The 365 Leadership Summit, gathering in Denver this September, is sharpening its focus on a persistent and costly corporate habit: purchasing AI software that nobody uses. With Microsoft Copilot and agentic AI poised to reshape enterprise workflows, the risk of "shelfware"—paid licenses that idly drain IT budgets—has become a central concern for CIOs and IT leaders. The two-day event at The Ritz-Carlton will dissect why so many AI initiatives fail to deliver promised productivity gains, despite a market that industry voices have pegged at $600 billion.
Prachi Mishra, an AI Business Solutions Engineer at Microsoft, and Geoff Ables, a Microsoft MVP and CEO of C5 Insight, set the tone in a preview podcast. They argue that the gulf between AI hype and return on investment isn't a technology problem—it's a leadership and execution gap. Their message: buying Copilot licenses is the easy part; making them indispensable to daily work requires a deliberate, human-centric adoption strategy.
The Hype versus the Hard Work
The AI gold rush has investors and executives betting big. Comments from a senior Nvidia executive ignited conversations around a $600 billion addressable market spanning chips, generative AI software, and enterprise platforms. Meanwhile, surveys reveal that many CEOs believe AI can automate a substantial portion of their routine tasks, from data analysis to scheduling. Yet these projections gloss over a stubborn reality: two-thirds of software licenses in large organizations go unused, according to multiple industry studies. The same pattern could easily engulf Copilot seats—thousands of dollars per user vanishing into thin air.
"We have to stop buying tools and then figuring out the problem later," Ables warns. "That's how you end up with shelfware. The organizations that get real value start with a defined business objective and then pilot ruthlessly."
The True Cost of Shelfware
Unused software is a recurring financial hemorrhage. One study from 1E found that the average enterprise wastes $7.4 million per year on redundant or underused applications. In the context of AI, the damage goes beyond dollars: shelfware erodes organizational trust in future technology investments and demoralizes teams who see no benefit from new tools. Mishra admitted that even inside Microsoft, early Copilot adoption required constant coaching and daily practice. "It's a muscle you have to build," she said. "If you don't invest in change management, the tool sits on the shelf."
To counter this, Ables recommends a procurement prerequisite: no new license purchase without an adoption plan, clear KPIs, and renewal-triggered reclamation reviews. This shifts the conversation from "how many seats?" to "how will we change work?"
Why Adoption Engineering Matters More Than the Technology
The summit's emphasis on "customer zero"—using Microsoft's own internal rollout as a blueprint—highlights that even the vendor learned hard lessons. Microsoft's public case studies document phased deployments, champion networks, and Copilot dashboards to track engagement. For Windows-centric IT leaders, the Copilot Dashboard in Microsoft 365 admin center provides telemetry on active usage, feature adoption, and time saved—metrics that go beyond vanity counts.
"You can't just email people a quick start guide," Mishra stressed. "You need role-based workshops, peer coaching, and daily prompts that turn novelty into habit." A practical checklist emerged from the discussion:
- Identify high-value scenarios (e.g., sales proposal generation, customer service escalation) where Copilot directly impacts revenue or efficiency.
- Run a 90-day pilot with baseline KPIs, then compare pre- and post-adoption metrics.
- Arm champions with templated prompts and feedback loops to iterate quickly.
- Scale only after playbooks are validated.
Governance Gaps That Derail AI Projects
Without governance, Copilot and AI agents become a security and compliance nightmare. The preview podcast stressed that identity and access controls must be layered in from the start. For Azure AD/Entra ID environments, that means least-privilege service principals for agents, conditional access policies that restrict sensitive data, and audit trails that capture every agent action.
"Governance isn't a roadblock; it's the rails that let you accelerate," said Ables. He advises building a lightweight Center of Excellence (CoE) that provides reusable security templates, connectors, and compliance guardrails. This CoE becomes the enforcer of data minimization—using synthetic or anonymized datasets for early-stage agent training—and ensures that human-in-the-loop checkpoints exist for high-risk decisions like financial approvals or legal filings.
