‘Garbage in, chaos out’: overcoming the curse of bad AI data

Brian Shannon, Chief Technology Officer, Flexera; Kimberly Floss, Senior Director, Data and AI Product Management, PepsiCo

---

Nowadays there is one surefire way for a business to maintain a competitive edge – by becoming AI-first. This is less about using AI as simply one of many tools, and more about integrating it into every aspect of a business operation, so that it becomes a driver of evolution and growth. 

 

Organisations eyeing this transformation have to assess their AI readiness – their capacity and willingness to undertake the kinds of system-wide changes needed to really become AI-first.

 

They’ll consider things such as the capability of their employees and their capacity for short-term disruption; their IT infrastructure; and whether they have the right workplace culture for such a change.

 

But there’s another aspect of this transformation that often flies under the radar: the quality of their data.

 

Data downtime costs big

 

Because AI is trained on curated datasets, it can only ever be as good as that data. It doesn’t matter what AI model is being used, or who is operating it – if the data is poor, whether messy or missing key detail, then the output won’t be reliable. And poor data leads to ‘data downtime’ – systems can still run, but what they’re turning out is incorrect in ways that can be disastrous.

 

Businesses have learned the hard way the consequences of bad data. Cases abound of organisations suffering tens of millions of pounds in losses – or causing clients to suffer losses – because of data downtime.

 

“With AI, there’s the old adage, ‘garbage in, garbage out’,” says Brian Shannon, Chief Technology Officer at Flexera, which works with some of the world’s leading companies to maximise the impact of their tech. “What I like to say is, it’s ‘garbage in, chaos out’.”

 

Before being used for training purposes, data needs to be vigorously cleaned, standardised and validated.

 

Perhaps the most effective way of ensuring high-quality data, says Shannon, is to create feedback loops in AI systems so that you can tell whether you’re receiving the answers you expect from the AI, or hallucinations.

 

The more you feed into that loop, the more the model is refined. “Without the loop it is very difficult to know that your data and AI is good,” he says.

 

Good governance equals good data

 

Among Flexera’s clients is PepsiCo. Kimberly Floss, Senior Director of Data and AI Product Management at PepsiCo, takes a broad view of what constitutes good data.

 

“Good data must be good quality, good cataloguing and good governance,” she says. Additionally, businesses must “put in sustainable processes to ensure that they build confidence in the data from day one, and that that level of confidence continues over time.”

 

It all comes down to the integrity of the product. "It’s more of a quality versus quantity kind of conversation,” adds Floss, “going for the whole estate without those other processes around it wouldn’t add as much value."

 

Shannon, a staunch advocate of businesses across the board making the transition to being AI-first, says that things such as an AI council or AI governance board, or even an AI centre of excellence, can all help to ensure quality is high. But incorporating these also demonstrates a willingness to put AI front and centre of a business operation – the first step to becoming AI-first.

 

“When you have that [AI-first] endpoint in mind, the rest follows,” Shannon says. “So you need AI governance – security, knowledge of where data is going and how it’s being used – as well as data quality and data lineage. In other words, as it goes through your system, do you know how it’s being transformed?

 

“With those elements you can really build yourself as an AI-first organisation.”

 

The road to AI-first

 

Businesses don’t simply become AI-first overnight. It’s a process of maturing as an organisation. But it does require that special attention be paid to the central foundation of AI readiness: data.

 

“It’s not just about your engineers going and experimenting with AI and implementing a few widgets here and there,” Shannon reflects. “It’s a transformation of your entire company.”

---

See how Flexera can help your organisation build a strong AI foundation with reliable data