Guide 35 mins

Portfolio-Wide AI Operating Model for Energy

Build a scalable AI operating model across energy portfolio companies. Diligence, capability rollout, governance, and exit positioning benchmarks.

The PADISO Team ·2026-06-07

Introduction: Why Energy Portfolios Need a Unified AI Operating Model
AI Diligence Framework for Energy Acquisitions
Assessing Current-State Technology and Talent Across Holdings
Building the Portfolio AI Capability Layer
Governance, Risk, and Compliance at Scale
Value-Creation Playbook: AI Rollout Across Holdings
Measuring and Benchmarking AI Impact
Exit Positioning and Diligence-Ready Tech Stories
Regional Execution: Hubs and Fractional Leadership
Summary and Next Steps

Introduction: Why Energy Portfolios Need a Unified AI Operating Model

Private equity firms managing energy portfolios face a unique challenge: their holdings span generation, distribution, retail, and services—each with different operational maturity, technology debt, and regulatory complexity. Yet all of them sit at the intersection of massive opportunity and existential risk when it comes to artificial intelligence.

The opportunity is real. According to the International Energy Agency analysis of AI and the energy sector, AI is already driving material value across electricity systems, energy efficiency, demand forecasting, and grid optimisation. The U.S. Department of Energy’s report on opportunities for a modern grid and clean energy economy documents concrete use cases: AI-driven grid planning, real-time forecasting, predictive maintenance, and resilience modelling are moving from pilot to production across leading utilities.

The risk is equally clear. Uncoordinated AI adoption across a portfolio creates fragmented data architectures, duplicated effort, inconsistent governance, and exposure to regulatory and reputational damage. A single portfolio company shipping an unaudited model into production, or failing to document AI decision-making in a regulated asset, can trigger cascading compliance failures and loss of stakeholder trust.

A portfolio-wide AI operating model solves both problems simultaneously. It creates a shared playbook for diligence, capability-building, and value creation that works across holdings of different sizes and maturities. It establishes clear governance and risk controls that satisfy regulators and boards. And it compresses the time from acquisition to AI-driven value creation from months to weeks.

This guide covers the practical mechanics of building and executing that operating model. We focus on outcomes—revenue uplift, cost reduction, time-to-ship, audit readiness—and avoid generic frameworks. We draw on experience working with energy operators, PE-backed portfolio companies, and platform engineering teams across North America and Australia.

AI Diligence Framework for Energy Acquisitions

The Three-Layer Diligence Stack

Traditional tech diligence looks at code quality, infrastructure, and team. AI diligence adds three critical layers: data readiness, model governance, and AI decision exposure.

Layer 1: Data Readiness

AI is useless without data. During diligence, you need to answer: What data does the target company have? Where does it live? How clean is it? Can it be moved or integrated?

Energy companies typically have data scattered across SCADA systems, historian databases, ERP systems, and spreadsheets. A generation company might have 20 years of hourly turbine telemetry, but it’s siloed in a proprietary historian with no API. A retail energy provider might have customer usage data in a data warehouse, but it’s not linked to billing or grid-edge device data.

During diligence, conduct a data inventory:

Operational Technology (OT) data: SCADA, PLC, historian systems. Document the historian vendor (Wonderware, OSIsoft PI, Ignition), data retention policy, and export capability. Test whether you can extract a month of data without breaking production.
IT data: ERP, CRM, billing, HR systems. Identify the systems of record, data quality (completeness, timeliness, accuracy), and whether they’re cloud-hosted or on-premise.
Edge and IoT: Smart meters, grid sensors, substation monitors. Understand the protocol (Modbus, OPC-UA, MQTT), data frequency, and whether the company has a centralised data lake or distributed collection.
Regulatory and audit data: Compliance logs, security events, change records. Energy companies in regulated jurisdictions must maintain audit trails; understand what’s being logged and for how long.

Score data readiness on a simple scale: Ready (accessible, documented, >80% complete), Workable (accessible but needs cleaning or integration), Blocked (siloed, proprietary, or requires vendor unlock).

Target: At least 60% of data should be Ready or Workable. If >40% is Blocked, budget 3–6 months for data infrastructure work before AI can move from pilot to production.

Layer 2: Model Governance

If the target company already uses AI or machine learning, you need to know:

What models are in production? (Forecasting, anomaly detection, classification, optimization.)
Who built them? (In-house team, vendor, consultant.)
How are they monitored? (Drift detection, performance tracking, retraining cadence.)
What happens if they fail? (Fallback to manual process, alert to operator, autonomous action.)
Who owns the model? (Data science team, ops team, vendor.)
How are they documented? (Model card, training data description, assumptions, limitations.)

Many energy companies run models with no governance at all. A wind farm might have a predictive maintenance model that drifts after 18 months because retraining is manual and sporadic. A retail provider might have a demand forecasting model that was built three years ago, never updated, and no one remembers how it works.