AI Agents: From Buzzword to Business Process
One of the summit's headliner topics is the rise of AI agents—semi-autonomous digital coworkers that can orchestrate multi-step tasks across Dynamics 365 and Microsoft 365. Mishra likened them to "an autonomous PhD student: they can research, draft, and iterate, but you still need to validate their work." This shift demands new leadership skills.
Managers must learn to delegate structured tasks to agents: define clear objectives, supply context, set constraints, and then inspect outputs. For IT teams, this means version-controlling agent prompts, monitoring performance, and establishing rollback protocols. The summit will feature case studies where agents reduced EDI transaction errors or automated invoice processing, but the hosts cautioned that such gains require rigorous testing and exception handling.
A Playbook for Leaders: Nine Steps to Measurable ROI
Drawing from the podcast and event preview, a actionable sequence emerged for leaders determined to avoid the shelfware trap:
- Define a single, prioritized business outcome—not "roll out Copilot," but "reduce contract review cycle time by 30%."
- Run a 90-day pilot linked to that outcome with explicit KPIs.
- Adopt a customer-zero mindset—test governance and UX on an internal group first.
- Stand up a nimble CoE that supplies templates, security policies, and reusable integrations.
- Build an adoption program with champions, localized training, and daily practice prompts.
- Instrument both technical and business metrics—usage dashboards and outcome KPIs.
- Reclaim unused licenses during each renewal cycle.
- Design human-in-the-loop checkpoints for all high-risk agent actions.
- Iterate relentlessly—convert pilot learnings into scalable playbooks.
The Summit Agenda: What to Expect
Scheduled for September 18-19 at The Ritz-Carlton, Denver, the 365 Leadership Summit promises peer-led sessions that move beyond vendor pitches. Confirmed tracks include Copilot adoption case studies from organizations that have moved the needle, change-leadership workshops, and deep dives on AI governance. The curated attendee list targets C-suite and senior IT leaders responsible for Microsoft 365 and Dynamics 365 environments.
"This isn't a tech conference; it's a transformation conference," an organizer noted. "Every session answers the question: how did you get real people to use this every day?"
Critical Analysis: Where the Approach Falls Short
While the summit's focus on behavior change and governance is refreshing, several risks remain unaddressed. First, vendor narratives—like the $600 billion market estimate—can seduce executives into buying forward without rigorous internal validation. Organizations must resist the "oh wow, we need to keep up" reflex and instead anchor investment in their own micro-wins.
Second, anecdotal productivity claims (one client allegedly cut a worker's workload by 30%) are illustrative at best. Without transparent before-and-after data and methodology, such stories can mislead. Leaders should demand hard evidence from peers.
Third, the CoE model can become a bottleneck if overbuilt. "A CoE should be a capability engine, not a command center," Ables cautioned. "Provide guardrails, then let teams run."
For Windows-Centric IT Leaders: Specific Actions
- Leverage native tools: The Copilot Dashboard and Microsoft 365 admin telemetry already exist—use them to spot inactive users and target enablement. Set up alerts for low adoption across your tenant.
- Harden agent identities: Enforce conditional access for any service accounts, require multi-factor authentication where possible, and use Privileged Identity Management to time-box elevated permissions.
- Test with clean data: During early agent development, use synthetic or anonymized datasets to avoid exposing sensitive customer information.
- Tie training to outcomes: Rather than reward raw click counts, incentivize managers on measurable business KPIs that Copilot influences—such as faster resolution times or reduced manual data entry.
Conclusion
The conversation at the 365 Leadership Summit is a timely antidote to AI hype fatigue. Generative AI and Copilot hold enormous potential, but they are not magic. They require a deliberate fusion of executive vision and gritty execution—the kind that turns skeptical employees into daily users and transforms pilot projects into enterprise standards. The playbook is clear: start with the problem, pilot mercilessly, govern by design, and kill any license that doesn't pull its weight. For IT leaders willing to do the unspectacular work, the summit offers a roadmap to turn AI investment from a cost center into a competitive advantage.