What are the practicalities when moving from generative artificial intelligence (genAI) experimentation to genAI in production? How do you identify the typical obstacles that stand in the way of success? And how can you best navigate a path around and over those obstacles? 

To address these questions and more, Tech Monitor associate editor Jon Bernstein was joined by Mark Oost, Vice President, AI & Generative AI Group Offer Leader at Capgemini, and Michael Taylor, Field CTO, AI/ML at Snowflake, for the latest edition of our Power of the Possible series.

Power of the possible? Leaders still believe

The conversation began with a quick recap. Given this series began last year, it’s worth asking whether business leaders still have faith in the transformative potential of genAI? Or have they entered what Gartner describes as a ‘trough of disillusionment’? 

Both speakers were bullish in response. 

“Business leaders are a bit more immune to some of the hype around general artificial intelligence, and the hyperbole that happens in political spaces,” observed Snowflake’s Michael Taylor. “They are focused on the practical benefits that are finally starting to emerge.” As Taylor goes on to explore, genAI’s ability to cut through the mass of data organisations possess is providing practical benefits. This applies both in customer-facing use cases, such as optimising call centres, and back-end use cases including more personalised onboarding of new employees. 

Capgemini’s Mark Oost agreed that genAI remains salient. “It’s still very, very relevant to our clients,” he said, noting that when he’s with a client it’s not uncommon to discover that “95% of the people in the room” are using genAI in some form. “It’s holding up to the hype at the moment,” he added.

Taking genAI live: typical deployments

Asked which use cases organisations are choosing to take live, Oost said there were commonly two types of deployment – value cases and efficiency cases. The former is where value is added “to existing services that [my customers] give to their clients.” The latter focuses on delivering productivity gains by applying AI to internal knowledge. 

One big area of interest is where businesses previously had to “churn through” a wealth of information in order to efficiently create and apply data insights. As an example, this applies to call centres where organisations can now deliver personalised responses that often address similar – or exactly the same – questions. Although this may seem like a simple “efficiency” use case, Taylor insisted that for those that run call centres it “is utterly transformative” with businesses seeing positive impacts to their bottom line.

Taking genAI live: the obstacles

The path from proof of concept to production is not without its challenges, as both guests acknowledged. Among the obstacles are a lack of relevant in-house skills and growing complexity. On the latter, Oost noted that the introduction multiple types of AI coupled with agentic AI – where agent often interacts with agent – is creating potential difficulties. “It becomes more and more complex by the day,” said Oost. “We are seeing our clients really struggling to comprehend what is about to happen.” 

Taylor echoed Oost’s concern about in-house skills, choosing to focus on one skill in particular: scoping. “[It’s about] understanding how to scope these projects: what makes a nice, bounded solution.” This is particularly the case with agentic AI. Organisations need to think about which activities should be replaced with which tool and how those tools should interact with each other. “It’s about defining your needs, and defining your expectations,” Taylor said. Organisations should think, too, about the need for observability in respects of human-AI collaboration.

Filling the skills gap

Capgemini has introduced a solution called RAISE in part to address the in-house skills challenge. An accelerator enabled by Snowflake, Oost described Capgemini RAISE as “starter pack with building blocks” designed to ease and hasten the deployment of agents. “It’s already pre-built, pre-packaged with your technology in it which means we can very quickly scale up different solutions at the client side.” For the customer, it brings a range of expertise together, from data science capabilities to cloud and integration specialisms. 

Building on the advantages of external experience, Taylor said: “Consultancy is at its best when you’ve got people … who have done it before somewhere else, and they take those learnings and repeat them with each subsequent client.”

Final thoughts

Asked to leave viewers with a final thought or piece of advice, Capgemini’s Oost urged those who have yet to undertake their genAI journey to wait no longer. “If haven’t started already, you’ll be quite behind because 90-95% of organisations are either working with it or testing it out.” 

Offering his own final thought Snowflake’s Taylor said: “It’s the old adage: ‘When is the best time to start? Two years ago. When’s the next best time to start? Today’. There’s no substitute for getting in there and trying things out.” 

It is in that spirit that Snowflake makes it easier for organisations to get up and running with genAI. 

To learn more, download Snowflake’s Ultimate Guide To Data + Ai For Industries 2025 here.