Score model governance:

Mature: Documented models, automated retraining, drift detection, clear ownership, audit trail.
Developing: Models exist, some monitoring, ownership is clear but documentation is sparse.
Immature: Ad-hoc models, no monitoring, unclear ownership, no documentation.

Immature governance is not a deal-killer—it’s a value-creation opportunity. But it tells you that the company is not yet operating at the maturity level required for a portfolio-wide AI programme.

Layer 3: AI Decision Exposure

This is the compliance and reputational risk layer. Ask:

Are any AI models making or advising decisions that affect customers, regulators, or safety?
Are those decisions logged, explainable, and auditable?
What happens if a model makes a bad decision? (Customer complaint, regulatory breach, safety incident.)
Has the company disclosed AI use to customers or regulators?

In energy, AI decision exposure is high. A demand response model that automatically sheds load to a customer’s facility without clear notification is a regulatory risk. A grid forecasting model that informs dispatch decisions but is not documented or validated is an operational risk. A customer churn prediction model that drives pricing decisions without transparency is a reputational risk.

Score exposure:

Low: AI is used for analytics and recommendations only; humans make all material decisions.
Medium: AI informs some operational decisions; decisions are logged and auditable.
High: AI makes autonomous or semi-autonomous decisions affecting customers or operations; documentation and auditability are unclear.

High exposure requires immediate governance investment post-acquisition. Plan for 4–8 weeks of audit-readiness work via frameworks like the NIST AI Risk Management Framework, which provides structured guidance for energy operators managing AI risk.

Diligence Checklist

Before signing, complete this checklist:

Data inventory complete; >60% of operational data is Ready or Workable.
Model governance assessment complete; no undocumented models in production.
AI decision exposure mapped; no autonomous decisions affecting customers or safety without audit trail.
Technology team interviews complete; clear understanding of AI capability, gaps, and appetite.
Regulatory and compliance status confirmed; no outstanding AI-related audit findings.
Budget estimate for AI infrastructure, governance, and capability-building (typically 3–12 months, $500K–$2M depending on portfolio size and complexity).

Assessing Current-State Technology and Talent Across Holdings

The 90-Day Tech Audit

Once you own a portfolio company, you have 90 days to conduct a thorough tech and talent audit. This is not a detailed engineering assessment—it’s a rapid scan to understand baseline maturity and identify quick wins.

Focus on five dimensions:

1. Technology Stack and Debt

Map the current architecture:

Core systems: What are the systems of record? (ERP, SCADA, historian, CRM.)
Data infrastructure: Is there a data warehouse, data lake, or just point-to-point integrations?
Cloud vs. on-premise: What’s cloud-hosted? What’s legacy on-premise?
Integration patterns: How do systems talk to each other? (APIs, ETL, batch file transfers, manual.)
Tech debt: What’s preventing velocity? (Monolithic systems, manual processes, outdated databases, poor observability.)

For energy companies, common debt includes:

Historian systems that are not cloud-integrated, making real-time analytics hard.
ERP systems (SAP, Oracle) that are tightly coupled to operations, making modernisation risky.
Siloed data: generation data in one historian, grid data in another, customer data in a third.
No observability: operations teams run on email alerts and manual log-checking.

2. Engineering Capability and Team Structure

Interview the CTO, VP Engineering, and technical leads:

Team size and composition: How many engineers? What disciplines? (Backend, frontend, data, DevOps, security.)
Hiring velocity: Can they hire? What’s the local talent market?
Retention: Are people leaving? Why?
Process maturity: Do they have CI/CD, code review, testing? Or is deployment manual and risky?
Cloud readiness: Are engineers comfortable with cloud platforms? Do they have DevOps culture?
AI readiness: Do they have data engineers? Data scientists? Have they shipped AI before?

In energy, common gaps include:

Heavy reliance on domain experts (operators, engineers) with limited software engineering discipline.
Lack of data engineering capability; data is managed by DBAs or operations teams, not engineers.
Limited cloud experience; infrastructure is on-premise or in private clouds.
No AI capability; machine learning is seen as a research activity, not a production discipline.

3. Data Maturity and Accessibility

Build on the diligence data inventory:

Can you extract data from SCADA, historians, and ERP systems without breaking production?
Is there a data warehouse or data lake? If so, how is it governed? What’s the latency (real-time, hourly, daily)?
What’s the data quality? (Completeness, timeliness, accuracy.) Run a sample data extraction and assess.
Are there data governance policies? (Data ownership, retention, access control.)
Do you have data engineers who understand the data and can build pipelines?

