Digitisation is process optimisation

It’s April Cools’ again. To celebrate, some friends and I made content outside our normal brands. Here goes mine. (Needless to say, this post does not represent the opinion of my employer.)

The other day, I was trying to open an account as a new customer of an online-only bank. Naturally, the onboarding process required a KYC (know your customer) check, vulgo, I had to provide my identity. The bank’s website informed me that I need to jump on a video call to do that. It took a few moments to get connected with a customer service agent, who walked me through various steps of holding my (physical) identity card in front of the camera, tilting it a few times, showing various numbers of fingers, saying particular phrases and so on. The process took maybe ten minutes. And voilà, I had my new bank account.

Sounds like a success story of digitisation, right?

Absolutely not. That whole story is completely bonkers: a sad indictment of the state of the industry. Because even though the onboarding took only ten minutes, it is wildly inefficient.

No worries, though: help is on the way.

I’m sure many bank managers have looked at this here KYC procedure and decided that ✨ AI ✨ will make it better. Just replace the human agent and have the ✨ AI ✨ carry out the repetitive work of telling many customers exactly the same thing and validating their responses. And sure, big tech and businesses invest eye-watering amounts of money into rolling out ✨ AI ✨.

You might very well think that these kinds of things are exactly what ✨ AI ✨ was made for. You couldn’t be more wrong. Let me explain.

Germany introduced the so-called “new identity card” in 2010. That is sixteen years ago. It contains a standards-compliant NFC interface and can be read by all modern smartphones. After certification, service providers can access certain data elements, such as name and address, through an open-source app. (There are also dedicated card readers that are supported by major desktop operation systems.)

Performing KYC using this electronic identity card takes literally seconds. It is a process that involves no customer service agents whatsoever, neither of the human nor the AI kind. It is perfectly secure since it relies on hardware security and cryptography. (Other European countries have similar cards, and there are interoperability standards, too.)

In other words, it just works.

However, there is a catch. There are some compliance costs involved, and it requires redesign of your business process.

So, the question is: what would win?

Painstakingly transforming your business process to use boring, proven, secure, and ubiquitous technology? Or just slapping ✨ AI ✨ on it?

I think you know the answer.

But optimising processes is unglamorous work. You need to hire people that have been in the industry for twenty years and have seen some shit. Sadly, those are more expensive than ✨ AI ✨ (at least in the short run). I would much prefer business to focus on the fundamentals.

We have seen this before. Fed up with convoluted payment processes, blockchain people have been selling ⛓ smart contracts ⛓ as a cure. They suggested to model business processes as code, but instead of executing said code on a computer, you would execute it on many computers simultaneously. I know, breathtaking levels of innovation going on here.

How well did that play out?

The point is: there is no free lunch. Firms love to wax poetically about digital transformation. (The only thing they love more is to complain about regulation, especially the European kind.)

But they are failing to address the inefficiencies of their own making.

Digitisation is process optimisation, and process optimisation is digitisation. There is no shortcut.

Let’s look at another scenario: say, your current process relies on PDFs containing data as text. Zeitgeist dictates to use an ✨ AI ✨ tool to parse such PDFs. However, this introduces non-determinism, exciting new error cases, and novel security flaws.

The better approach is to define machine-readable metadata standards. This is exactly what EN 16931 is trying to achieve for invoices. And what do businesses do? Complain about regulation mandating the use of EN 16931. You know, a standard whose adoption (perhaps even voluntarily) could’ve massively simplified processes, therefore driving down costs.

Of course, someone is already selling an ✨ AI ✨ tool to extract EN 16931 metadata from regular PDFs. Back in my day, if you would’ve asked a development team to implement a service consuming PDFs, you would’ve gotten laughed out of the room. (Unless the room is in the basement of a national library, where old documents are being scanned.)

Perhaps, at some point, someone will implement this in existing invoicing software. Maybe even in Java. It will take a few weeks, but it will be correct.

A software engineer can dream.