The generative AI ‘scramble’ is over – now it’s time to sort your infrastructure or fall behind
The early days of the generative AI boom were somewhat of a “scramble”, according to Thomas Meyer, General Manager and Group VP of IDC research, EMEA. But now the industry has entered its “pivot” era – and enterprises need to take a reasoned approach to investment.
Speaking at Huawei’s 2025 IDI Forum in Munich this week, Meyer detailed findings from a recent study conducted by the consultancy exploring what the priorities are for enterprise tech leaders dabbling in AI.
A key takeaway from the talk was that while many organizations are still in a nascent stage, bolstering the underlying foundational layers in an enterprise tech stack has become critical.
Organizations simply can’t afford to sprint haphazardly into AI transformation strategies, they need to build from the ground up in a responsible manner.
“Organizations are still in the starting blocks, but they are beginning to move to the AI pivot,” he told attendees. “So when you think about where your organization is and how you can compete successfully in the new era, then that is now the time to act.”
This is by no means an easy task, however, and one that requires significant investment on the part of any enterprise, regardless of size. However, a great starting point for any business looking to make the ‘pivot’ typically involves assessing the benefits of the technology on a case-by-case basis.
“Most organizations start at the point where they go ‘okay, by function, which cases are important?’ – so if you pick one, and I’ve picked a very popular one, the contact center, you then think about what are the key spaces that exist there.”
“It’s great to get real time summarization, it’s great to get multilingual support. So we can make these things more automated quicker. We can increase customer happiness, and obviously we can increase the level of productivity in organizations,” Meyer added.
Contact centers and customer service roles have been repeatedly identified as ripe for automation, particularly since the emergence of agentic AI in late 2024. Salesforce, for example, specifically highlighted the potential of the technology in customer services when it launched its Agentforce service in September last year.
But this is limited to specific roles within an organization, Meyer noted. On a broader, industry-wide scale, there are signs that AI adoption is gathering pace across a host of sectors.
These include the media industry, which Meyer said has “embraced the new era of AI very quickly”. Similarly, tech companies, telecom operators, and also financial services firms are all bullish on agentic AI, based on the consultancy’s findings.
The latter of these, he added, are beginning to show tangible practical use cases for the use of the technology in frontline operations.
“Finance has been at the forefront of technology for many years because there’s a clear advantage when looking at finance,” he said.
Drilling down deeper at an individual company level, Meyer said firms can be divided into ‘survivors’ and ‘thrivers’ based on their pace of adoption and success – and there are glaring differences between the two.
“What are the ones that are moving ahead really quickly? What are those organizations doing? They invest early and look at chatbots, and have done this for a long time. They also look at fraud, and ideally try to prevent fraud. They’re looking at governance very clearly.”
IDC’s findings, specifically on the use of AI to tackle fraud in financial services, aligns closely with previous research on adoption rates in the industry.
Analysis from UK Finance in late 2023 showed nearly three-quarters of banks had piloted the use of generative AI to improve fraud detection and risk analysis. But during this early stage of the technology, many noted they were struggling to achieve a solid return on investment.
Meyer highlighted issues around return on investment and success rates for AI adoption projects. On average, he said, organizations run “around 14 proofs of concept” – yet out of those, only five are put into production and just two are successful.
The infrastructure conundrum
While all use cases will be dependent on demand or suitability, the overall success of any project still depends on the underlying infrastructure, which was a key talking point at the IDI Forum and an issue Huawei was keen to address.
The firm launched its AI Data Lake solution, which seeks to address issues such as infrastructure visibility, data storage limitations, and enterprise access to relevant data for AI training and inference.
Meyer insisted that moving forward, enterprises must have a robust tech stack in place, or the aforementioned use-cases will be rendered pointless without the underlying infrastructure to support them.
“Scaling [AI] with people and convincing people that AI is a good thing, and working with AI is great, is just one side,” he said. “But when you bring it down in terms of what drives things forward and what holds things back, infrastructure is a big element – the top element in terms of moving AI along more quickly.”
Meyer said a common hurdle now faced by enterprises is how they can build “scalable areas of data” and data lakes to provide easy access to the data that is essentially the lifeblood of AI.
Similarly, questions over whether data is hosted in the right format – be that cold, warm, or hot – are also posing challenges. All three have a valuable role to play in AI training and inference, but once again infrastructure limitations create larger problems – particularly with regard to store and security.
Ultimately, the initial rush of the generative AI boom has created problems for many enterprises hoping to capitalize on the technology, Meyer noted. The hype and hyperbole surrounding the technology – and its business benefits – appears to have left gaps.
“I think many of us want to fly before we can crawl, walk, and run,” he said. “But the fact remains that to be successful in the intelligent era, you need to get the foundations right for AI running infrastructure and data.”
“Without data, it’s going to be very tough, and there is not much of a shortcut to take moving forward.”
MORE FROM ITPRO
Source link