4. Governance, Risk, and Compliance Posture

Understand the regulatory and compliance environment:

Is the company regulated? (FERC, NERC, state PUC, other.)
What compliance frameworks apply? (SOC 2, ISO 27001, HIPAA, other.)
What’s the current compliance status? (Certified, working toward, not started.)
What’s the security posture? (Vulnerability scanning, penetration testing, incident response.)
What’s the audit history? (Any findings? Any AI-related findings.)

For energy companies, compliance is often a maturity gap. Many smaller regional providers have minimal security infrastructure and no formal compliance programme.

5. Organisational Readiness for Change

Assess cultural and organisational factors:

Is the leadership team aligned on AI and modernisation? Or is there resistance?
How fast can they move? (Decision velocity, approval processes, risk tolerance.)
Are there union or labour considerations? (In some regions, automation can trigger labour concerns.)
What’s the customer and regulator relationship like? (Supportive of innovation, or conservative.)

Output: The 90-Day Tech Scorecard

Score each holding on a simple 1–5 scale across the five dimensions. This gives you a portfolio-wide view of maturity and allows you to cluster holdings by similar challenges.

Example:

Holding	Tech Stack	Engineering	Data	Governance	Org Readiness	Overall
Gen Co A	2	2	2	1	2	1.8
Retail B	3	3	3	2	3	2.8
Services C	2	3	2	3	4	2.8
Distrib D	4	4	4	4	3	3.8

This scorecard becomes your roadmap. Holdings with scores <2.5 need foundational work (data infrastructure, team building, governance). Holdings with scores >3.5 are ready for rapid AI rollout. Holdings in the middle are candidates for targeted capability-building.

Building the Portfolio AI Capability Layer

Shared Infrastructure vs. Distributed Execution

The core decision: Do you build one AI platform that all holdings use, or do you give each holding autonomy to build their own, with shared standards?

The answer depends on your portfolio composition. If you have 3–4 similar holdings (e.g., regional distribution utilities), a shared platform makes sense: one data warehouse, one ML platform, shared models for demand forecasting and network optimisation. If you have diverse holdings (generation, retail, services), distributed execution with shared standards is more pragmatic.

We recommend a hybrid model:

Shared infrastructure layer: Centralised data warehouse, ML platform (e.g., Databricks, SageMaker), model registry, and monitoring. This is typically a $1–3M investment (capex + opex) and takes 4–6 months to build.
Distributed execution layer: Each holding builds AI applications on top of shared infrastructure, with autonomy over use cases, models, and deployment. This allows for local optimisation while avoiding duplication.
Shared standards and governance: All holdings follow the same model governance framework, data governance policies, and compliance controls. This is enforced via templates, code reviews, and audits.

Building Shared Infrastructure: A 6-Month Roadmap

Months 1–2: Foundation

Decide on cloud platform (AWS, Azure, Google Cloud). For energy companies with on-premise infrastructure, we typically recommend AWS (mature energy sector partnerships, FERC compliance examples, strong data integration tools).
Provision data warehouse (Snowflake, Redshift, BigQuery). For energy use cases with high-frequency telemetry, Snowflake or ClickHouse offer better performance than traditional data warehouses.
Build initial data pipelines from each holding’s SCADA, historian, and ERP systems into the data warehouse. Start with one holding (fastest to value), then scale.
Set up ML platform (SageMaker, Databricks, or Vertex AI). Ensure it integrates with your data warehouse.

Months 3–4: Governance and Tooling

Build model registry and governance layer. Document all models: use case, training data, performance metrics, owner, deployment status.
Set up monitoring and alerting for model drift, data quality, and performance. Use tools like Arize, Fiddler, or custom dashboards.
Build compliance and audit tooling. Document all AI decisions, retraining events, and model changes for regulatory audit.
Establish data governance: define data ownership, access control, retention policies, and lineage tracking.

Months 5–6: Capability and Scaling

Train holding teams on the shared platform. Run workshops on data access, model development, and deployment.
Onboard first use cases from 2–3 holdings. Aim for quick wins: demand forecasting, equipment maintenance prediction, customer churn.
Establish support model: centralised data and ML platform team (4–6 people) provides infrastructure support; holding teams own use-case development.

Staffing the AI Capability Layer

You need a small central team to operate the shared infrastructure. Typical composition:

1 VP/Director of AI: Owns strategy, vendor relationships, governance, and board reporting.
2 Data Engineers: Build and maintain data pipelines, data warehouse, and platform infrastructure.
1 ML Engineer / Platform Engineer: Manage ML platform, model registry, monitoring, and deployment tooling.
1 Data Scientist: Develop reference models, validate use cases, and help holding teams with advanced analytics.
1 Compliance / Governance Lead: (Part-time or shared across portfolio.) Ensure all models meet governance, audit, and regulatory requirements.

Total: ~5 FTE, typically $800K–$1.2M annual cost (depending on location and seniority).

