard. Because we can’t do it.
And the reason for that?
Addresses.
I know. Not the villain you expected.
Historically, financial messages permitted unstructured, free-text formats for postal addresses because humans were involved at all steps of every process.
The free-text format, apparently, allowed people to claim they shared an address with the Financial Conduct Authority in London or (even better) list Buckingham Palace as their home and find their applications go through without a hitch because, humans may have been involved but that didn’t always mean they were paying attention.
Fraudsters or pranksters, it doesn’t much matter, because the ISO20022 structured address requirement put an end to all that.
The new standard requires breaking this data down into rigid, dedicated XML fields (such as isolated town, postcode and country blocks). But, because legacy databases, corporate ERP systems, and retail apps frequently generate messy or unformatted address text, moving directly to structured mapping results in severe data corruption, rejected payments, and operational gridlock.
In short: this hasn’t worked.
And it hasn’t worked because in order to move to a shiny new world, we need to transition. And to do that safely, we need to test. And to do that sensibly, we need to simulate legacy back-end systems. And that is hard.
So it hasn’t been done.
Recognizing these hurdles, SWIFT extended its migration deadlines, giving financial institutions a vital window to test and secure their architectures.
So now we have more time to solve this problem.
The problem is the same.
The legacy estates are the same.
The risk is the same (if not greater: amplified by the prolonged period of uncertainty).
The deadline moved but, as we already said, dates don’t matter other than to mark how long some things take. Dates don’t matter unless you turn them into milestones.
And turning this one into a milestone is hard. Because it is hard to test every system of record. And yet that is the exam question and the extra time is thankfully received but it will run out before we know it unless we get our skates on.
We have about a year.
And we have a technology that is very good at parsing data and we also have a collective need to demonstrate value.
And Iliad has a context model protocol that can actually execute the way you thought AI would work. Because they are willing to do the work to map the complexity and manage the context that will unlock the outcomes needed.
A problem that needs a solution.
A capability that needs material proof points.
And a window of time that could not be more perfectly aligned.