Digital Economy Dispatch #200 -- What's the Best Way to Succeed with AI-at-Scale? Lower Your Expectations!

Digital Economy Dispatch #200 -- What's the Best Way to Succeed with AI-at-Scale? Lower Your Expectations!
8th September 2024

After several decades in the tech industry, it’s fair to say that I’ve seen organizations take a few spins around the digital technology adoption roundabout. A promising new digital technology comes surging out of the starting gate with great fanfare. Scientists and engineers in R&D labs are demonstrating astounding results and winning new grants to do more. Academics are excitedly writing papers and rewriting their teaching modules to inform students of the issues and implications of these new advances. VCs and Private Equity companies are pumping in money to fuel a new family of startups with promising innovative products aimed at disrupting every conceivable application domain.

The primary effect of all this activity is that expectations for the new technology go through the roof. Commentators and consultants declare the old world is dead. Headlines from the news media predict rapid fundamental changes as we enter a brave new world. Everyone scans the social networks to see who’s doing what with whom.

However, within a short period of time questions begin to be asked. Progress slows and we begin to come up against delivery obstacles, leading to persistent requests from the business leaders and financial controllers to demonstrate value from these investments. In a scene reminiscent of the Jerry Maguire movie, there are increasingly urgent demands to “show me the money!”. Unfortunately, with deeper review it is found that the business cases and financial models for these technologies have been so inflated that they rarely stand up to much scrutiny. Well documented use cases delivering a clear ROI are conspicuous by their absence. Confidence in the much-touted advances dips. After a great deal of heated discussion and hand wringing, the digital technology roadshow rumbles on in an endless search for the next wave of innovation.

Is this now what we’re experiencing with AI? Have we built up expectations for AI to breaking point? If so, how can we reduce the disappointment with AI adoption and maximize the value of AI delivery at scale? Perhaps there is a simple solution to increase your success rate with AI: Lower your expectations!

AI’s Long and Winding Road

This new technology boom-and-bust scenario is so common that it named by industry analysts at Gartner as “the Hype Curve”. According to this framework, most technology shifts begin with excessive excitement but soon experience the “trough of disillusionment” as progress stalls. Eventually, ways to address the barriers are found (something Gartner refers to as “the slope of enlightenment”) and a slower, more measured way forward emerges to deliver measurable results and broaden adoption.

It is not hard to fit the latest AI trends into this pattern. We’ve all been shaken by the possibilities of what we’ve seen recently with AI. Many of us have discussed and written about the transformative potential of AI. After many years of research, emerging tools and technologies have gained widespread attention.

But I've also witnessed the increasing pressure on organizations to deliver AI-at-scale quickly, often leading to inflated expectations, increasing tensions, and suboptimal outcomes. The reality is that AI is not a magic bullet. It's a powerful technology that requires careful planning, execution, and continuous refinement. Increasingly we are hearing warnings about the challenges to be faced when broadening use of AI in today’s complex operating context. Just as worrying, we also see tensions rise as the urgency to deliver more with AI to live up to its billing meets resistance from the reality of delivering AI-at-Scale in complex, high risk operating environments. In some cases, leading to significant conflict.

Yet, the allure of AI's promise has created a climate where organizations feel compelled to rush headlong into implementation, often overlooking critical factors that can increase adoption risks and significantly impact success. It’s a perfect storm of technology-driven increases in capabilities being positioned as the answer to a growing list of business and societal needs for better services delivered faster.

The Myth of AI as a Silver Bullet

One of the most pressing challenges is the pervasive myth that AI can solve all problems, from eliminating fraud and waste to reducing bureaucracy and speed up service delivery in government. While AI can certainly support many of these needs, overblown comments and shared misconception can lead to unrealistic expectations and a sense of urgency that can drive poor decision-making. For example, expectations that AI can automatically predict customer preferences and optimize inventory levels without any human intervention are misplaced. While AI can certainly assist in these tasks, it's not a substitute for human judgment and expertise in handling demand management and supplier unpredictability.

The Interconnectedness of AI

Another common pitfall is the assumption that AI can be acquired and deployed as a standalone technology. In reality, AI is deeply intertwined with an organization's existing systems, processes, and culture. Implementing AI-at-Scale requires a comprehensive approach that addresses the entire ecosystem. Neglecting this interconnectedness can result in fragmented solutions and limited benefits.