For energy portfolios with specific regional needs—e.g., if you have holdings in Denver, Houston, and Calgary—consider fractional CTO leadership. A fractional CTO in Denver or Houston can provide part-time technical leadership, vendor selection, and hiring support without requiring a full-time executive hire. This is particularly useful during the first 12–18 months while you’re building the operating model.

Quick Wins: First AI Projects

Don’t wait for perfect infrastructure. Identify 2–3 quick-win projects in the first 90 days:

Use Case 1: Demand Forecasting

Relevant for: Retail providers, distribution utilities, generation companies.

Timeline: 6–8 weeks from data access to production model.

Impact: 2–5% improvement in forecast accuracy translates to $5–20M annual value for large portfolios (via better procurement, reduced imbalance costs, improved customer retention).

Approach: Use historical demand data (12–24 months minimum), add weather data (public APIs), and train a gradient-boosted model (XGBoost, LightGBM) or neural network. Deploy as a daily batch job that feeds into operations and procurement systems.

Use Case 2: Predictive Maintenance

Relevant for: Generation, distribution, transmission.

Timeline: 8–12 weeks (data collection + model development).

Impact: 10–20% reduction in unplanned downtime, $10–50M annual value depending on portfolio size.

Approach: Use SCADA and historian data (vibration, temperature, pressure, other sensor data) to train anomaly detection or failure prediction models. Deploy as real-time monitoring that alerts operations teams to degradation before failure.

Use Case 3: Customer Churn Prediction

Relevant for: Retail providers.

Timeline: 4–6 weeks.

Impact: 5–10% improvement in retention, $2–10M annual value.

Approach: Use billing data, customer service interactions, and usage patterns to predict churn. Deploy as a weekly batch job that identifies at-risk customers for proactive retention campaigns.

Each of these projects should be owned by a holding team (with central AI team support) and documented as a model card in the model registry. This builds muscle memory for the operating model and creates proof points for board and investor reporting.

Governance, Risk, and Compliance at Scale

The AI Governance Framework

Energy is a regulated industry. Your AI operating model must embed governance from day one, not bolt it on later.

Build a three-layer governance framework:

Layer 1: Model Development Governance

Every model that goes into production must follow a checklist:

This checklist is enforced via code review: no model goes to production without sign-off from the central governance team.

Layer 2: Data Governance

Data is the foundation of AI. Establish clear policies:

Data ownership: Every dataset has a documented owner (business unit, team) responsible for quality and access control.
Data lineage: Track where data comes from, how it’s transformed, and where it flows. Use tools like dbt (data build tool) or Collibra for lineage tracking.
Access control: Implement role-based access control (RBAC) in your data warehouse. Restrict access to sensitive data (customer PII, operational data) to authorised users.
Retention and deletion: Define how long data is kept and how it’s deleted when no longer needed. Energy companies often have regulatory retention requirements (e.g., 7–10 years for certain operational data).
Data quality: Establish SLAs for data completeness, timeliness, and accuracy. Monitor and alert on violations.

Layer 3: Compliance and Audit

Energy companies operate in regulated environments. Your AI operating model must be audit-ready.

Regulatory mapping: Map your AI use cases to regulatory requirements. (E.g., if you have a model that informs dispatch decisions, ensure it’s documented and validated per NERC standards.)
Audit trail: Log all AI decisions, model updates, and governance changes. Ensure logs are immutable and retained per regulatory requirements.
Third-party audit readiness: If you’re pursuing SOC 2 or ISO 27001 certification, ensure your AI governance aligns with audit requirements. Use frameworks like the NIST AI Risk Management Framework to structure your controls.
Incident response: Define how you respond to AI-related incidents (model failure, data breach, regulatory violation). Document and report per regulatory requirements.

Compliance Roadmap: SOC 2 and ISO 27001 for AI

If your portfolio includes regulated or customer-facing companies, SOC 2 Type II or ISO 27001 certification is increasingly expected. AI adds complexity to these audits.

Timeline: 6–12 months from starting point to certification.

Approach:

Months 1–2: Conduct a compliance gap assessment. Identify what controls are missing.
Months 3–6: Implement controls. Document policies, build tooling, train teams.
Months 7–9: Run internal audits. Find and fix remaining gaps.
Months 10–12: Engage external auditor. Work through audit findings. Achieve certification.

For AI-specific controls, focus on:

AI model risk management: Controls for model development, testing, deployment, and monitoring.
Data governance: Controls for data access, retention, and quality.
Audit logging: Controls for logging all AI decisions and model changes.
Incident response: Controls for responding to AI-related incidents.

Many energy companies use Vanta or similar compliance automation platforms to streamline SOC 2 and ISO 27001 audits. Vanta integrates with AWS, Azure, and other cloud platforms to automatically collect evidence of controls, reducing manual audit work.

