You may not be able to isolate every dollar of AI energy savings—but you can give the board a credible proof plan

This guide from Rohan Hall’s personal site addresses the management problem directly: an executive championed an AI-driven energy optimization system for Anaheim facilities, and an Orange County board now wants financial proof that is difficult to isolate. The first task is not to defend AI or predict what the board will do. It is to turn an emotionally charged challenge into a decision-quality review of performance, evidence and next steps.

Your professional risk cannot be determined from the information available. Boards can criticize a project because of weak performance, cost overruns, earlier decisions, governance failures or poor communication even when the project leader is candid. What you control now is the quality of the evidence, the clarity of attribution and the practicality of the corrective plan. If the discussion is becoming a broader dispute about near-term returns, use the framework for explaining long-term AI value to a quarterly focused board to separate strategic potential from claims about current savings.

That statement is stronger than either extreme: claiming that all improvement came from AI or conceding that the system produced no value because the exact contribution is uncertain. A similar attribution problem arises when leaders must connect AI trend analysis to sales: the correct response is to map the chain of effects and test each link, not jump from system activity to financial impact.

Ask what the board must decide. It may be considering whether to keep the system operating, modify its controls, deploy it to more facilities, renegotiate a contract or stop further investment. Each decision requires different evidence. Expansion requires repeatability across facilities. Continued operation requires expected benefits to justify ongoing costs and operational risk. A decision to stop should consider transition costs and any operational effects that would disappear with the system.

Next, define “savings” precisely. Reduced electricity consumption is not automatically equal to reduced expense. The bill can also reflect tariffs, demand charges, time-of-use periods, taxes and other factors. Likewise, lower cost does not necessarily prove lower consumption if rates or operating schedules changed. Keep physical performance, billed cost and avoided cost in separate columns until the causal relationships are clear.

This discipline also applies outside energy management. The reporting approach for quantifying efficiency gains from AI automation is useful because time saved, activity completed and money retained are related measures, but they are not interchangeable.

The core question is counterfactual: what would the Anaheim facilities have consumed and paid without the AI-driven optimization system? A simple before-and-after comparison is rarely sufficient. Weather, occupancy, production volume, operating hours, equipment replacements, maintenance, shutdowns, tariff changes and manual operating decisions can all move the result. List every material change rather than letting the board discover one during questioning.

If comparable buildings, equipment or operating periods exist, use them as reference points—but do not pretend they are identical. A control facility can strengthen the case only when its differences are documented. Where no credible comparison exists, say so and improve measurement prospectively. The same principle appears in predictive-maintenance cost measurement, where avoided failures are difficult to prove unless maintenance events, downtime and asset behavior are tracked consistently.

The board does not need a tour of the AI model. It needs a compact explanation of the business objective, investment, observed outcome, confidence in attribution, remaining risk and requested decision. Put the conclusion first and retain the detailed methodology for questions. If the strongest evidence is operational rather than financial, label it that way instead of converting it prematurely into dollars.

Prepare an appendix that lets a skeptical director trace the headline number back to source data. Reconcile the calculation with utility bills and financial records, explain exclusions and retain versions of the analysis. Invite finance to critique it before circulation. That is more credible than asking the board to accept a model owned solely by the project team. For another example of translating an indirect operational outcome into financial language, see how to measure an AI customer-service chatbot’s return.

Open with the gap before anyone else does: “The system’s contribution cannot yet be isolated to a single exact figure with the available baseline.” Then move immediately to what is known. This signals control of the issue rather than concealment. Do not lead with technical complexity, accuse stakeholders of misunderstanding AI or argue that innovation should be exempt from financial scrutiny.

When asked why the weakness was not identified earlier, distinguish between an execution failure and a measurement-design failure. Accept responsibility for choices within your control, but do not speculate or absorb responsibility for undocumented decisions. Bring a chronology of the original approval, measurement assumptions, scope changes and prior reporting. If continued investment has already been rejected, the guidance on building a stronger AI investment case after rejection can help restructure the next proposal around measurable decision gates.

Avoid presenting secondary indicators as though they were cash. Better control consistency, reduced manual intervention or improved visibility may matter operationally, but the board is entitled to ask how they affect cost, risk or future decisions. This is also why saying an AI tool simply “feels better” is inadequate; the framework for articulating the value of AI-assisted recruiting shows how to move from subjective improvement to observable workflow and economic measures.

If the current data cannot support clean attribution, propose a prospective test rather than repeatedly reworking a weak retrospective analysis. Establish the baseline before the next comparison period, identify variables that must be captured, define the financial conversion method and agree in advance on what would count as success. Where operationally safe, compare facilities, equipment groups, schedules or staged deployment periods. The test design should reflect facilities and engineering constraints, not override them.

Assign accountability across the functions that own the evidence. Facilities should validate operating conditions and interventions. The technical team should document when recommendations or controls were active. Finance should approve rate assumptions and cost treatment. An independent analyst may help challenge the methodology when the project team’s incentives could be questioned. Establish decision thresholds before results arrive so success is not redefined afterward.

If the system’s value is expected to emerge over a longer horizon, keep that strategic case separate from the current savings claim. Rohan Hall examines AI alongside other emerging technologies in The Convergence of AI and the Top 10 Emerging Technologies, but a broader technology thesis should never substitute for facility-level evidence.

Rohan Hall simplifies complex AI ideas so they are accessible and useful in decision-making. His broader work, ventures, writing and podcast are available through Rohan Hall’s home page. He founded OceSha Ventures and its AI-first solutions, which builds and operates course creation, branded academies, AI assistants such as Lumi, and business intelligence for businesses and organizations.

OceSha AI is the course and academy creation platform Rohan built. For organizations that need to turn an internal framework—such as an energy-measurement method, governance standard or board-reporting process—into structured learning, OceSha AI’s course and academy platform is the relevant offering. That role is educational and organizational; the facility’s savings case still depends on its own operational and financial evidence.

The larger lesson is that AI leadership requires both technical understanding and institutional trust. A system can be sophisticated and still fail executive scrutiny if its value mechanism, baseline and accountability were never designed clearly. Rohan’s work across AI and other emerging technologies informs this perspective, but your immediate priority is straightforward: produce a transparent account of the project and a credible route to a decision.