The Expert View: Business Leaders Identify Best Practises for Succeeding with AI at Scale

Artificial intelligence (AI) can offer transformative potential, but organisations face challenges to deliver the value at scale that can turn it into an enterprise-grade capability. Identifying how to get this balance right was the focal point of a recent Business Reporter dinner at the House of Lords, hosted by SoftServe and Google for industry leaders.

 

As the discussion unfolded the complex challenges of successfully deploying AI became clearer. We will look in more detail how these ranged from the way businesses approach data readiness and process intelligence, to validation, guardrails and transparency.

 

Human importance

 

Given the impact AI will have and the access it needs to have across an enterprise, it was agreed that the success of AI will hinge not just on architecture and algorithms. People, processes and the ability to build trust at every step will remain equally important. In many ways this means people are as essential to making AI work as the tech itself. The emphasis must be on making sure employees not only know how to use AI and agentic tools but how to manage them as if they were employees. Everyone will be both a techie and boss.

 

Introducing the event, Scott Kemp, Director of Industry Solutions at SoftServe, noted that because of these holistic challenges, the success of AI will increasingly depend on strong partnerships. “We’ve learned that organisations need a strong partner with AI experience to resolve complex problems, as AI can be an extremely complex problem.”

 

Saffi Ali, Principal Architect at Google Cloud, framed the goal of the evening, saying he was keen to understand the problems business faced and the obstacles to scaling. Executives replied that, while AI was becoming embedded in almost every strategic business roadmap, turning that ambition into a reliable, enterprise-grade capability remains a work in progress.

 

Delivering value

 

Many felt that, in principle, AI could be “almost ready to reshape everything” - so there is a great sense of urgency to get projects working. But, as one said: “We don’t have three years to plan. We have to build the plane and fly it.”

Many reported successes with various forms of AI, whether newer generative AI or older machine learning technologies. In manufacturing, predictive tools already determine what will be produced tomorrow across hundreds of lines. One said, “Tomorrow we’ll make 200 million things, and the forecast for what to make is done by machine-learning tools.” While in supply-chain planning, AI can now optimise shipping container loads and distribution flows more quickly and efficiently than humans.

 

Saving time and money

 

Healthcare and pharmaceuticals are particularly active as AI can summarise clinical-trial data in two weeks, compared to the months taken previously, accelerating submission pipelines. AI is also increasingly used for drug-development, where shaving even a few weeks off years-long R&D cycles brings meaningful commercial and clinical advantage.

 

Others discussed using AI to consolidate years of brand research to generate marketing insights, evaluate new concepts or provide rapid initial assessments of whether an idea is worth pursuing. Car manufacturers use AI to assemble risk profiles across factory networks, spotting issues invisible before human observers.

 

Across these examples, a trend emerged with a shift from AI as a tool to AI as a collaborator, especially for long-chain, data-intensive tasks that humans struggle to manage at speed.

 

Deployment challenges

 

However, despite some strong use cases, deployment remains hard. Concerns about job security meant many were reluctant to embrace a technology perceived as a replacement rather than an enabler. But, working with AI in industrial environments, many realised that human supervision and interaction will be essential for years, meaning adoption is as much about confidence as capability.

 

Participants said change management is a challenge, while fear of hallucinations (AI errors) was pervasive. They said many teams understand AI isn’t flawless but still might not question AI-generated answers when they should. This can quickly erode confidence, unless leaders make clear early on that issues are inevitable and can be addressed.

 

Data readiness is also critical. AI “survives on data,” one said, so organisations must be prepared to open access to data and adapt processes to new workflows. But they must also be selective as AI isn’t needed everywhere, and poor processes cannot be redeemed by adding AI on top. Process intelligence is therefore essential preparation - mapping workflows, identifying bottlenecks, and understanding where AI can meaningfully intervene.

 

Validation and risk

 

As AI moves into critical workflows, the question of validation becomes unavoidable. But what AI should be compared against? Is it a perfect outcome, or a typical human performance that includes error, as there are risks in both. Organisations should therefore generate sufficient hypothetical edge cases, or “black swan events”, to better understand model behaviour, but these can be hard to create.

 

Metrics also matter. Before deploying AI at scale, leaders must define the measures a system must excel at and ensure it has access to the background information a human would naturally bring to the task. This then forms part of a broader AI assurance process, involving both internal and external auditors who assess performance and governance.

 

Many stressed that this readiness should flow in two directions, both preparing the organisation to work with AI and preparing AI to work with the organisation.

 

Conclusion

 

In summary, Saffi Ali echoed the urgency heard earlier. “Customers tell me that they are having to solve these problems at pace. They need to adopt AI in the next year, otherwise they will start counting how much revenue they’ll be losing.”

 

For Scott Kemp, the path forward rests on balancing two intertwined challenges. “The tech challenge is about laying the foundations - paving the road for others to drive on. The business challenge is to let teams uncover the real problems and run with the solutions. People are much more engaged when they feel they have ownership of the solution.”

 

The good news was that not only are businesses embracing AI, despite the challenges, but that fears of people displacement were allayed by the realization that humans are still going to have a central role in ensuring AI is safe and successful. All agreed that not only is AI here to stay, but the responsibility to make it work safely and effectively depends on everyone.