Value-Creation Playbook: AI Rollout Across Holdings

The 12-Month Value-Creation Roadmap

Once you’ve built the foundation (shared infrastructure, governance, first quick wins), you’re ready to scale AI across the portfolio. A typical 12-month roadmap looks like this:

Months 1–3: Quick Wins and Team Building

Launch 2–3 quick-win projects (demand forecasting, predictive maintenance, churn prediction) as described above.
Hire or contract central AI team (VP AI, data engineers, ML engineer, data scientist).
Conduct tech audits across all holdings.
Establish governance framework and model registry.

Months 4–6: Infrastructure and Capability Building

Complete shared data infrastructure (data warehouse, ML platform, monitoring).
Build data pipelines from all holdings into central data warehouse.
Train holding teams on platform and governance framework.
Identify 5–10 AI use cases across portfolio (prioritised by impact and feasibility).

Months 7–9: Scaling and Operationalisation

Launch 3–5 use cases in parallel across different holdings.
Establish support model and SLAs for central platform team.
Implement monitoring and alerting for all models in production.
Conduct first governance audits and compliance checks.

Months 10–12: Optimisation and Exit Positioning

Measure and report on AI impact (revenue, cost, efficiency gains).
Identify opportunities for model reuse and consolidation across holdings.
Plan next phase: advanced use cases, new technology (e.g., generative AI, agentic AI).
Prepare AI narrative for investor reporting and exit diligence.

Use Case Prioritisation Framework

Not all AI use cases are created equal. Prioritise based on:

Impact: What’s the potential value? (Revenue uplift, cost reduction, efficiency gain.) Aim for use cases with >$1M annual impact.
Feasibility: How hard is it to execute? (Data availability, team capability, technical complexity.) Prioritise easy wins first.
Time-to-value: How long until we see results? Aim for projects with 3–6 month payback periods.
Scalability: Can we reuse the model or approach across multiple holdings? Favour use cases that apply to >2 holdings.
Regulatory risk: Are there compliance or safety implications? Prioritise lower-risk use cases early; build governance muscle before tackling high-risk applications.

Use a simple 2x2 matrix:

High Impact / High Feasibility: Do first (quick wins)
High Impact / Low Feasibility: Do second (after building capability)
Low Impact / High Feasibility: Do third (build momentum)
Low Impact / Low Feasibility: Don't do (or defer)

Typical Portfolio AI Use Cases by Segment

Generation (Fossil and Renewable)

Predictive maintenance: Reduce unplanned downtime by 10–20%.
Demand forecasting: Improve procurement and dispatch by 2–5%.
Efficiency optimisation: Reduce fuel costs or improve renewable output by 3–8%.
Grid stability prediction: Anticipate frequency or voltage issues.

Distribution and Transmission

Network optimisation: Reduce losses and congestion by 2–5%.
Fault prediction: Anticipate equipment failures and plan maintenance.
Demand response: Automatically manage load to balance grid.
Customer outage prediction: Proactively manage restoration and communication.

Retail (Energy Supply)

Demand forecasting: Reduce procurement costs by 2–5%.
Customer churn prediction: Improve retention by 5–10%.
Pricing optimisation: Increase margins by 1–3%.
Fraud detection: Reduce theft and billing errors by 5–10%.

Services (Maintenance, Consulting, Software)

Predictive maintenance: Improve service margins and customer satisfaction.
Resource optimisation: Allocate technicians and equipment more efficiently.
Lead scoring: Improve sales conversion and pipeline quality.
Sentiment analysis: Monitor customer satisfaction and brand health.

Execution Model: Central Team + Holding Teams

Who builds and operates the AI models?

Central AI Team (5 FTE):

Owns shared infrastructure (data warehouse, ML platform, monitoring, governance).
Provides data engineering and ML platform support to holding teams.
Develops reference models and best practices.
Ensures compliance and governance across all models.
Reports on portfolio AI impact and strategy to executive team.

Holding Teams (2–3 FTE per holding, depending on size):

Own use-case identification and prioritisation.
Partner with central team to develop and deploy models.
Own model monitoring, retraining, and operations.
Own stakeholder engagement and change management.
Contribute data engineers and domain experts to model development.

This model avoids two extremes: (1) centralised model-building, which is slow and disconnected from business context, and (2) completely decentralised AI, which leads to fragmentation and duplicated effort.

Fractional CTO Leadership for Holdings

If a holding lacks engineering leadership or is too small to justify a full-time CTO, fractional CTO services provide cost-effective technical guidance. A fractional CTO in Perth or Sydney can work 1–2 days per week on architecture, hiring, vendor selection, and AI strategy—without the overhead of a full-time executive.

This is particularly valuable for:

Regional distribution utilities (often smaller, limited engineering bench).
Acquired companies (where the founding team has left or moved on).
Holdings in transition (during tech debt paydown or platform migration).

