ABSTRACT

An energy audit is the process of examining how a building uses energy and identifying the measures that can reduce unnecessary consumption. The process normally includes utility-bill analysis, a site inspection, equipment evaluation, and the calculation of potential savings. Artificial intelligence (AI) may improve parts of this process by analyzing large sets of building data, identifying unusual operating conditions, and predicting future energy use. However, AI does not eliminate the need to verify equipment conditions or confirm savings through engineering measurements. This paper describes how AI can be incorporated into the normal energy audit process without replacing the judgment of the energy auditor. It discusses energy forecasting, fault detection, HVAC control, and renewable energy management. A hypothetical 600,000-kWh-per-year commercial building is used to explain the financial analysis of an AI-supported operational improvement. Under the stated assumptions, annual gross savings are $7,200, annual net savings are $4,800 after recurring service costs, and simple payback is 2.5 years. These are illustrative values, not measured project results. The paper proposes a practical sequence for selecting, testing, and verifying AI applications in existing commercial buildings.

INTRODUCTION

The energy audit process is divided into three main parts: preliminary energy use analysis, site visit, and post-site work. Preliminary energy use analysis involves collecting and reviewing utility bills, identifying unusual energy consumption, and comparing the building with facilities of similar use. The site visit is the process of checking the information collected during the preliminary analysis and inspecting the equipment that consumes energy. Post-site work involves calculating energy and cost savings and preparing the recommendations for the building owner.

These activities remain necessary even when advanced software is available. An AI system can identify a possible operating problem, but the system may not know whether a sensor is defective or whether the building is being used differently from the way it was used in the past. That information often requires a site visit or confirmation from the facility operator.

Research has identified applications of AI in building design, maintenance, operation, and control [1]. Fault detection systems have also demonstrated energy savings in operating buildings. In one study involving 26 organizations and approximately 550 buildings, the median reported savings among users of fault detection and diagnostics systems was 8% [2]. This finding indicates that the technology can be useful, but it does not establish a savings rate for every building.

The purpose of this paper is to describe a method for adding AI-assisted analysis to the existing energy audit process. Particular attention is given to small and medium-sized commercial facilities where the cost of installing an advanced building automation system may be difficult to justify.

PRELIMINARY ENERGY USE ANALYSIS WITH AI

Preliminary energy use analysis (PEA) is the first phase of the audit, and it must be completed before conclusions are made about the building’s performance. The process involves compiling historical energy data, calculating energy-use indicators, examining operating conditions, and determining whether additional engineering analysis is necessary.

AI can assist with this phase when the auditor has reliable information to analyze. Utility-bill records, interval-meter data, outdoor temperature, building operating hours, and occupancy information can be used to develop an energy baseline. AI-based forecasting may identify periods when actual consumption is substantially higher than expected under similar conditions.

Before using an AI model, the data must be checked. Missing records, inconsistent dates, incorrect meter units, and equipment changes can affect the results. The auditor should also establish a simpler reference model, such as a weather-adjusted regression, and determine whether the AI model improves the analysis enough to justify the additional work.

For example, a model may indicate that a building uses more energy overnight than similar nights in the baseline period. This observation should be treated as a reason for investigation. It could be caused by unnecessary HVAC operation, but it could also result from an approved cleaning shift, a refrigeration load, or new equipment. The next phase of the audit is needed to determine the actual cause.

SITE VISIT AND EQUIPMENT VERIFICATION

A site visit is the process of verifying the preliminary analysis and inspecting the systems that use energy. During the visit, the auditor should review the HVAC equipment, lighting, building controls, operating schedules, major electrical loads, and maintenance records. Equipment conditions must be recorded accurately because they determine whether a proposed measure can be implemented.

AI may help prioritize the inspection. For example, if an analytical system identifies an air-handling unit that consumes excessive power under certain weather conditions, the auditor can examine the unit’s schedule, temperature sensors, dampers, filters, and control sequence. The AI finding directs attention to a possible problem. The auditor must still confirm that the problem exists and determine the appropriate corrective measure.

Portable instruments and temporary monitoring devices can be especially important in buildings without a building automation system. If the available information is limited to monthly utility bills, there may not be enough detail for reliable equipment-level AI fault detection. In such cases, obtaining better measurements may provide more value than purchasing a complicated analytical platform.

