Digital Economy Dispatch #176 -- Agile Lessons for Delivering AI-at-Scale

Digital Economy Dispatch #176 -- Agile Lessons for Delivering AI-at-Scale
24th March 2024

There are now widespread discussions about the opportunities and challenges of adopting AI broadly across large established organizations (LEOs) in the public and private sectors. Much is expected from this wave of change. Along with increasing hype from industry commentators about AI’s disruptive impact, from many quarters a high level of expectations is being set about how fast and furious its effects will be felt by workers across these organizations.

As a result, pressure is building to meet ambitious goals for the return on investment (ROI) in AI. For example, reports from the UK government point to major productivity boots that could automate up to 84% of repetitive service transactions, and lead to two-thirds of civil service jobs being affected over the next 15 years.

Whether these ambitions will be realised or not remains to be seen. However, it is clear that progress will depend on answering critical questions about how to scale AI adoption in practice: What is the best way to covert lessons from experiments and pilot studies with AI into organization-wide change? What barriers must be overcome to introduce AI into established working practices? How can we turn early wins with AI into substantial measurable successes? And so on.

These conversations have a very familiar ring to them. For those of us involved in digital transformation efforts over the past decade, they bring back many memories of addressing similar challenges when moving to “agile at scale”. And it appears that those participating in the current wave of AI adoption are wearing similar battle scars. Perhaps by reviewing the lessons from previous efforts to scale Agile we can provide important pointers to help those engaged in today’s AI struggles.

Reviewing Agile at Scale Experiences: Lessons for AI adoption

Successfully deploying AI across a complex enterprise poses significant challenges. All forms of substantive change face resistance. For LEOs pursuing highly disruptive digital changes, the barriers can be substantial. What can we learn from previous largescale digital change efforts?  

One of the most interesting places to explore is to consider the way many organizations have adopted more agile ways or working. Particularly in areas of software and systems delivery, the Agile approach has gained significant foothold over the past 20 years. But not without having to address many concerns about its scope, range, impact and applicability across diverse, complex environments.  By examining the experiences of organizations adopting Agile at scale, we can glean valuable insights that can be applied to navigating the potential pitfalls of large-scale AI adoption.

Agile, a methodology emphasizing iterative development and rapid feedback loops, has become a cornerstone of modern software development. While the core principles are well-established, achieving "Agile at Scale" presents a different hurdle. Similar to AI, scaling Agile requires not just adopting new technologies, but also optimizing processes for collaboration across diverse stakeholder groups, revising decision making practices, and adjusting management structures to support the new ways of working.

While there are many aspects to how agile approaches became widely used, based on my own experience over the last 20 years in enterprise software delivery, I have found that Agile adoption in LEOs often started with uncoordinated individuals and small developer teams making rapid advances in short, focused delivery efforts. Subsequent efforts to scale these successes across the organization required overcoming several key hurdles:

  • Resistance to Change: Traditional, plan-driven mindsets can clash with Agile's emphasis on dynamic planning and rapid iterations. This created a "progressives vs. traditionalists" divide, hindering widespread adoption.

  • Misaligned Support Teams: Resource managers, financial teams, and other supporting functions struggled to adapt their practices to Agile's less rigid planning and progress tracking. They perceived it as disruptive to their established workflows.

  • Middle Management Challenges: Project managers, analysts, and other mid-level roles were found to feel threatened by the potential loss of control associated with empowered Agile teams.

These kinds of challenges bear many similarities to reports starting to emerge from corporate efforts to adopt AI-at-scale. Recent surveys such as the Digital Leaders Attitudes to AI survey that took place in December 2023 indicate AI is a major topic among digital leaders, with most survey respondents reporting weekly discussions and interactions with AI, and over a third using it daily. However, while awareness of AI is high, many surveyed organizations haven't identified practical uses for it or assessed its broader business impact. This lack of clear strategy extends to generative AI, with most organizations lacking policies to govern its use.

Furthermore, It is also clear that implementing AI-at-scale faces hurdles common to digital transformations in large organizations. While ROI concerns exist (almost half unsure of positive impact), bigger issues lie in talent acquisition/retention and integrating AI into existing workflows (both cited by over half as significant barriers).