Typical engagement: 6–12 months, $150K–$400K, focused on building engineering capability and executing a specific technical roadmap.

Measuring and Benchmarking AI Impact

The AI Metrics Dashboard

You can’t manage what you don’t measure. Build a portfolio-wide AI metrics dashboard that tracks:

Business Metrics (What matters to the board and investors)

Revenue impact: Models that directly increase revenue (churn prediction, pricing optimisation, lead scoring). Target: $1–5M per holding per year.
Cost impact: Models that reduce costs (demand forecasting, predictive maintenance, network optimisation). Target: $2–10M per holding per year.
Efficiency impact: Models that improve operations without direct revenue or cost impact (faster decision-making, better resource allocation). Target: 5–15% improvement in key operational metrics.
Time-to-value: How long from project kickoff to production model delivering value. Target: 6–12 weeks for quick wins, 3–6 months for complex projects.

Operational Metrics (What matters to the operations and AI teams)

Models in production: Count of models deployed and actively in use. Target: 2–3 new models per holding per year.
Model uptime: Percentage of time models are available and performing. Target: >99%.
Model accuracy: Accuracy, precision, recall, or other relevant metric depending on use case. Target: >90% for most use cases, >95% for critical safety applications.
Data freshness: How current is the data feeding the models? Target: Real-time or hourly for operational models, daily for analytics.
Retraining frequency: How often are models retrained? Target: Monthly or quarterly for most models, more frequent for models in volatile environments.

Governance Metrics (What matters to compliance and audit teams)

Governance compliance: Percentage of models that have completed governance checklist. Target: 100%.
Audit readiness: Percentage of models that are audit-ready (documented, monitored, compliant). Target: 100%.
Incident response time: How quickly do teams respond to model failures or data quality issues? Target: <4 hours for critical models.
Data governance compliance: Percentage of data that is properly classified, governed, and access-controlled. Target: 100%.

Benchmarking Against Industry

How does your portfolio compare to peers? Industry benchmarks (from McKinsey, Gartner, and others) suggest:

AI adoption: 35–50% of large energy companies have >5 models in production. (You should aim for >3 per holding within 12 months.)
Time-to-value: Leading companies achieve 3–6 month payback periods. (Laggards take 12+ months.)
Cost of AI infrastructure: $500K–$2M per year for portfolio-wide infrastructure supporting 10–20 holdings. (This includes cloud, platform, and team costs.)
ROI on AI: 200–400% annual ROI for well-executed programmes. (This means $1M invested in AI infrastructure and team should generate $2–4M in value.)

For detailed benchmarks and real-world examples, see the Google Cloud roundup of real-world generative AI use cases from industry leaders, which includes energy-sector examples and case studies.

Reporting to the Board and Investors

Monthly or quarterly, report on:

Portfolio AI Impact: Total value created (revenue + cost + efficiency) year-to-date and projected for full year.
Model Count and Status: How many models are in production? How many are in development? What’s the pipeline?
Key Wins: Highlight 1–2 successful use cases with concrete numbers (e.g., “Predictive maintenance model at Gen Co A prevented 3 unplanned outages, saving $2.5M in downtime costs”).
Governance and Compliance: All models are governance-compliant and audit-ready. No compliance findings.
Roadmap and Next Steps: What’s coming next? (New use cases, new technology, new holdings.)

Keep it simple. The board cares about outcomes (revenue, cost, risk reduction), not technical details. A one-page summary with a chart is more effective than a 50-slide deck.

Exit Positioning and Diligence-Ready Tech Stories

Building a Diligence-Ready Tech Narrative

When you exit, buyers will conduct tech diligence. A well-executed AI operating model becomes a value multiplier if it’s properly documented and packaged.

For each holding, prepare a Tech Story that covers:

Technology Foundation: What’s the current state of the technology stack? (Cloud, data infrastructure, systems of record.)
AI Capability: What AI models are in production? What value are they creating? How are they governed?
Team and Talent: Who built and operates the technology? Are they staying post-exit? (Retention agreements.)
Roadmap: What’s the vision for the next 3–5 years? (New platforms, new use cases, new markets.)
Compliance and Risk: What’s the compliance status? (SOC 2, ISO 27001, regulatory audit results.) Are there any known risks?

Diligence Checklist for Exit

Before you go to market, ensure:

All models are documented in the model registry with current performance metrics.
Data governance is clear: who owns each dataset, access controls are enforced, data quality is high.
Compliance status is current: SOC 2 or ISO 27001 certification (if applicable), no outstanding audit findings.
Tech debt is mapped and prioritised: buyers understand what needs to be fixed and why.
Team is stable: key technical leaders have retention agreements or clear succession plans.
Roadmap is realistic: next 12–24 months of development is planned and resourced.
Competitive positioning is clear: how does the technology differentiate in the market?

