← All insights
Energy

The Operational Performance Gap: Why Commercial Buildings Waste Energy They Know How to Save

The Operational Performance Gap: Why Commercial Buildings Waste Energy They Know How to Save

The operational performance gap is the difference between how a commercial building’s systems are designed to run and how they run day to day. Most of the energy lost to that gap comes from problems the industry already knows how to fix: economizer dampers stuck closed, heating and cooling running against each other, schedules that no longer match occupancy, and manual overrides left in place. According to ENERGY STAR, commercial buildings and industrial plants often waste up to 30 percent or more of the energy they use through inefficiencies. Lawrence Berkeley National Laboratory (LBNL) reviewed commissioning results from 643 buildings and found median whole-building energy savings of 16 percent in existing buildings, with a median payback of 1.1 years.

Commissioning recovers those savings, but only for as long as a building keeps running the way it was commissioned. A commissioning project captures a building’s performance at one point in time. As setpoints are overridden, occupancy patterns change, and control sequences are edited, the building drifts away from its commissioned state and the savings erode. Keeping the savings therefore means recommissioning on a regular cycle, which turns a one-time project into recurring work for engineering teams.

Engineering teams rarely have the hours for that recurring work. Building engineers typically spend their days on comfort calls, preventive maintenance, and tenant requests, which leaves little time to review raw trend data or work through lists of faults.

Fault detection and diagnostics (FDD) software took on part of that work by automating fault finding: FDD monitors equipment data continuously and flags faults as they occur. But a flagged fault is only the starting point for engineering work. Before anyone can act on a fault, an engineer has to confirm that the fault is real, find its root cause, and judge whether the fix is worth making. FDD made engineers aware of more faults without reducing the work of diagnosing each one.

The scarce resource in closing the operational performance gap is therefore analytical capacity. Information about faults is abundant: buildings with FDD generate more of it than engineering teams can analyze and synthesize into action. Agentic AI is now being deployed to take on that analysis. This article traces the causes of the gap, how FDD moved the workload downstream, and how agentic AI takes on the diagnosis that remains.

Causes of the operational performance gap

Avoidable waste in commercial buildings traces back to faults that go unnoticed, performance that drifts after tune-ups, and too few experienced engineers to keep up with either.

Faults that never trigger a complaint

Many of the costliest HVAC faults leave occupants comfortable but still show up in the utility bill. An economizer damper stuck closed forfeits free cooling whenever outdoor air is cool enough to meet the cooling load. Simultaneous heating and cooling, such as a reheat coil warming air the air handler has just cooled, spends energy on both processes while the zone still holds setpoint. Neither condition typically trips a building automation system (BAS) alarm, so both can run for months before anyone investigates.

Economizer failures alone are widespread: a 2014 paper by Kristin Heinemeier of UC Davis, presented at the ACEEE Summer Study, compiled field studies of economizers that found failure rates from 43 percent to 100 percent, and put the average at about 60 percent. Faults like these surface only when someone goes looking for them in trend data.

Drift after commissioning and tune-ups

Buildings that perform well after retro-commissioning or a controls tune-up frequently give back part of the gain within a few years. Operators override setpoints to answer hot and cold calls and never release the overrides, tenants change their hours, and control sequences get edited without documentation. LBNL’s evaluation of a Sacramento Municipal Utility District retrocommissioning program for large commercial buildings found aggregate electricity savings of approximately 10.5 percent in the second year after retro-commissioning, diminishing to approximately 8 percent by the fourth year. Across the 8 participating buildings, only 59 percent of the recommended measures had been implemented in the first place.

Both findings point to the same constraint. Recommended changes depend on engineering time to implement, and keeping them in place depends on someone checking that each change holds.

Fewer engineers covering more buildings

Many experienced building engineers are nearing retirement, and organizations frequently lack trained replacements to operate and maintain building systems, a trend consultants described in a 2019 article in IFMA’s Facility Management Journal. The judgment the work requires, such as telling a failed sensor from a real fault or tracing a comfort complaint back through an air handler to the chiller plant, takes years to build. Portfolios frequently run with one engineer covering several buildings, and many mid-market properties have no dedicated engineering staff. Faults and drift accumulate fastest where coverage is thinnest. Owners adopted FDD software in part to stretch that thin coverage, by monitoring equipment data around the clock in buildings without engineers on site.

FDD made faults easier to find and moved the labor downstream

FDD delivered real savings, but it also created a new body of analytical work that lands on the same engineers.

