Enersponse managed more than 300 MW of electricity consumption across over 3,800 customer sites in California during its 2024 programme year. When independent evaluators measured the demand reductions its customers provided during grid events, they estimated an average of 34.3 MW of response per event hour. Under specified August grid-planning conditions, the same portfolio was estimated to be capable of delivering 32.8 MW.
Those numbers might tempt someone to conclude that nearly 90% of the flexibility had disappeared between identifying the resource and delivering it. That would be a serious misunderstanding. The 300 MW describes the underlying electricity consumption of participating customers, not the amount those customers had agreed to curtail. The 34.3 MW represents measured reductions assembled into an average-event-hour estimate, while the 32.8 MW is a modelled planning capability under standardized conditions. Each answers a legitimate question, but they cannot be treated as consecutive stages in a conversion funnel.
The figures come from Demand Side Analytics’ April 2025 evaluation of Enersponse’s demand-response resources during October 2023 through September 2024. The portfolio included commercial, industrial and agricultural customers across the territories of Pacific Gas and Electric, Southern California Edison and San Diego Gas & Electric. Enersponse participated in California’s Demand Response Auction Mechanism, Demand Side Grid Support programme and resource-adequacy arrangements. Its resources were dispatched in response to instructions from the California Independent System Operator, CAISO.
This is a useful case because it provides more than a company’s announcement about installed or enrolled capacity. The evaluators examined actual electricity-meter data, operating conditions, dispatch records, customer characteristics and different methods for estimating the electricity customers would otherwise have consumed. They then developed projections for how the portfolio might perform under specified planning conditions.
The evaluation establishes that commercially organized demand response can provide meaningful electricity-system services. It also illustrates how much information is needed before a broad claim about flexible resources becomes a defensible statement about available capacity, delivered response or future performance.
The difficulty becomes apparent when the portfolio is examined in more detail. More than half of Enersponse’s participating sites were supermarkets and other retail facilities, but most of the measured demand reductions came from fewer than 1,000 agricultural and pumping sites. Counting customers was a poor proxy for counting flexibility. A large collection of relatively small retail loads contributed differently from a smaller group of customers with substantial electricity-consuming equipment that could be interrupted or rescheduled.
That is not evidence that supermarkets are inherently poor flexibility resources or that agricultural pumping always offers a better service. The operating circumstances differ. Refrigeration, lighting, ventilation, cooling and other commercial loads are governed by customer requirements and equipment controls. Agricultural pumping depends on water requirements, irrigation schedules, weather and pumping-system configurations. Even within one customer category, location, temperature, timing and operating practices can change the useful response.
The distinction becomes particularly relevant when an investor or policymaker starts with the size of a customer population. Thousands of sites, hundreds of megawatts of underlying consumption and a substantial annual electricity bill may make an attractive presentation. They establish that equipment and demand exist, but not how much can be changed when the electricity system requires a particular service.
Enersponse’s actual dispatch records expose another problem with interpreting aggregate figures. CAISO did not instruct the entire portfolio to respond simultaneously during the evaluation year. Different resources were called on different days, for different durations and in different electrical locations. An event on August 27, 2024, involved 2,639 dispatched sites and produced an estimated 28.26 MW of demand reduction. It was the broadest dispatch of the year, but it still excluded some resources and programmes.
The evaluation’s 34.3 MW average-event-hour figure was constructed by estimating performance at individual sites during their dispatched hours and aggregating those estimates. That makes it useful for understanding portfolio performance across the programme year. It does not mean that every site was observed delivering its contribution during one common hour, or that the portfolio’s entire capacity had been demonstrated under a simultaneous stress event.
The evaluators identified this limitation themselves. They recommended that Enersponse arrange at least one annual test in which all resources are dispatched during the same event hours. They also recommended introducing deliberate variation in event timing, duration and conditions so that the programme could better establish how the portfolio performs under different circumstances.
That recommendation is more important than it might appear. Aggregation can reduce the uncertainty associated with individual customers, but it does not eliminate the need to know what the combined portfolio can deliver. A collection of measured responses from different days is evidence of operating capability. A coordinated test under a relevant common condition would establish something different. Neither should be presented as evidence of every operating circumstance the portfolio might encounter.
The planning estimate provides a further distinction. Demand Side Analytics estimated that Enersponse could provide 32.8 MW of demand reduction under August conditions with weather expected to occur about once every two years, assuming full dispatch of the relevant resources. The evaluators used measured performance and models that incorporated weather, customer characteristics and other factors. This was an estimate of capability under defined assumptions, not another measurement of electricity actually curtailed during 2024.