AI as a Value Multiplier

A portfolio company with a mature AI operating model can command a 10–20% valuation premium over peers with no AI capability. This premium is justified by:

Recurring revenue from AI-driven products or services: If you’ve built AI-powered offerings (e.g., a demand forecasting service for other utilities), this is a new revenue stream.
Cost reduction: AI-driven efficiency gains improve margins and cash flow.
Competitive moat: If your AI models are proprietary and hard to replicate, they create defensibility.
Buyer optionality: Buyers can leverage your AI infrastructure and models to create value in their own portfolio.

Quantify the premium: If your company is worth $100M on earnings, and AI is generating $5M in annual value (revenue + cost reduction), the AI premium could be $10–20M (2–4x revenue multiple).

Case Studies and Proof Points

Before you go to market, prepare 2–3 case studies that demonstrate AI impact:

Case Study 1: Quick-win project with clear ROI. (E.g., “Demand forecasting model improved forecast accuracy by 3%, reducing procurement costs by $2.5M annually.”)
Case Study 2: Complex, high-impact project that showcases technical capability. (E.g., “Predictive maintenance platform built on real-time SCADA data, preventing 5+ unplanned outages per year.”)
Case Study 3: Governance and compliance success. (E.g., “Achieved SOC 2 Type II certification with AI-specific controls, enabling new customer segments and revenue opportunities.”)

Each case study should include: use case, timeline, team, technology stack, results, and lessons learned. See PADISO’s case studies for examples of how to structure and present technical wins.

Regional Execution: Hubs and Fractional Leadership

Geographic Considerations for Energy Portfolios

Energy portfolios are often geographically dispersed: a generation company in Texas, a distribution utility in Colorado, a retail provider in Canada. This creates challenges for centralised AI team support.

Consider a hub-and-spoke model:

Central hub (Sydney, New York, or another major tech centre): Central AI team (VP AI, data engineers, ML engineers, data scientists). Owns platform, governance, and cross-portfolio strategy.
Regional hubs (Denver, Houston, Calgary, Perth): Fractional technical leadership (fractional CTO or VP Engineering) who supports 2–3 holdings in that region. Owns local execution, hiring, vendor relationships.

This model combines the efficiency of centralised infrastructure with the responsiveness of local execution.

Fractional CTO Services by Region

If you have holdings in specific regions, fractional CTO services provide cost-effective technical leadership without the overhead of full-time executives.

For energy companies in Colorado and Wyoming:

A fractional CTO in Denver can provide 1–2 days per week of technical leadership, focusing on:

Architecture and platform decisions for renewable energy and grid-edge applications.
Hiring and team-building in a competitive tech market (Boulder, Denver, Fort Collins).
Vendor evaluation and management (cloud, data platforms, AI/ML tools).
AI strategy and roadmap development.

Typical engagement: 6–12 months, $150K–$350K.

For energy companies in Texas:

A fractional CTO in Houston can provide technical leadership focused on:

OT/IT integration and industrial architecture for oil, gas, and power generation.
Regulated data handling and compliance (FERC, NERC, state regulations).
Historian and SCADA integration with cloud platforms.
Hiring in a tight labour market (competition from oil/gas, aerospace, healthcare).

Typical engagement: 6–12 months, $150K–$400K.

For energy companies in Canada (Calgary, Edmonton):

A fractional CTO in Calgary or Edmonton can support:

Energy transition and renewable integration.
Operational data platforms and time-series analytics.
Hiring and team-building in a regional market.
Cross-border compliance and regulatory strategy.

For energy companies in Australia (Perth, Sydney):

A fractional CTO in Perth can support mining and energy companies with:

OT/IT integration and industrial architecture.
SCADA and historian data pipelines.
Predictive maintenance and asset optimisation.
Hiring discipline in a competitive tech market.

A fractional CTO in Sydney can provide broader strategic support for:

AI strategy and roadmap development.
Board-ready tech stories and investor reporting.
Compliance and governance frameworks.
Cross-portfolio technical strategy.

Building a Distributed Team Model

With fractional CTO leadership and a central AI team, you can scale technical leadership across a geographically dispersed portfolio without hiring a full-time CTO at every holding.

Typical structure for a 5–10 holding portfolio:

Central hub (1 city): 5–6 FTE (VP AI, 2 data engineers, 1 ML engineer, 1 data scientist, 1 governance lead). Cost: $1M–$1.5M.
Regional hubs (2–3 cities): Fractional CTO (0.5–1 FTE equivalent) per region. Cost: $300K–$600K total.
Holding teams: 2–3 FTE per holding (engineer, data engineer, operations liaison). Cost: $300K–$600K per holding.

Total portfolio AI capability cost: $2–3M for a 10-holding portfolio. This should generate $10–30M in annual value (3–5x ROI).

