Too many businesses to count, in just about every sector imaginable, have by now spent months, if not years, and thousands, if not millions, fine-tuning their AI tools and strategies.

The numbers suggest that this approach is already paying off. A study from Google last summer found that 74% of enterprises using genAI report ROI within the first year, many of them at 6% or more.

Likewise, research from Snowflake published only last month found that a staggering 93% of UK firms are now reaping efficiency gains from the adoption of genAI, with 69% using genAI tech in their security operations.

So, with all the hype surrounding the revolutionary promise of AI, why are so many businesses claiming that the technology is failing to deliver? 

In October, research from the Boston Consulting Group found that 74% of global companies are struggling to achieve value, while other studies show that few AI projects are making it out of the planning stage, leaving the average ROI for AI-focused investment sitting at just 2.5% over the last year.

With studies and data from big-hitters both making cases for and against the immediacy of AI’s efficacy in an enterprise setting, it’s easy to feel a bit confused. 

Clearly, there’s a lot of nuance here – with AI being integrated into tech stacks across so many industries at so many different levels, the metrics for success (and failure) are invariably all over the place. 

That’s before you even consider that with the amount of money being spent on AI projects, there’s a serious case to present its success in as favourable a way as possible.

All that to say, if your company’s AI efforts haven’t lived up to expectations, you’re in good company. But according to Orlando Machado, chief data and AI officer at the LEGO Group, there are a few guiding principles that can help businesses refocus their strategies and real value in AI.

Speaking at this year’s DataFest at Edinburgh’s Assembly Rooms – an event that also marked the tenth anniversary of the Data Lab – Machado offered AI entrepreneurs some insights derived from his decades of experience in data science, statistical modelling, machine learning, and more latterly AI, to help businesses cut through the hype and focus on delivering meaningful, measurable value.

“When something is moving at a dizzying pace, it can be difficult to navigate, and it can be difficult to prioritise, and it can be hard to understand what will still be true in months, in years to come,” said Machado. 

“It can be very hard to work out how to influence AI in a company where there are cultures and structures and agendas, all sorts of things that don’t change at quite such a dizzying pace.”

With a packed audience of tech leaders and enthusiasts, Machado offered three key lessons designed to unlock the business potential of this incredibly fast-moving technology.

Build for impact, not perfection

“An algorithm that’s perfect in a lab environment might not always be appropriate in a production environment,” warned Machado. 

It’s a reality check that should hit home for many tech leaders. There’s often a temptation to rush to the end game  –  perfect the model, tune it endlessly, and optimise every inch of output, a timely and expensive process. 

But in the real world, budget constraints, implementation costs and resource management are consistently among the top roadblocks to digital transformation, with 45% of firms citing AI overheads as a key challenge to deploying the technology, according to a 2024 KPMG study.

More than two years on since the release of ChatGPT instigated a new AI goldrush, these kinds of roadblocks mean that two-thirds of companies are still failing to scale the technology across their organisation.

Rather than spending time and money on chasing ever-new AI projects, at least 30% of which are predicted to be abandoned by the end of this year, companies should instead focus on keeping engineering spend under control.

Evidence suggests that plenty of firms are now taking that approach. A recent AWS survey found that most organisations are opting to use pre-trained, off-the-shelf AI models and simply layering custom applications on top, rather than building solutions from scratch. 

It’s faster, cheaper, and, crucially, can prove more practical.

“Don’t forget subtraction,” adds Machado.

“Don’t forget taking away the noise, taking away the clutter, taking away the distractions, taking away the jargon.”

In crafting AI tools and strategies, simplicity is best. That was a message reiterated by more than one speaker at DataFest, as well as industry experts at other recent events.

“Lift the lid on AI, and you have any number of techniques, and each one of these techniques is fascinating and it’s powerful, whether you’re talking about agentic or RAG or LLMs,” said Machado. 

“They all have the ability to distract because each one of those techniques can be a rabbit hole that you can fall down that stops you from making an impact.”  

Solve for today’s problems, not yesterday’s

Have you ever heard of the Netflix Prize? Neither had most of the DataFest audience, but the story behind it illustrates Machado’s next lesson.

Launched back in 2006, the Netflix Prize offered data scientists the chance to compete for a $1 million dollar jackpot if they could find a way to improve the movie recommendation algorithm the company was reliant on.

Thousands of teams from around the world entered, with the contest being won by a group named BellKor’s Pragmatic Chaos in 2009, who managed to improve the predictability of Netflix’s movie preferences by 10.09%.

But that’s where the story only becomes more interesting.

“Netflix didn’t ever use the algorithm,” said Machado.

While one of the main problems with the algorithm was that Netflix found it too costly and difficult to scale, Machado points to another, perhaps more obvious issue.

Netflix officially launched its streaming service in 2007, not long after it invited data scientists to compete for the Prize, but by the time the teams were refining their algorithms to improve DVD rental queues, the business model had already moved on. 

What Netflix really needed wasn’t better curation for DVDs mailed out days later, it was a recommendation engine that could instantly serve up what people wanted to watch right now.

“It turns out that this dramatically shifted the requirements of the Netflix algorithm, which meant it was making the wrong recommendations,” Machado points out.

With research from Fortune last year finding that nearly 75% of corporate AI initiatives fail to deliver because of a rush to implement without alignment to business processes, Netflix essentially giving away $1 million is a reminder that technology should supplement a business model, rather than define it, something Machado sums up in just a few words.

“AI strategy has to start with a business strategy.”     

Recommended reading

• ‘Sycophant-y’ ChatGPT Rollback Sparks AI Personality Debate
• How to Avoid Wasting Months on AI That Doesn’t Work
• Report: More Than 50% of Fraud Is Now Driven by AI

Learn fast and go again

Over the past decade, AI adoption has skyrocketed, more than doubling between 2017 and 2022, with the release of the first incarnation of ChatGPT arguably fuelling explosive growth in the years since, with the latest figures showing that 78% of organisations use AI in at least one business function, up from 55% just two years ago.

Maintaining a competitive edge in AI means moving quickly, and according to Machado, that begins by cultivating the right mindset.

“Learning fast and going again is an important cultural aspect if you want to take advantage of any fast-moving technology.”

But simply jumping on the bandwagon isn’t enough. The businesses seeing real ROI on their AI investments are the ones prepared to fail fast, learn fast, and move on. 

That kind of agility is becoming a hallmark of AI maturity. According to IBM’s 2024 AI in Action report, organisations leading the way in AI have 40% more applications in production than their less mature peers, and are able to move into deployment more quickly thanks to clear governance frameworks and serious investment in people.

For Machado, it’s people, rather than technology itself, that is key in unlocking business value from AI.

“To me, the thing that is not changing is that great AI starts with great conversations.”

“This was also true when we were talking about machine learning, data science analytics, computer simulation and statistical modelling,” said Machado.

“The skillful, impactful work came when people were having conversations, technicians and non-technicians. So, although it can be extremely exciting to think about a new development in AI, that’s not always the most relevant thing to talk about when you’re trying to make an impact.” 

It’s not just about deploying bigger and better models, it’s about building a business structure that’s comfortable testing ideas, learning from what doesn’t work, and trying again with something better.