Salesforce UK&I CTO on the genAI upskilling challenge facing businesses

While the term ‘skills gap’ is often heard around new technology and increasingly used in relation to AI, what does this mean in practical terms?

Paul O’Sullivan, UK & Ireland Chief Technology Officer at Salesforce cites the tech company’s research from last year that found close to two-fifths – 38% – of UK workers are either using, or planning to use, generative AI in their roles, but a majority of these (62%) also believe they lack the skills to use the technology accurately and safely.

“That’s a huge risk,” says O’Sullivan. “…Even individuals within the workforce that might not have direct access to the technology itself need to have an awareness and an understanding of what’s happening with the technology.”

O’Sullivan was speaking at the official launch of Salesforce’s new AI Centre, where the company hosted an intensive full-day workshop on building generative AI applications, something that forms part of a new pledge by the company to upskill 100,000 developers globally.

“It’s a really exciting time, but we’ve got to get this right,” said O’Sullivan, referring to the upskilling challenge ahead.

The data conversation

“AI won’t be successful without the data that fuels it,” O’Sullivan said. Speaking at a panel discussion during the launch event, O’Sullivan observed that when working with Salesforce customers to implement generative AI, “We … start on the AI conversation, and everyone’s excited about the technology and the potential it holds – but then, we come back to the data conversation.”

He noted that over the decades, a lot of organisations fragmented their data into “different siloed pockets” in a bid to become more agile and keep pace with consumer demand, “and what we’re now seeing is, enterprises are going on a journey to say, ‘I’ve done abstraction to increase agility – I’m now having to do orchestration to bring my data back together.’ This is why we’re seeing a huge investment … in data platforms.”

From there, an additional challenge is activating the data; O’Sullivan estimated that as much as 90% of customer data is typically stored in an unstructured format. “…there is a lot that we’re doing in making sure that we can leverage the data that exists not only in the CRM, but across the enterprise, and then activate it – both structured and unstructured.”

This may also require a process of change management and learning to bring the organisation’s workforce along on that journey of evolving their data practices.

Change management “is going to be really important”

“I think [a] key thing is – the change management element in the workforce in the future is going to be really important,” O’Sullivan said. “How do we make sure people put [the technology] to use in a consistent way that equally supports the right levels of productivity and efficiency?”

He recalled asking Salesforce customers how many prompts it typically takes before they get the response they’re looking for from ChatGPT – most cited between four and five attempts. “If you think about that on a cost scale – that’s four or five times your cost before you get the outcome you’re looking for,” he said.

However, achieving a level of “maturity” with generative AI prompting can lead to a more truly conversational experience. “…If over a period of time, I learn and develop as an individual that actually, if I put in particular objects within the appropriate prompt in the right places, it will significantly improve my accuracy,” he said. “… at that point, we have that conversational unlock [in which the user is] interacting with the technology seamlessly.”