Abhay Gupta is cofounder and CEO of Bidgely, evolving energy analytics for utilities with the power of data and artificial intelligence.
The utility industry has long favored deliberate pacing, which is a necessary by-product of managing mission-critical infrastructure across multi-decade asset life cycles. Yet, surging power demand from data centers and electrification now demands rapid technological evolution from a sector built on long-range planning. Utility C-suites, in response, are elevating artificial intelligence (AI) from an experimental novelty to a top strategic priority.
But without an industry-standard playbook for enterprise AI, even the most successful experiments get trapped in pilot purgatory. Gartner IT research reveals that despite 94% of power and utility CIOs planning to expand AI budgets, data readiness issues stall or kill over 60% of enterprise projects before commercial deployment. MIT’s NANDA Initiative similarly shows that up to 95% of generative AI pilots fail to deliver measurable impact due to decentralized decision-making and fragmented grid data. The bottleneck isn’t necessarily a lack of algorithms—more often, it’s the absence of specialized talent, governance frameworks and organizational readiness.
To break this gridlock, more C-suites are establishing AI centers of excellence (CoE) as their internal brain trust to set standards, upskill staff and drive enterprise adoption.
Balancing Partner Speed With Internal Ownership
To build effective AI CoEs, leadership frequently turns to global system integrators like Accenture, Deloitte, EY and Wipro. However, outsourcing this architecture can create hidden long-term risks. Because consultancy business models naturally favor continuous engagement, utilities without a clear transition strategy may be prone to high operational costs, solutions detached from grid-specific realities and little internal intellectual property.
To manage this, utility leaders can treat CoEs as cross-functional innovation hubs rather than standard IT governance bodies. Where a standard IT committee acts as a policy gatekeeper, an innovation hub focuses on operational impact.
While team size will vary based on a utility’s goal, keep in mind that a team with too few members will likely lack the multidisciplinary range needed to scale beyond single pilots. Conversely, too many members can slow down centralized decision-making and create the very bureaucratic drag the CoE was built to eliminate. Consider including:
• A dedicated CoE director to bridge executive vision and operational execution and align technical road maps with business units
• A data governance specialist to ensure model transparency, auditability and data privacy across both IT and OT environments
• Two to four lead AI architects and data engineers to manage models and infrastructure
• Two to four subject-matter experts drawn directly from substation automation, transmission and distribution, or field operations
• Two to four workforce liaisons who ensure tools are integrated into daily operations
To keep vendors accountable and protect long-term independence when building this team, utility leaders should also ask three fundamental questions up front:
1. What specific platform IP and operational models will our internal team own outright on day one, and what is your 12-month road map for training our staff to take over model governance?
2. How does this platform integrate directly into our existing cloud data environment without relying on proprietary vendor wrappers or custom connectors that lock us into your ecosystem?
3. How will this solution turn valuable data, like raw meter data, into automated, actionable workflows for our customer service representatives, grid planners, marketers, field technicians, etc., and how are we measuring user adoption?
Talent, Mindsets And Change Management
Beyond vendor management, culture and change management represent notable failure points for enterprise AI deployments. Boston Consulting Group (BCG) also revealed that 70% of an AI transformation’s value is driven by workforce changes, talent and process redesign, whereas technology infrastructure accounts for 20% and the underlying algorithms just 10%.
Regulated utilities also face acute talent friction. Competing on base salary alone against major tech firms for scarce data scientists is a losing battle. Combined with an operational culture built around risk mitigation—where engineers naturally view black-box decision-making with skepticism—the barrier to scaling AI becomes cultural rather than technical.
Resolving this friction requires utility leaders to treat AI adoption as an operational transformation rather than a software deployment. Structuring a CoE that pairs cross-functional utility domain experts directly with data engineering talent lets utilities turn black-box skepticism into algorithmic transparency and ensure AI solutions are built with front-line operators rather than imposed upon them.
What ‘Good’ Looks Like In Practice
Begin by scoping a high-impact anchor use case. It’s important to resist the urge to boil the ocean with too many disparate initiatives. Instead, select a narrow, high-ROI operational challenge, such as predicting transformer degradation or detecting unannounced EV chargers. By restricting data ingestion exclusively to the inputs required for that specific problem, teams can realistically deliver measurable value within their desired time frame.
With the first use case underway, the CoE can establish proper data governance. This could include defining protocols for data quality, access controls, transparency, auditability and model validation. Establishing this during the first build helps create a reusable compliance framework that dramatically accelerates the approval process for every subsequent AI initiative.
Next, define metrics that drive business unit ownership. These metrics should align directly with KPIs, such as average handle-time reductions, first-contact resolution rates, SAIDI impact and avoided peak power purchasing costs, rather than just technical model performance.
Keep in mind that AI models are most valuable when outputs flow into existing enterprise platforms, like feeding appliance-level load profiles straight into CRM, ADMS or DERMS systems. When usage analytics identify an aging HVAC unit or a new EV charger, the system should automatically trigger actionable workflows for customer service representatives or field technicians.
Escaping AI Pilot Purgatory
Ultimately, escaping pilot purgatory and supporting long-term scale will require a clear handoff from the CoE to the core business unit. Once a model achieves operational stability, ongoing monitoring and day-to-day business ownership can transfer to the front-line team while the CoE retains light central oversight. Project funding then shifts from central IT innovation grants directly into operational business unit line items, cementing AI as a permanent operational capability.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?