These issues are perhaps most often seen when AI systems are introduced without sufficiently investing in improving an organization’s existing data infrastructure. For instance, in financial services an organization might develop a sophisticated AI model to assess risk, but if the underlying data is inaccurate or incomplete, the model's predictions will be unreliable. Similarly, if the organization's existing IT infrastructure cannot handle the computational demands of AI, the deployment process will be slow, fragile, and unstable.

The Skills Gap in AI

Furthermore, the skills gap in AI is a significant barrier to adoption. Many organizations lack the necessary expertise to develop, deploy, and maintain AI systems effectively. This shortage of talent can lead to reliance on external consultants, which can be costly and time-consuming. Additionally, it can hinder the organization's ability to adapt to emerging AI trends and technologies.

For example, in many domains we have seen companies hire a consulting firm to develop AI-powered predictive analytics and management systems. However, if the company doesn't have the in-house expertise to maintain and update the system, it soon becomes reliant on the consultants for ongoing maintenance, upgrades, and support, which can be expensive and time-consuming.

Managing Expectations for AI Success

To mitigate these risks and maximize the value of AI, it's essential to be realistic in following the path to AI-at-Scale to manage expectations. Instead of focusing on rapid, large-scale deployments, organizations should prioritize small, incremental steps while developing the structures, skills, and infrastructure that will be required to deploy AI in a robust and responsible way for long term success. By starting with focused use cases and gradually expanding their AI initiatives, they can reduce the risk of failure and build a solid foundation for future growth.

For example, in the case of one healthcare client I am supporting, they have begun their AI-at-Scale journey by implementing an AI-powered chatbot to answer patient questions and provide basic information collated from the past few years’ customer service records. This is a controlled use of AI where many issues with scaling AI adoption can be understood, shared, and institutionalized. Once the chatbot is successful, the organization can expand its AI capabilities to include more complex core administrative tasks, including adjustments to supply chain management with healthcare suppliers, staff recruitment and training, and a host of different logistics and planning tasks.

Investing in Data Quality and Governance

Moreover, in many largescale AI efforts, recent surveys have highlighted that investment in data quality and governance is frequently under estimated. Clean, reliable data is the lifeblood of any AI system. Organizations must prioritize data management practices that ensure data accuracy, consistency, and security. Additionally, establishing robust governance frameworks can help mitigate risks and ensure ethical AI development and deployment.

In working with a large UK government agency recently, one of the first and most effective steps in moving to AI-at-Scale has been to implement a data quality assurance process to identify and correct errors in customer data. This is now being synchronized with their existing data governance policy to protect sensitive customer information and ensure compliance with relevant regulations.

Building a Strong AI Team

Finally, as found in all complex change programmes, building a strong AI team is essential. This involves investing in talent development, fostering a culture of innovation, and collaborating with external experts as needed. By assembling a skilled and motivated team, organizations can overcome the challenges of AI adoption and unlock its full potential.

While the human issues involved in AI delivery are often emphasized in an organization’s future strategy and plans, in practice there is often inadequate attention to ensure these are appropriately prioritized. For example, in discussions with a defence technology company recently I was pleased to hear that they had invested in offer AI training programs to its employees to develop their AI skills and knowledge. This included partnerships with universities and research institutions to collaborate on AI projects and stay up-to-date with the latest advancements. However, in discussion with employees on the ground, such efforts where quickly dismissed when placed against existing deadlines, workloads, and client pressures. Few employees had made use of them and instead turned to more informal, just-in-time learning mechanisms on an as-needed basis.

Expecting Less, Delivering More

Despite the immense potential of AI, we must be careful not to expect too much too quickly. Moving from pilots and experimentation with AI adoption often faces challenges due to inflated expectations and a lack of realistic planning. While AI can be a powerful tool, it's not a magic bullet. Organizations must approach AI implementation with a measured and deliberate strategy, focusing on incremental steps and addressing key factors such as data quality, governance, and talent development. By managing expectations and investing in foundational elements, organizations can increase their chances of successfully leveraging AI-at-Scale.