The measured 34.3 MW average-event-hour estimate was slightly higher than the modelled 32.8 MW planning estimate. That is not evidence that the portfolio overperformed a 32.8 MW contractual obligation. The numbers refer to different operating conditions and calculations. Their relative closeness provides useful information about the consistency of the evaluation and forecasts, but it does not remove the need to assess particular weather conditions, resource availability, event duration and performance requirements.
The same distinction applies to projections of future capacity. The study estimated that the portfolio’s planning-condition response could increase to approximately 39.8 MW by 2026, assuming annual growth of about 10% in enrolment and resource development. That was a forecast published in 2025, not an observation of 2026 delivery. A company might achieve the forecast, exceed it or fall short as customers join, leave or change their operations. An investor should not give projected resource growth the same evidential weight as measured performance.
California’s resource-adequacy arrangements add another layer. The evaluation reports that Enersponse had been awarded 30 MW of net qualifying capacity for October 2023 through June 2024 and 40 MW for July through September 2024. It also reports an award of 10 MW for the relevant 2025 summer period. Those figures establish the amount recognized through particular qualification and procurement arrangements for specified periods. They are not measurements of how much demand response occurred during every dispatch event.
The change in awarded qualifying capacity cannot be interpreted from this evaluation alone as the physical disappearance of 30 MW of flexible demand. Qualification, procurement decisions, programme participation and commercial arrangements can change independently of the equipment at customer sites. It would be just as inappropriate to treat an award of 40 MW as proof that 40 MW was available and delivered in every qualifying hour.
This is why the distinction between physical potential and operating service matters. The Flexibility Mesh separates eleven delivery states, from theoretical capability through connection, controllability, participation, qualification and network permission to availability, activation, verified response and repeatable performance. The states identify different claims that must be supported by appropriate evidence. They are not eleven successive losses from a single original quantity.
Capability, permission, availability and verified response are different claims, not an attrition funnel.
Flexibility Mesh animation tracing physical capability through enrolment, qualification, network permission, availability, activation and verification. It represents distinct service evidence states rather than percentages of capacity lost.
An installed battery may be technically capable of discharging into the grid but lack permission to export. A fleet of vehicles may be compatible with managed charging while only a portion is connected and available during a particular operating window. An industrial site may have equipment capable of reducing demand but no contract, control system or operating procedure that allows the system operator to rely on it. Those are materially different reasons why technical capability has not become a usable service.
Some of those gaps represent investment opportunities. Better controls, clearer network rules, improved customer agreements and lower integration costs can make existing equipment useful without adding a corresponding amount of new physical capacity. Other gaps are enduring limits. A refrigerated warehouse cannot promise unlimited curtailment without affecting its primary function, and a vehicle cannot provide its battery to the grid without preserving sufficient energy for the owner’s next journey.
The available resource is the portion that can meet the named service during the required operating period. That is still not the same as the resource actually instructed to respond. A system operator may maintain a substantial pool of qualified and available capacity while calling only a fraction of it because the requirement is smaller, another resource is less expensive or reserves are being held for a contingency.
An available resource that was not called has not failed to deliver an instruction. Its availability may still have value, especially where readiness is itself a contractual obligation. By contrast, a resource that receives a valid dispatch instruction and provides insufficient response, responds too late or cannot sustain the required duration has a performance problem. Measuring those two situations using the same denominator would confuse procurement utilization with operational reliability.
The Flexibility Mesh makes this distinction explicit. Its available-to-activated stage measures dispatch utilization. The next stage measures response against the instruction, including magnitude, timing, duration and recovery. Repeatable performance asks whether the resource has demonstrated the relevant service over an appropriate observation window, rather than relying on a successful isolated event. A resource contracted for rare emergencies may require periodic readiness tests even when the system has no reason to activate it for years.
California’s broader operating evidence reinforces the need for that discipline. In its September 2026 report on demand-response performance during 2025, CAISO’s Department of Market Monitoring reported approximately 1,330 MW of demand response counted toward system resource adequacy, about 3.1% of the total. On days with strained grid conditions, approximately 77% of scheduled demand-response capacity in real time was reported as performing as scheduled.
That figure concerns a different resource population and a different performance denominator from the Enersponse evaluation. It should not be applied as a discount factor to Enersponse’s 32.8 MW planning estimate or used to claim that all California demand-response resources had a uniform 23% failure rate. It does demonstrate why market operators track instructed performance independently of qualifying capacity, and why the proportion delivered against a schedule matters for reliability.
