Artificial Intelligence Integration Fails to Transform Core Operations

Large corporations and government agencies have largely failed to integrate artificial intelligence into their core operations, according to Marc Warner, Accenture's global chief technology officer.

Warner made this statement during a recent virtual event for Fortune Brainstorm AI, where he was joined by three other executives from prominent companies: AXA's Matthieu Caillat, Mastercard's Greg Ulrich, and Mistral AI's Marjorie Janiewicz. They discussed the disappointing returns on investment from AI projects.

Despite widespread adoption of AI, few organizations have undergone a fundamental transformation to put it at the heart of their operations. Warner pointed out that most companies are using AI in auxiliary tasks such as marketing, note-taking, and transcription.

The executives suggested that the problem lies not with the technology itself but rather with the organizational changes required to fully harness its potential. Changing core business processes is a high-risk endeavor that many companies seem unwilling or unable to undertake.

Warner drew an analogy between the current state of AI adoption and the early days of industrialization, when electricity was initially used only for lighting before being integrated into production processes.

The key to unlocking significant returns on AI investment lies in fundamentally redesigning workflows, rather than merely layering AI onto existing processes. This is according to Ulrich, who claims that Mastercard has made substantial progress in this regard by adopting a more agile approach to product development.

By breaking down large teams into smaller units and streamlining turnaround times, Mastercard has been able to significantly accelerate its innovation cycle. However, Ulrich warns that even with these advancements, the process can quickly become bogged down if one step is accelerated at the expense of others. This creates a backlog that can be just as problematic as the original slow pace.

Warner's perspective on AI transformations shifts from viewing them as technology projects to understanding them as politics projects. He suggests that this shift in thinking is not yet widespread, and that many organizations still struggle with integrating AI into their existing structures.

Ulrich emphasizes the importance of CEO sponsorship in driving change within large organizations. At Mastercard, weekly updates are provided to the CEO's direct reports to ensure that everyone is aligned on the company's AI goals. However, Ulrich concedes that there is no one-size-fits-all solution to addressing the politics of organizational change.

When considering accountability for AI agents that go wrong, Warner notes that it's not a simple matter of assigning blame to one individual or department. Instead, he suggests that responsibility falls on multiple parties, including business owners for an agent's value, technologists for its functionality, and chief information security officers for its security.

Companies should focus on progress rather than assigning blame for their slow adoption of artificial intelligence, according to experts in the field.

Warner believes that companies should give credit to those who attempt to implement AI solutions, even if some efforts fail. He cautions against dwelling too much on fault, suggesting that organizations should prioritize learning from their experiences instead of engaging in finger-pointing exercises. However, he does make an exception for critical incidents such as major security breaches.

Janiewicz agrees with Warner's sentiment, pointing out that the more important question is whether a company has control over the technology it adopts and can set its own guidelines to ensure safe usage. Her company, Mistral, offers open-weight models that allow customers to run them on their own infrastructure, giving businesses greater autonomy.

The pace of technological change means that plans are often short-lived. Ulrich notes that consulting tools Mastercard developed about a year ago have become outdated and are no longer useful today, necessitating the rebuilding of new ones.

AXA's Caillat highlights the insurer's experience with AI, which has involved starting with 400 potential use cases to build enthusiasm before narrowing its focus. AXA aims to generate between €500 million to €700 million in recurring pretax value from AI by 2029 and is currently working on utilizing its vast collection of unstructured documents, totaling around 4 billion.

The adoption of artificial intelligence by companies is progressing at a slow pace, according to Accenture's estimates. The consultancy firm forecasts that the current rate will continue until around 2030 without any significant acceleration.

Energy constraints in Western countries may become a pressing issue as early as the 2030s, posing potential challenges for businesses.