Agile-at-Scale by Example: Bosch and USAA

From a practical perspective, looking at the experiences of LEOs and their Agile adoption journeys allows us to glean further insights that are useful as we consider AI-at-scale. In their review of scaling Agile adoption, Darrel Rigby, Jeff Sutherland, and Andy Noble point to several important characteristics of successful Agile adoption efforts in large organizations. By looking at experiences of several organizations, they focus on 3 key levers in scaling Agile effectively:

  1. Leading Agile by Being Agile. Agile teams are self-organizing with close customer connection, allowing faster innovation and freeing up senior leaders for strategic work. To lead agile transformations effectively, senior leaders should act as an agile team themselves, focusing on understanding customer needs and removing roadblocks.

  2. Getting Agile Rolling. Large companies must implement substantial changes such as Agile in phases. They start small, measure the impact, and then decide whether to expand based on a cost-benefit analysis focused on value creation and organizational challenges. This allows for adjustments and avoids overwhelming changes.

  3. Building Agility Across the Business. Creating isolated agile teams is just one piece of the puzzle. Successful agile companies also focus on changing how these teams work with traditional structures to avoid slowdowns and ensure innovations are implemented. This means continuous adjustments in multiple areas such as project management, HR, contracts management, and procurement.

Two examples, drawn from their article, illustrate these concepts in practice and offer important insights into ways that LEOs can face the challenges of achieving AI-at-scale.

One organization they highlight is Bosch, a large global product and technology supplier with over 400,000 associates, Bosch initially struggled to implement Agile because they tried a dual-organization approach where some departments were Agile while others remained traditional. This created conflict and hampered the overall transformation.

Later, Bosch formed a steering committee with members from the board of management. This committee acted more like an Agile team itself, with members working collaboratively to remove roadblocks and solve problems. They also created a taxonomy to identify all the potential Agile teams across the company. Bosch’s experience highlights the importance of senior leadership buy-in and collaboration when adopting Agile at scale. It emphasizes that Agile should not be implemented in a way that creates a two-tiered system within the organization.

A second example they describe is USAA, a large US-based banking and insurance organization providing services exclusively to military, veterans, and their families. USAA organizes their large workforce into a taxonomy of several hundred teams. While at first glance this may seem overwhelming, this consistent approach across the whole organization helps them avoid confusion and finger-pointing by clearly defining the landscape of teams and their activities. Then, when new projects are defined, project leaders can readily determine which team is responsible for each part of the customer experience, across all channels (phone, website, app).

This is especially important because USAA focuses on customer journeys that may cross several traditional departmental lines. The taxonomy connects agile teams to the people accountable for results, ensuring everyone is working together to deliver a seamless omnichannel experience delivering value to the organization and the client.

Lessons Learned from Agile at Scale for Successful AI Adoption

The experience of adopting Agile at scale offers valuable insights for organizations navigating the challenges of large-scale AI deployment. Both initiatives require not just implementing new technologies or practices, but also fostering a cultural shift that embraces continuous learning, adaptation, and collaboration.

One of the key takeaways is the importance of focusing on empowerment rather than disruption. Similar to concerns around Agile replacing traditional project management roles, fears of AI displacing human workers can hinder adoption. By framing AI as a tool to enhance existing functions and empower stakeholders to understand its contribution, organizations can foster a more positive and collaborative environment.

Another crucial lesson is the value of a phased implementation approach. Just as Agile adoption often starts with successful pilot projects, AI initiatives should begin with targeted use cases that showcase the technology's value proposition. This allows for incremental scaling, continuous improvement, and adaptation of existing workflows to address concerns and ensure a smooth integration of AI.

Building bridges between technical teams and supporting functions is another critical element. The challenges faced when attempting Agile at scale, where resource managers and finance teams struggled to adapt to Agile's flexible planning, mirror potential hurdles with AI adoption. Fostering open communication and ensuring a clear understanding of how AI disrupts existing processes and roles is essential for successful integration.

By applying these lessons learned from Agile adoption, organizations can be better prepared for the complexities of large-scale AI adoption. A focus on collaboration, empowerment, and a phased approach are three of many lessons we can learn that will help unlock the full potential of AI while minimizing disruption and maximizing the value it delivers to the organization.