There are several other reasons a portfolio can appear larger on paper than it proves to be under a specified service. A customer may participate in more than one programme, with obligations that cannot always be fulfilled simultaneously. A battery may reserve part of its state of charge for resilience or a different contract. An industrial process may have to recover production after curtailment, while a thermal load may create rebound demand. The electricity system cannot credit the same resource, stored energy or customer action in full to several incompatible commitments.
The Enersponse evaluation is useful in part because it acknowledges the complexity of estimating delivered response. Demand reduction is not measured by observing a separate stream of electricity flowing out of an appliance. It is calculated against an estimate of the consumption that would otherwise have occurred. Weather, agricultural water conditions, customer operations and normal load variation affect that calculation. The evaluation used site-level statistical modelling and additional data sources to improve the estimates, while recommending further operating tests to strengthen the evidence.
For a utility purchasing flexibility, the appropriate evidence depends on the service. A local network constraint requires resource availability at the affected electrical location and time. Resource adequacy requires dependable contribution during defined scarcity conditions. A balancing service requires response under the applicable dispatch rules. A portfolio can be useful in all three settings, but performance in one does not automatically establish its contribution to the others.
An investor examining an aggregator or flexibility platform should ask for similarly explicit records. The starting point is the physical customer and asset inventory, with duplicate records and overlapping service rights reconciled. The next questions concern how much is connected, controllable, enrolled and qualified for particular services, followed by how much capacity is actually available during the relevant operating windows. Dispatch histories and independent performance measurements then establish the evidence for delivered service.
The analysis should distinguish customers leaving a programme from those temporarily unavailable. It should show how available capacity has been used, how resources have responded when called and whether performance has remained credible as the portfolio changed. Forecast additions, market awards and future recruitment targets belong in the assessment, but not in the same column as commissioned, operating and verified capability.
There is a commercial consequence, although it should not be confused with the accounting problem examined in last week’s article. A technically possible megawatt has no automatic claim on procurement revenue. A qualified megawatt is closer to participating in a market, but may still face limited activation, incompatible commitments, weak customer economics or insufficient payments. Proven performance makes a business case more credible without guaranteeing that the contract will cover acquisition, controls, settlement, customer compensation and financing costs.
Better measurement can reduce an inflated market estimate. That is healthy when the alternative is investing on the assumption that every connected device is a dependable resource. Better controls, customer propositions and market rules can also expand the real market by making previously inaccessible resources genuinely deliverable. The objective is not to force the measured market downward, but to distinguish useful future opportunities from capability that can already be relied upon.
The California case demonstrates why these distinctions are worth making. Enersponse had access to more than 300 MW of underlying customer demand, and its independently evaluated portfolio delivered meaningful reductions under actual dispatch conditions. The 34.3 MW average-event-hour estimate and 32.8 MW planning-condition estimate provided defensible measures for particular purposes. Neither was a measurement of a simultaneous full-portfolio event, and the evaluators recommended further testing to close that evidence gap.
The more useful flexibility market is the one described in terms of named services, compatible commitments, operating conditions and defensible performance evidence. It may be smaller than an estimate constructed from every theoretically controllable battery, vehicle, building or industrial process. It is more valuable to decision-makers because it can support serious planning, procurement and investment judgments.
The Flexibility Mesh report and its interactive architecture provide a reference for tracing the relationships from physical capability through control, qualification, network permission and verified delivery. They do not accredit individual projects or forecast the profitability of an aggregator. That work requires the service-specific operating and commercial evidence discussed here.
Next week’s article will examine where much of that usable flexibility originates and why its limits matter. Industrial facilities, commercial buildings, data centres and electric fleets exist to provide production, services, computing, comfort and mobility. Their flexibility comes from the operating room left after those primary functions are protected. Counting that room accurately is the beginning of a credible flexibility market.
This is the third article in the Flexibility Mesh series. TFIE Strategy provides independent assessment of electricity-system strategies, flexibility portfolios and investment assumptions. For discussion of a specific programme or investment thesis, contact michael@tfie.io.
The 300 MW enrolled vs 34.3 MW measured gap is the same pattern I keep seeing in data-center power announcements — headline capacity vs what is actually scheduled to be delivered. Generac's Amazon deal was announced as $8B, but the SEC filing schedules only $2.4B of deliveries across 2027-28; the rest is a ceiling, with warrants vesting as Amazon actually spends. Your Enersponse case makes the same point from the grid side: the number that matters is the measured, dispatched one, not the enrolled one. Is there a standard disclosure you'd want — measured average-event-hour alongside every enrolled MW claim — or would that just get gamed into a different headline?