Subscribe
About

The value of an AI centre of excellence

Erwin Bisschops, Data Practice Lead at Entelect.
Erwin Bisschops, Data Practice Lead at Entelect.

AI – particularly generative AI – is a hot topic across organisations. However, taking a haphazard approach and rushing into generative AI projects will not drive AI success; the better approach is to implement AI strategically, with a multidisciplinary team within an AI centre of excellence.

This is according to Erwin Bisschops, Data Practice Lead at Entelect, who specialises in building end-to-end AI capabilities within organisations.

Bisschops says: “The likes of ChatGPT unleashed a tsunami of discussions about generative AI, and now everyone wants an LLM. I see too many proof of concepts being executed in organisations – they might have 10 or 20 initiatives under way at same time, yet few reach a full production environment. They would be more successful if they focused on just one or two initiatives and made sure they ended up in applications end-users used every day, fully integrated into relevant business processes.”

Hampering AI adoption

Bisschops says there are several factors undermining efforts to roll out generative AI in organisations.

“Organisations don’t realise AI is very broad – it encompasses and impacts data science, machine learning, statistics, organisational culture and people. Generative AI may look so easy when people are using Gemini or ChatGPT, but those solutions are just the tip of the iceberg. It’s more complex when it comes to deploying AI to solve specific problems within an organisation,” he says.

Other hurdles include a lack of appropriate skills and no centralised AI function, Bisschops says.

COEs for AI success

An AI centre of excellence (COE) is the best starting point for an organisation wanting to adopt generative AI, says Bisschops.

“Over the past decade, we have seen organisations adopting a more decentralised data analytics architecture. Inevitably, they have started to implement AI in a decentralised fashion too. However, the challenge with this is they may be using multiple toolsets, they aren't sharing learnings across the organisation and governance of the AI is not centralised. At early stages of maturity, a centralised model is preferable – it allows organisations to lay down the rules of the game, benefit from economies of scale and identify what works and what doesn’t.”

Bisschops notes there are six key pillars in effective AI COEs: people, culture and integration, learning and development, recruitment, delivery and business development.

“A COE is a very important starting point to stay in control of AI. It should be driven by business – normally someone like the chief data and analytics officer,” he says.

He says the COE should help drive culture change within the organisation, and upskill business leadership to understand what will work, as well as on the real challenges of AI. “The COE should also prepare leadership for the possibility that AI initiatives could fail,” he says.

“The COE must drive learning and development programmes to improve AI skills, using approaches like knowledge-sharing sessions and events, and mentoring and reverse mentoring. It should manage talent acquisition and internal and external recruitment for AI initiatives. In the culture and integration component, the COE should market these centralised AI efforts – perhaps through videos, newsletter communications and simply ‘walking the business corridors’. The COE must ensure it has effective two-way communication channels. Crucially, the COE must manage AI delivery guidelines, playbooks and templates, demo projects and monitor and review implementations to drive maturity forward,” he says.

Despite the need for multidisciplinary team involvement, Bisschops says successful AI implementations need not be lengthy, complex projects. “Entelect has implemented these kinds of solutions in as little as three months. It’s about scoping it correctly, managing expectations and taking a multidisciplinary, end-to-end approach across business analysis, quality assurance, user experience design, data science and data engineering, and software development to successfully implement AI.”

Share