MAIN APPLICATIONS OF AI IN BUILDING ENERGY MANAGEMENT

Energy consumption forecasting

Energy forecasting is the process of estimating future consumption based on historical energy use and conditions that affect building load. AI models may include outdoor temperature, occupancy, calendar information, and equipment schedules. The forecast can be used to identify unusual consumption and evaluate how changes in building operation affect energy use.

Forecast quality should be tested with data that were not used to develop the model. The auditor should report the prediction error and determine whether the forecast is sufficiently accurate for the intended

decision. A model that provides an attractive graph but cannot make reliable predictions should not be used to calculate savings.

Fault detection and diagnostics

Fault detection identifies operating conditions that differ from expected performance. Diagnostics is the process of investigating the probable cause. Examples include an economizer that does not function properly, an HVAC unit running after hours, or heating and cooling systems operating at the same time.

Some tools use established engineering rules, while others use statistical or machine-learning models. The two methods can also be combined. AI may identify patterns that are difficult to detect manually, but incorrect sensor data can create false alarms. Every significant finding should be reviewed before repair work is recommended.

HVAC control optimization

AI can also be used to evaluate how HVAC systems are controlled. Possible applications include supply-air temperature reset, equipment scheduling, ventilation control, and demand management. A study conducted at Oak Ridge National Laboratory evaluated control methods based on ASHRAE Guideline 36 and reinforcement learning in a research building [3]. The results demonstrate the potential of advanced control, but the measured performance in that facility should not be treated as typical savings for other buildings.

Changes to controls must preserve equipment safety limits and necessary indoor environmental conditions. A building owner should not allow untested AI software to override protective controls or operate major equipment without appropriate supervision.

Solar power and battery storage

Where a building has solar photovoltaic equipment or battery storage, AI may be used to forecast generation, estimate demand, and recommend charging or discharging schedules. These functions may improve the use of generated electricity or help control demand charges. However, the financial value depends on the utility tariff, battery constraints, weather, and the actual load profile. Savings should be calculated separately from the performance of other conservation measures.

A PRACTICAL IMPLEMENTATION PROCESS

The proposed process for introducing AI into a commercial building is divided into six steps. The steps are intended to ensure that the building has adequate data and that the proposed application solves an identified operating problem.

Step 1 — Establish the baseline. Collect at least twelve months of energy data where possible. Record building area, business activities, major equipment, operating hours, and relevant weather information.

Step 2 — Check the information. Review data quality and identify missing information. Determine whether interval meters, sensors, or temporary monitoring equipment are required.

Step 3 — Perform the conventional audit. Inspect the equipment, verify the building’s actual operation, and identify potential energy conservation measures. An AI tool should not replace this inspection.

Step 4 — Apply AI to a defined problem. Select an application such as abnormal-load detection or equipment scheduling. Compare the tool’s findings with the available engineering evidence and a simpler analysis method.

Step 5 — Implement approved measures. Have qualified personnel make appropriate changes to equipment operation or control sequences. Test the completed work and document the settings.

Step 6 — Verify the savings. Measure the post-implementation results, make appropriate adjustments to the baseline, and determine whether the value of the savings justifies the cost of the AI application.

ILLUSTRATIVE FINANCIAL ANALYSIS

Consider a hypothetical commercial building that uses 600,000 kWh of electricity annually. Assume an average variable electricity charge of $0.15/kWh, giving an annual energy charge of $90,000. The owner considers an AI-supported monitoring and fault detection system with an initial cost of $12,000 and annual software and support costs of $2,400. For the purpose of this example, assume that corrective actions identified through the system reduce annual electricity use by 8%.

The 8% value is a scenario assumption informed by published research on fault detection and diagnostics [2]. It is not a measured result or a guarantee for the hypothetical building.

ItemAssumption
Existing electricity use600,000 kWh/year
Variable electricity charge$0.15/kWh
Existing annual electricity energy charge$90,000
Assumed operational energy reduction8%
Initial implementation cost$12,000
Annual software and support expense$2,400

Annual electricity savings are 600,000 × 8% = 48,000 kWh. The annual gross energy-cost savings are 48,000 × $0.15 = $7,200. After subtracting the annual software and support expense, the net annual savings are $4,800. The simple payback is $12,000 ÷ $4,800, or 2.5 years.