Summary and Next Steps

The Portfolio-Wide AI Operating Model: Key Takeaways

Start with diligence: Understand data readiness, model governance, and AI decision exposure in each holding. This informs your roadmap and budget.
Build shared infrastructure: A centralised data warehouse, ML platform, and governance framework creates efficiency and consistency across holdings. Budget 4–6 months and $1–3M.
Establish governance from day one: AI governance is not optional in energy. Use frameworks like the NIST AI Risk Management Framework to structure controls. Plan for SOC 2 or ISO 27001 compliance within 12 months if you have regulated or customer-facing holdings.
Launch quick wins early: Demand forecasting, predictive maintenance, and churn prediction are high-impact, achievable projects that build momentum and proof points. Target 2–3 in the first 90 days.
Scale with distributed execution: Central AI team owns infrastructure and governance; holding teams own use-case development and operations. This balances efficiency with local ownership.
Measure and report obsessively: Track business impact (revenue, cost, efficiency), operational metrics (model count, uptime, accuracy), and governance metrics (compliance, audit readiness). Report monthly to the board.
Use fractional CTO leadership for distributed portfolios: If you have holdings across multiple regions, fractional CTO services provide cost-effective technical leadership without full-time overhead.
Build for exit: Document your AI capability, governance, and impact clearly. A mature AI operating model is a 10–20% valuation multiplier.

90-Day Action Plan

Week 1–2: Assessment and Planning

Conduct tech and talent audit across all holdings. (Use the 5-dimension framework above.)
Identify 2–3 quick-win AI use cases.
Define centralised AI team composition and hiring plan.
Assess compliance and governance gaps.

Week 3–4: Foundation

Hire VP AI or fractional CTO leadership.
Select cloud platform and data warehouse technology.
Kick off first data pipeline from fastest-moving holding.
Establish governance framework and model registry template.

Week 5–8: Execution

Launch first quick-win project (demand forecasting or predictive maintenance).
Begin hiring data engineers and ML engineers.
Build data pipelines from 2–3 additional holdings.
Conduct governance training with holding teams.

Week 9–12: Scaling

First quick-win model goes to production.
Second and third use cases in active development.
Central AI team is 50% staffed.
Compliance and governance baseline is established.

Getting Started

If you’re building a portfolio-wide AI operating model for energy, you don’t need to do it alone. PADISO works with PE firms and portfolio companies across North America and Australia on exactly this challenge.

Our team has built shared AI infrastructure for multi-holding portfolios, led tech diligence on energy acquisitions, and helped companies achieve SOC 2 and ISO 27001 compliance with AI-specific controls. We provide:

AI Strategy & Readiness: Help you design a portfolio-wide AI operating model tailored to your holdings and timeline.
Fractional CTO Leadership: Provide technical leadership for specific holdings or regions via part-time CTO advisory.
Platform Engineering: Build shared data infrastructure, ML platforms, and governance tooling.
Security Audit & Compliance: Guide you through SOC 2, ISO 27001, and AI-specific governance frameworks.

If you’re in Sydney or Australia, our Sydney-based AI advisory team can help you design and execute your AI operating model. If you’re in North America, we have teams in Denver, Houston, Calgary, and other energy hubs.

Reach out for a 30-minute conversation about your portfolio, challenges, and roadmap. We’ll help you move from strategy to execution.

Appendix: AI Operating Model Checklist

Pre-Acquisition Diligence

Data inventory: >60% of operational data is Ready or Workable.
Model governance: No undocumented models in production.
AI decision exposure: Mapped and understood.
Team interviews: Clear understanding of AI capability and gaps.
Regulatory status: No outstanding AI-related audit findings.
Budget estimate: 3–12 months, $500K–$2M.

90-Day Tech Audit

Technology stack and debt assessed.
Engineering capability and team structure evaluated.
Data maturity and accessibility assessed.
Governance, risk, and compliance posture understood.
Organisational readiness for change evaluated.
Scorecard completed for all holdings.

Shared Infrastructure Build (Months 1–6)

Governance Framework

Model development checklist created and enforced.
Data governance policies documented.
Compliance and audit framework established.
Incident response plan documented.
Training and change management plan created.

12-Month Rollout

2–3 quick-win projects in production (Months 1–3).
Shared infrastructure complete (Months 4–6).
3–5 use cases launched across holdings (Months 7–9).
AI impact measured and reported to board (Months 10–12).
Roadmap for next phase created.

Exit Preparation

Tech story documented for each holding.
All models documented in model registry.
Compliance status current (SOC 2, ISO 27001, regulatory audits).
Tech debt mapped and prioritised.
Team stability and retention agreements in place.
2–3 case studies prepared with concrete ROI.
Competitive positioning and differentiation clear.

Want to talk through your situation?

Book a 30-minute call with Kevin (Founder/CEO). No pitch — direct advice on what to do next.

Book a 30-min call