Top of mind for organizations in the midst of this AI boom is how to leverage AI to support their business goals without wasting time on tools or projects that deliver more hype than substantive return. Gartner’s 2023 report Become an AI-First Organization: 5 Critical AI Adoption Phases suggests that while CIOs and other tech executives are brimming with enthusiasm for AI initiatives, actual deployment rates lag.

Conversations with Stack Overflow for Teams customers, not to mention the volume of discussion on AI topics across our Stack Exchange network, reflect something similar: huge enthusiasm for the business potential of AI, but significant uncertainty about how to undertake the practical steps of implementing an AI project—or even what kind of AI project would best support business goals.

Everybody wants to be an AI-first organization, but not everyone is quite sure what that means or what it requires.

At Stack Overflow, it’s our view that fostering a knowledgeable community of practice around AI is how you become an AI-first organization. That means that AI is productively incorporated into every facet of the business ecosystem, while AI experts, best practices, and technologies are made available for new projects. We also know that the bedrock of a great AI system is high-quality data. That’s why investing in a knowledge base like Stack Overflow for Teams is crucial in building your knowledge foundation.

In the above-mentioned report, Gartner stresses the importance of forming a community of practice (COP) to enable knowledge-sharing and innovation around AI. The report also suggests well-defined adoption phases and indicators for AI success, with an eye toward allowing organizations to assess the capabilities and performance of AI tools, measure their overall business impact, manage costs, and mitigate the inevitable risks. In this article, we’ll explore more of the Gartner report and show how Stack Overflow for Teams can act as a strategic partner in your AI initiatives.

AI initiatives generate enthusiasm—but many don’t make it to production

AI implementations are challenging on several levels. Instead of following adoption phases familiar to any organization that’s rolled out a new tool or technology, AI implementations are unfolding unpredictably, forcing teams to learn the hard way what works and what doesn’t. While Gartner organizes their five phases into an orderly chronology, AI adoptions in reality aren’t always unidirectional. You might move forward, backward, or linger in one phase.

When it comes to AI projects, companies frequently start out with unrealistic expectations, from how long the project will take to deploy to what it will cost to what it will actually accomplish. Software and technology companies are moving rapidly, whereas more highly regulated industries like healthcare, finance, and government are inching forward, reluctant to move fast and break things in such a new and untested space.

While your adoption may not unfold in perfect alignment with Gartner’s phases, defining and carefully monitoring the phases of AI adoption will help you sidestep major pitfalls and increase the likelihood of a successful implementation.

A successful AI adoption requires proof of value for the AI and a level of business trust. Bear in mind that your investment in people, process, and technology will likely exceed your AI returns at first, so setting expectations across the organization, board, and senior leadership is important.

A successful AI adoption has 5 critical phases

To maximize the success of an AI implementation and avoid snags, Gartner suggests that companies check their deployment progress against five phases of AI adoption: planning, experimentation, stabilization, expansion, and leadership.

Planning: In the initial phase, AI teams at your organization socialize the AI ideas to surface the most promising use cases and identify a business champion to act on and benefit from the first AI solutions. Based on our own experience incorporating AI into our businesses processes, combined with what we’ve learned from conversations with industry leaders and our own customers, we recommend using the planning phase to:

• Consider what data you have that would be useful.
• Decide whether you want to build your own AI mode, use an open-source one, buy one from a third party, or simply interact with an API (check out our other article where we outline some options!).
• Determine whether your use cases are customer-facing or internal.

Experimentation: To evolve from planning to experimentation, it’s best to have somewhere between three and six use cases, ideally centered around a single theme and a small cluster of stakeholders. Here are some options for experimentation:

• Explore AI playgrounds/sandboxes available online.
• Run a small AI model locally, and scale it up if the results are promising.
• Create a simple retrieval-augmented generation (RAG) model as an initial proof of concept.

Stabilization: In the stabilization phase, a multidisciplinary AI team has established a virtual center of excellence (COE), or at least a community of practice, to share best practices and facilitate training/upskilling for core AI roles. In this phase of your AI adoption, you might:

• Decide whether you’ll do your own embedding or rely on a vector database.
• Consider the ongoing costs of using an API or hosting your own model.
• Be aware that even after training, prompt engineering and reinforcement learning from human feedback (RLHF) may still be required to produce high-quality results.

Expansion: As organizations gain confidence and experience, they’re able to start applying AI to higher-risk, higher-return projects. Successful line-of-business projects scale to the entire organization, new AI projects multiply and move faster, and new roles appear. If you’re moving to a pilot program or a small public beta, we would recommend:

• Carefully monitoring inference costs and latency. As with cloud computing, AI budgets can change significantly based on usage, and even a small amount of latency can discourage users from coming back.
• Creating a feedback loop so users can rate the model’s output or provide feedback to improve the system over time.

Leadership: Evolving from the expansion phase into the leadership phase requires demonstrating success with high-risk/high-return AI projects, expanding your data sources, and cultivating a top-down AI-first culture that encourages rapid innovation.

Concrete examples of successful AI projects are still scarce, which means that organizations that do it right can emerge as thought leaders, like the first companies built on the SaaS model or cloud computing. If you can bring tangible examples and strong results to public attention, you may be rewarded with inbound interest, media opportunities, and new markets or partners.

How Stack Overflow for Teams can help

As you advance along the AI adoption curve, Gartner recommends that you invest in human, process, and technology frameworks that ease the transition between phases, such as a funded community of excellence (COE), practical knowledge of cloud AI APIs, and executive awareness that leads to knowledge and skills sharing. Stack Overflow for Teams is perfectly positioned to be a strategic partner in these efforts.

It’s purpose-built as a place for technologists to collaborate and share learnings across all critical phases of AI adoption and deployment. Stack Overflow for Teams can help:

• Capture and document knowledge and best practices, allowing you to develop a COE for AI.
• Support innovation, rapid training, and upskilling in AI with access to experts and high-quality, community-vetted information.
• Enable collaboration across multidisciplinary roles and teams.
• Plan and implement enterprise-wide AI literacy education.

Learn more about how your organization can successfully leverage AI with our Industry Guide to AI.