Step 2: Launch a pilot

Successful pilots deliver quick wins while building organizational capability. Choose projects carefully, showcase cross-functional potential, and learn rigorously from every experiment.

Choose initial projects carefully

We suggest selecting one or two initial projects that demonstrate AI's value while building organizational capability. These pilots should span different business functions to showcase AI's versatility. For example, building an internal chatbot might answer common IT and support questions, reduce help desk burden, or deploying a coding agent like Claude Code could accelerate software development while improving code quality across multiple business units.

When choosing your pilot project, ensure they abide by these two critical guidelines:

• Evaluate potential pilots against clear ROI: Measure the potential business impact in specific terms like cost reduction or time savings, technical feasibility, user receptivity or process disruption, and resource requirements including both technology costs and implementation effort.
• Prioritize projects where failure creates minimal business disruption: avoid selecting customer-facing applications or mission-critical processes for initial pilots regardless of their potential value.

Once you've aligned on your chosen pilot, structure pilot teams with clear roles and dedicated time commitments rather than treating AI as an additional responsibility for already-stretched resources. Assign a pilot lead who owns outcomes and coordinates across functions, technical resources who handle implementation and troubleshooting, business users who provide domain expertise and test real-world scenarios, and executive sponsors who remove organizational barriers and maintain stakeholder alignment.

Showcase cross-functional potential

Design pilots that reveal possibilities beyond their immediate scope. When legal teams see marketing's success with content generation, they begin imagining contract automation applications. When engineering demonstrates code review acceleration, finance envisions automated report generation. This cross-pollination of ideas drives organic expansion beyond initial implementation plans.

Create structured opportunities for cross-functional learning that accelerate insight transfer across the organization, such as a monthly "AI Showcase" where pilot teams highlight their projects. These sessions should include live demonstrations of AI capabilities, before-and-after comparisons of work processes, and open discussion about challenges encountered and solutions discovered.

Navigate common pilot challenges

Even well-designed pilots encounter obstacles. Anticipating common challenges and preparing response strategies prevents minor issues from derailing valuable initiatives. We outline several of the most prevalent, below.

• User resistance and adoption gaps emerge when team members feel threatened by AI or find new workflows disruptive. Address this through early involvement in pilot design, hands-on training that builds confidence rather than just explaining features, and celebrating early adopters who become internal advocates. When a team member discovers a better way to use the technology, amplify their insight across the organization.
• Data quality and availability issues often surface mid-pilot when teams discover their existing data isn't structured for AI applications. Rather than pausing the pilot, implement pragmatic workarounds such as starting with better-quality data subsets while simultaneously improving broader data practices, or using AI to help clean and structure the problematic data itself.
• Integration complexity with legacy systems frequently exceeds initial estimates. Build in technical flexibility by starting with manual handoffs between AI and existing systems if necessary, then automating connections as you prove value. Perfect integration shouldn't block initial learning.
• Scope creep threatens pilot timelines when teams discover additional applications mid-implementation. Maintain discipline by documenting new ideas in a "next phase" backlog rather than expanding current scope. This preserves pilot timeline integrity while ensuring good ideas aren't lost.

Conduct pilot post-mortems

Once each pilot concludes, the real work of learning begins, moving beyond surface metrics to understand the deeper story of what unfolded and why.

Start by examining the numbers but also evaluate quantitative metrics. In fact, these "anecdotal" proof points are often more informative, for instance, moments when users discovered unexpected benefits, the friction points that emerged in practice, the workarounds teams invented when the technology didn't quite fit their workflow.

On the technical side, probe system reliability, integration snags, data quality surprises, and infrastructure needs that only became apparent under real-world conditions. Understanding user adoption requires detective work—why did some teams embrace the technology while others quietly resisted? What practical barriers emerged?