FDD moved fault finding from periodic audits to continuous monitoring, and the results are measurable. In the U.S. Department of Energy’s Smart Energy Analytics Campaign, run with LBNL, participants using FDD achieved median energy savings of 9 percent after two years, against 4 percent for participants using energy information systems (EIS).

But the same campaign also documented the labor behind those results. LBNL’s synthesis of the campaign’s third year notes that an energy management and information system (EMIS), the category that includes FDD, “does not directly produce savings, but requires action upon the analytic results.” The report identifies a lack of staff time to review dashboards and reports, and to investigate and implement findings, as a common challenge. Participants reported a median of 9 hours of in-house labor per month per building to operate FDD. Across a 50-building portfolio, that rate comes to approximately 450 labor hours per month, close to 3 full-time positions.

The investigative share of those hours goes to triage and root cause analysis. A chilled water bypass valve stuck open at the plant can starve several air handlers of chilled water, raising faults on each air handler and on dozens of zones downstream. An engineer has to rule out sensor error, group the related faults, trace them to a common root cause, estimate the savings a fix would return, and rank that fix against the rest of the queue before deciding what to change. FDD tools that add findings without doing that analysis lengthen a queue engineering teams already struggle to clear. In practice, labor hours shifted from walking the plant to triaging alerts, and savings came to depend on how much of the queue a team could resolve. The resulting backlog is often described as the insight-to-action gap: owners of buildings with FDD frequently know about more faults than their teams have the hours to fix.

Agentic AI takes on triage and root cause analysis

Agentic AI for building operations narrows the gap by doing the triage and diagnostic work FDD leaves to engineers, continuously and across portfolios.

An AI agent is software that carries out a multistep task toward a goal, choosing each step based on what the previous one found. Fault detection rules, by comparison, test one condition at a time. Agentic AI platforms use AI agents that reason across BAS trend data, equipment relationships, schedules, weather, and utility rates. Rather than issuing an alert per fault, these platforms group related faults by common cause, trace root causes through connected equipment, estimate the energy and cost savings of each change, and rank the resulting projects by dollar impact. Engineers receive a prioritized list of diagnosed changes, typically adjustments to setpoints, schedules, and control sequences that cost little or nothing to make. Some platforms also write accepted changes back to the BAS within limits operators define.

Because agentic platforms run continuously, they can also catch drift: a platform can check whether a change holds after implementation and flag a regression when an override or a schedule edit undoes the work. When the platform compares energy use against a baseline normalized for weather and occupancy, the same monitoring verifies the savings each change delivers.

Evaluating platforms by resolution

For owners and asset managers, the useful test of any operations platform is how much of the operational performance gap it closes, counted in resolved problems and verified savings. As the LBNL campaign found, an analytics tool produces savings only when someone acts on its findings, so the number of faults a platform detects says little about the savings it returns. The following criteria help compare platforms on those terms:

  • Resolution rate: the share of findings that reach a completed change, and how long resolution takes.
  • Diagnostic depth: whether the platform groups related faults and identifies root causes, or passes each alert to an engineer for triage.
  • Engineering hours: the in-house time required per building per month to operate the platform, with LBNL’s median of 9 hours for FDD as a reference point.
  • Savings verification: whether savings are measured against a baseline after implementation, and how regressions get caught.
  • Fit with existing systems: whether deployment works with the BAS already installed, across mixed vendors and equipment vintages, without the rip-and-replace of controls.

The Agentic AI Buyer’s Guide covers these criteria in more depth.

How Noda closes the gap

Noda is an agentic AI platform for commercial building operations that works as a virtual building engineer, taking on the triage, root cause analysis, and savings verification that engineering teams rarely have the hours to run across portfolios. Noda is deployed across 300M sq ft of real estate owned and operated by leading global firms. Each site connects through an on-site gateway, the Noda LaunchPad, which connects to the BAS already in place over open protocols including BACnet/IP, Modbus, and MQTT. Noda’s fault detection rules turn equipment data into Insights, and on a regular schedule the AI Project Generator, Noda’s multi-agent pipeline, groups each building’s recent Insights into Projects with root causes identified and annual savings estimated. Command Center puts Projects in front of engineering teams in priority order, ranking first by urgency, a measure that combines equipment health impact with opportunity size, and then by estimated annual savings. Once a Project is complete, Noda measures energy use against the Project’s baseline and reports Realized Savings. Engineering teams start from diagnosed, ranked Projects, and their hours go to carrying out the changes that close the operational performance gap.