The result depends on the assumed savings and service cost. If the system identifies few useful measures, the payback will be longer. If implementation requires additional meters, controllers, or repairs, those expenses must be included. Demand-charge savings, maintenance benefits, and tariff changes have been excluded from the calculation because there is not enough information to estimate them reliably.

POST-SITE WORK AND MEASUREMENT AND VERIFICATION

Post-site work includes calculating the recommended savings, documenting the equipment changes, and preparing the final report. Where AI has been used to identify an energy conservation measure, the report should distinguish between the AI observation, the finding confirmed during inspection, the corrective work completed, and the savings established through measurement.

The basic relationship used in measurement and verification is:

Verified energy savings = Adjusted baseline energy use − Measured reporting-period energy use

The baseline may require adjustments for changes in weather, occupancy, business hours, or equipment. The appropriate method depends on the proposed measure and the amount of energy it affects. The International Performance Measurement and Verification Protocol provides recognized approaches for

this analysis [4]. A model’s prediction should not be presented as verified savings unless the required post-implementation measurements have been completed.

Savings should also be checked over time. Building schedules, equipment operation, and occupancy change. An AI system may help monitor these changes, but a responsible person must review alarms and complete corrective actions. Monitoring does not save energy if identified problems are never corrected.

IMPLEMENTATION LIMITATIONS

The principal limitations of AI-based energy management are the quality of the data, the cost of installation, the reliability of the model, and the ability of the building staff to use the results. An older building may not have interval meters or dependable sensors. The cost of adding these components must be compared with the amount of energy that can reasonably be saved.

Cybersecurity and protection of building operating information must also be considered. Systems connected to HVAC controls or electrical equipment should have suitable access restrictions and defined responsibilities. The energy used by cloud computing and other AI infrastructure is another factor, although the relevant share for a particular building application must be evaluated rather than assumed [5].

This paper presents an implementation method and a hypothetical economic example. It does not establish that every AI installation will be cost-effective. Additional field studies should compare buildings using conventional audits with buildings receiving AI-assisted monitoring and should document installation costs, verified savings, and persistence over time.

CONCLUSIONS

AI can help energy auditors review building data, identify unusual operating conditions, and monitor whether corrective actions continue to produce savings. It is most useful when it is applied to a specific technical problem and supported by reliable measurements. The conventional audit process remains necessary because equipment conditions, operating requirements, and safety issues must be confirmed at the site.

The approach described in this paper begins with preliminary energy use analysis, continues through site inspection and targeted AI analysis, and ends with post-site calculations and verification. The illustrative calculation produces estimated annual net savings of $4,800 and a 2.5-year payback, but actual project results will depend on the building and the cost of implementation. The recommended practice is to use AI as an engineering support tool, document the assumptions, and verify the outcome before reporting energy savings as achieved.

REFERENCES

  1. Yussuf, R. O., and Asfour, O. S. (2024). “Applications of artificial intelligence for energy efficiency throughout the building lifecycle: An overview.” Energy and Buildings, 305, 113903. https://doi.org/10.1016/j.enbuild.2024.113903
  2. Lin, G., Kramer, H., and Granderson, J. (2020). “Building fault detection and diagnostics: Achieved savings, and methods to evaluate algorithm performance.” Building and Environment, 168, 106505. https://doi.org/10.1016/j.buildenv.2019.106505
  3. Yoon, Y., Amasyali, K., Li, Y., Im, P., Bae, Y., Liu, Y., and Zandi, H. (2024). “Energy performance evaluation of the ASHRAE Guideline 36 control and reinforcement learning-based control using field measurements.” Energy and Buildings, 325, 115005. https://doi.org/10.1016/j.enbuild.2024.115005
  4. Efficiency Valuation Organization. International Performance Measurement and Verification Protocol (IPMVP),

Core Concepts 2022. https://evo-world.org/en/products-services-mainmenu-en/protocols/ipmvp

  1. International Energy Agency. Energy and AI (2025). https://www.iea.org/reports/energy-and-ai

AUTHOR BIOGRAPHY

Mtijan M. Kamara, C.E.M., is a Certified Energy Manager with extensive experience in energy management, energy auditing, and sustainable energy solutions. His expertise spans conventional and renewable energy systems, with a focus on energy efficiency, conservation, and cost-effective energy solutions. Throughout his career, he has worked in research, project management, and consulting, advising organizations on improving energy performance, reducing operating costs, and promoting sustainability.