Arcadian's AI success: Start small, solve real problems

There's a simple reason behind a claim that most AI rollouts are bound for failure. In a gold rush complete with easy promises of big productivity gains, the basics are ignored in favour of the perceived advantage of rapid action. But as with any technology since the dawn of time, caution being the watchword while sticking to good basic business principles is always going to deliver better results.

That's according to Arcadian Digital director Adrian Randall. He agrees that while an enduring (for all the wrong reasons) justification for action, 'shiny new thing' makes for a terrible business case. "You really want to start with well-defined problems that AI can solve. And you want well-defined processes and solutions to those problems that deliver a measurable return on investment," he sets out.

His observation prompts the next question: Sounds like business analysts should get involved up front, in much the same way they do for any other complex technology project? "Very much so, actually, and we're in the process of hiring for exactly that role," Randall confirms.

AI has burst onto the scene with almost astonishing rapidity, but that in no way negates Gartner's well-known Hype Cycle. Sure, from 2023 to present, AI has gone from curiosity to production in what seems like record time (set aside the fact that AI is conceptually far older). But that hasn't come without an equally rapidly accelerated set of challenges and disappointments.

While most put AI somewhere on the Peak of Inflated Expectations, Gartner itself puts it into the Trough of Disillusionment. What that means in practical terms is a gap between those expectations, and the delivery of clear value and returns on money invested.

Some of the risks are prominent and public: Deloitte's expensive AI-generated government report. The lawyer citing non-existent case law. The vibecoder deleting an entire application. The endless low-effort AI trash on LinkedIn.

Other risks don't make the headlines; poorly executed projects that annoy people at work rather than making them 'more productive'. AI targeted at the wrong things, where processes are incompletely mapped or no mapping was done at all, leading to vast gaps between the drawing board and in the field.

There's no need to throw out the baby with the bathwater though, notes Randall. "We've delivered several successful projects that have in common what could be called the basics. Clearly identified problems or tasks where AI can and does deliver a measurable advantage. A business case based on numbers. And defined objectives rather than vague approaches which really come down to not much more than 'everyone is doing AI, so we should too'."

Not, it seems, rocket science, even if the technology is the latest greatest and the talk of the town.

Randall adds that AI projects needn't be grandiose, either, advocating instead for a 'slow and steady' approach where early wins drive future investment. "What often comes as something of a surprise is that these kinds of projects generally aren't big capex initiatives but instead serve as demonstrators of where and how AI adds value."

He says this helps build confidence in AI capabilities, noting that trust remains an issue to be dealt with. "We address trust in part by looking for consistency of results, so we'll use several LLMs, and look for consistency in the outputs. If there isn't alignment, there could be a problem."

There's the by-now tired cliché that we want AI to do our dishes and fold the laundry so we can write and do art. In other words, AI has the potential to change jobs in unwanted ways, for example making the writer an editor scouring for AI mistakes, or the lawyer reduced to associate fact checking case law. "That's actually an interesting point," reckons Randall. "And it probably shows that when looking at any AI project, you want to target the things you'd rather not be doing. Instead of the high value tasks that people actually enjoy."

That's detail, really. The bottom line is that common to every successful project, whether AI or otherwise, is the clear definition of need, accurate costing, and disciplined delivery.