AI from exploring to transforming: Introducing the AI Maturity Framework
Karthik Ramakrishnan Karthik Ramakrishnan
November 5 7 min

AI from exploring to transforming: Introducing the AI Maturity Framework

This article was co-authored by Karthik Ramakrishnan, Head of Advisory & AI Enablement, and Cyrielle Chantry, Lead Strategist.

Though many enterprises have started AI projects, results are falling short. Gartner researchers project 4 out of 5 AI projects will fail, while practitioners at McKinsey found fewer than 1 in 10 organizations are ready to put AI into production. The successful AI products most of us can point to are helping consumers, not CEOs.

A commonly cited roadblock for enterprises to execute on AI is missing or unlabelled data. At other times, it might be skills gaps of technical and business teams, the alignment of stakeholders, technology maturity, and so on.

What we’re finding in practice is that it’s rarely just one challenge that holds back AI. Operationalizing AI requires maturity across multiple dimensions, including strategy, data, technology, people and governance. Each dimension needs to be harmonized and fit for purpose.

While solutions are now emerging for individual challenges in the AI journey, it’s still not widely understood how the journey unfolds across different stages of maturity, or what it takes to get from one stage to the next.

We’re closing this gap now. Here are the five stages of AI maturity that we’ve identified by combining what we’ve learned from building enterprise AI products and tools, as well as advising dozens of clients on their own unique AI efforts.

Five stages of organizational maturity for AI

Organizational maturity for AI advances through five stages, from exploring what AI can and cannot do, to transforming the whole enterprise to work smarter with AI. At each stage, the organization’s ability to operationalize AI becomes more finely tuned and impactful.

The AI maturity framework blueprint.

In practice, advancing from stage to stage requires organizations to level up across all five dimensions of Strategy, Data, Technology, People, and Governance. Each dimension has significant depth.

In the following sections, we focus on the relationship between just one dimension — People — and the organization’s ability to apply AI in practice.

Stage 1: Exploring

Organizations enter the Exploring stage of their AI journey when they shift from general awareness of the technology to targeted questions about what problems or opportunities it can help address.

Companies in this stage might range from having zero budget for AI to having a chartered role dedicated to adopting it. In our experience, what unifies Exploring companies is their common lack of experience judging a good AI opportunity from a bad one. For example, data scientists can run Tensorflow with GPU acceleration using a toolkit like NVIDIA CUDA® Deep Neural Network library (cuDNN), but don't know where to go next after tutorials. Business leaders aren't sure how to help.

Techniques for closing this gap vary for different teams and organizational structures. Business as well as technical leaders usually need help separating hype from reality for AI techniques like deep learning, reinforcement learning and transfer learning. Data science teams might need less help understanding these techniques at a technical level, but still need to discover what cooperation is needed from the business to deliver models in production.

Progress in the Exploring stage tends to be driven by ambitious individuals or teams who focus on building informed interest and buy-in.

Stage 2: Experimenting

In the Experimenting stage, organizations get more intentional about developing a detailed hypothesis of where and how AI should be applied.

Since organizations in this stage are still figuring out what works and what doesn’t, we often see a concentration of AI efforts with a single team that’s able to operate independently. These teams usually can’t yet perform at the scale of a Center of Excellence (CoE), but they should be starting to work across functional boundaries. For example, they might have early conversations about interconnecting data across the organization or interview employees to validate system designs.

As experiments unfold, the organization hones in on which AI opportunities are desirable, feasible and viable. For example, data scientists and developers start using cloud infrastructure to share know-how and results — as well as to leverage GPU power beyond the confines of their laptops.

Small wins naturally emerge in this stage as side projects become Proofs of Concept (POCs) or the organization starts testing commercial tools or products. Teams that make the swiftest progress are careful to maintain focus on learning how to deliver AI models in production — not just delivering small wins for their immediate impact alone.

Stage 3: Formalizing

Organizations enter the Formalizing stage after successfully deploying their first AI projects into production.

These first projects may be relatively small, running just one or a small handful of AI models each, yet in order for them to move forward, they must still meet a long list of demands.

Models must be performant as well as safe, while workflows and job roles must have been updated to incorporate human judgment “in the loop” to monitor and control performance over time. The AI system needs to be supported with relevant data flows to operate in real time. For each AI model put into production, business and technical teams must be aligned on vision, goals and budgets.

In the Formalizing stage, these needs are still being met ad hoc compared to later stages. For example, data may not be delivered via a specialized data lake, but by bespoke system integrations. Similarly, the organization will have started hiring and training for AI talent, but may not have a company-wide plan for updating roles.

Stage 4: Optimizing

Organizations transition to Optimizing for AI as they absorb the changes necessary to reliably select, deliver and manage AI projects for positive ROI.

More than in previous stages, organizations in the Optimizing stage are using AI to improve internal operations as well as market offerings, such as new products and business models. It also necessitates new infrastructure, whether from traditional cloud providers or in a data center with tools like NVIDIA DGX. Teams need help managing end-to-end development, delivery and optimization of AI models over the long-run.

This increasing scope of impact puts more pressure on organizations to enhance AI governance. As the collection of AI models deployed in production increases, so also does the complexity of interactions between these models, further adding to risks. Tools and processes to effectively monitor and control the risks from AI must keep pace, while techniques like explainable AI and adversarial defenses for AI grow more valuable. These commitments are worthwhile because they’re the ticket price of revealing the benefits and potential of interconnected systems for exponential learning.

Across the organization, AI challenges and opportunities are being experienced by most employees rather than only dedicated teams. Organizations in this stage especially need to have a plan for leading employees in this new normal, such as by proactively transitioning jobs and skills.

Stage 5: Transforming

The Transforming stage of AI maturity is where the promise of AI for business and society meet for the biggest impact. It's our north star for what we're trying to help organizations achieve with AI. Few organizations have reached this stage in practice, and it’s not clear if any are yet delivering on its full potential.

At this stage, companies have transformed to work smarter with AI. They operate safer and cleaner than ever before, while creating more value for society and tackling problems that couldn’t be solved before. Industry analysts and researchers are still debating what this looks like in practice. For example, organizations that transform with AI are expected to leverage systems of intelligence to augment human intelligence and unleash collective intelligence between humans and machines — but each of these terms is still being actively defined.

More practically, we expect companies that are successfully Transforming with AI to be spending more time advancing the state of the art, both technically and in the arena of AI ethics.

Working smarter, together

We’re excited to share this vision of the five stage AI maturity — and we invite you now to help us make it even better.

AI maturity framework survey

Please take our survey to learn more about your own organization’s AI maturity and to help us define AI maturity in industry.

Take the Survey

Of course, if you want help operationalizing AI in your business or would like to learn more about AI maturity, please get in touch!