Storage dispatch optimization software succeeds when every charge-and-discharge decision trades off competing objectives — arbitrage revenue, cycle degradation, and grid-service obligations — instead of chasing one revenue-maximization number. A dispatch engine that only chases the largest hourly price spread builds assets that ace a backtest and fail a warranty claim within two years.

Quick answer: Treat battery dispatch as constrained multi-objective optimization — arbitrage, degradation cost, and grid-service commitments, bounded by hard operational limits — and surface every trade-off as an explicit product decision before an engineer encodes a solver. Ship a decision-explainability layer alongside the dispatch signal itself, because operators will not trust a black box managing a multi-million-dollar asset.

Why "Maximize Revenue" Is the Wrong Spec for Dispatch Software

A dispatch algorithm told to simply maximize revenue will find the mathematically optimal way to burn through a battery's warranted cycle life for a few extra dollars of arbitrage spread. The correct product spec is a constrained optimization across arbitrage, grid-service revenue, and degradation cost, bounded by contractual and hardware limits.

Most first-draft BESS product requirements documents still describe dispatch with a single adjective — "smart," "optimal," "revenue-maximizing" — as if there were one number to climb. That framing hides three separate, partially conflicting objectives inside what looks like one knob.

Energy arbitrage is a bet on future price spreads that depend on volatile day-ahead and real-time market forecasts — the same forecasting fragility explored in our piece on AI energy demand forecasting and megawatt errors. Degradation is a real cost that never appears on this month's P&L but shows up years later as capacity fade and warranty disputes. Grid-service commitments — frequency regulation, capacity payments, resource adequacy — carry contractual availability obligations that constrain when and how much state of charge (SoC) can be spent chasing arbitrage.

The multi-revenue reality regulators created

This tension exists because storage assets are now allowed to stack revenue streams in the first place. FERC Order 841 (2018) directed U.S. wholesale markets to open participation models for storage, which is precisely what let a single battery earn arbitrage, capacity, and ancillary-service revenue simultaneously. Rocky Mountain Institute has written extensively on this "value stacking" reality, and its core finding is simple: the more revenue streams a battery stacks, the more its dispatch schedule is constrained by each stream's rules, not freed by the extra revenue.

A PM who treats dispatch as one optimization target is implicitly assuming those constraints away. They don't disappear — they just get discovered later, usually by an asset manager reading a warranty exclusion clause.

The Objective Function: Turning Trade-offs into Product Requirements

Good dispatch requirements name every term in the objective function explicitly — arbitrage revenue, service payments, and a degradation cost curve — rather than hiding them behind a vague instruction to "optimize." The PM's job is to make the trade-offs a product decision, not an implementation detail an engineer infers.

Use this as a working framework when scoping a dispatch feature or reviewing a vendor's optimization engine:

  1. Name every revenue and cost term. Arbitrage ($/MWh spread), regulation mileage payments, capacity clearing revenue, and a degradation cost term expressed in dollars per equivalent full cycle (EFC) or per unit of throughput.
  2. Attach a real degradation cost curve, not a flat number. Degradation is nonlinear with depth of discharge (DoD), C-rate, temperature, and state-of-charge dwell time — a flat cents-per-cycle estimate understates the cost of aggressive shallow cycling at high SoC.
  3. Define the hard constraint set before the soft objectives. State of charge operating band, maximum daily cycles, reserved capacity for must-offer commitments, and any interconnection limits are non-negotiable inputs, not optimization outputs.
  4. Choose an explicit ordering or weighting. Decide, and document, whether grid-service commitments are a hard constraint (must always be met) or a soft objective that competes for capacity with arbitrage — this single decision changes the dispatch engine's behavior more than any tuning parameter.
  5. Require a decision-attribution output. Every dispatch interval should be able to report which term in the objective function drove the action — this becomes the explainability layer covered later.

How the objectives actually compare

Revenue streamWhat drives the price signalTypical degradation pressureConstraint the PM must productize
Energy arbitrageDay-ahead / real-time price spreadHigh if pursued via frequent, deep cyclesSoC reserve band; max cycles per day
Frequency regulationISO regulation mileage / performance paymentHigh-frequency shallow cycling, elevated C-rateSub-second to few-second response time; availability %
Capacity / resource adequacyCapacity market clearing priceLow in normal operation; battery mostly on standbyMust reserve a minimum SoC through the peak window
Ancillary reserves (spin / non-spin)Reserve clearing priceLow to moderate; rarely triggers full dischargeAvailability commitment; dispatch notice time

The table's takeaway is blunt: the revenue streams that pay the most reliably (capacity, reserves) are also the ones that cost the least in degradation, while the stream with the highest headline upside — arbitrage — is the one most capable of quietly consuming warranted cycle life. A dispatch product that doesn't weigh these differently is optimizing for the wrong column.

The Warranty Trap: How Revenue-Max Dispatch Quietly Destroys Cycle Life

Unconstrained revenue-maximizing dispatch can be economically "correct" against the price signal it sees while still violating the operating envelope a battery's warranty assumes — converting a covered asset into an uninsured degradation liability. This is the single most consequential mistake a dispatch product can make, and it's invisible until an annual capacity test comes back short.

Illustrative scenario: Consider a 4-hour, 100 MWh system whose warranty covers a fixed lifetime EFC budget and excludes accelerated fade caused by operating outside a defined SoC and temperature window. A revenue-max dispatch engine, chasing every small intraday price spread, finds it more profitable to run two or three shallow cycles a day at high average SoC than one deep cycle at a moderate SoC band. The daily EFC throughput can look similar on paper. The actual stress on the cells is not.

Research from NREL and Sandia National Laboratories on lithium-ion aging has repeatedly found that degradation does not scale purely linearly with throughput. Calendar aging accelerates disproportionately when a pack is held at high state of charge for extended periods, and cycling stress compounds with elevated C-rate and temperature rather than adding on top of it.

EPRI's work on battery performance and warranty modeling makes the same point from the asset-owner side. Most commercial storage warranties cap annual EFC throughput and explicitly exclude fade caused by operating outside a specified SoC/temperature envelope, precisely because that operating pattern is disproportionately damaging.

A revenue-max objective function has no term that represents "future warranty claim risk." It will happily trade a small, real arbitrage gain today for a large, deferred degradation cost that lands on the asset owner's balance sheet, not the software vendor's. The fix isn't a smarter solver — it's a degradation cost term that's actually in the objective function, calibrated against the warranty's real exclusions, not an afterthought bolted on once an operator complains.

Constraints as First-Class Product Decisions

Dispatch constraints — SoC operating bands, daily cycle budgets, reserved capacity, hardware-reported limits — are product decisions with commercial consequences, not defaults an engineer picks for convenience. Treating them as configuration surfaces the trade-offs to the people who own the financial and contractual risk.

Build the constraint set as an explicit, per-asset configuration layer, not a hardcoded assumption:

  • State-of-charge operating band. A 10-90% band instead of 0-100% trades a small amount of usable energy for a meaningful reduction in calendar aging — this ratio should be a contract-level setting, not a firmware constant.
  • Cycle and throughput budgets. Daily and annual EFC caps tied directly to the specific warranty terms in force for that asset, not a generic industry rule of thumb.
  • Reserved capacity for must-offer commitments. Capacity-market and resource-adequacy contracts often require the battery to show up during system peak regardless of what arbitrage would otherwise prefer — this is a hard constraint, never a soft preference.
  • Interconnection and telemetry limits. Utility interconnection agreements frequently cap ramp rate or require specific telemetry cadence; getting this wrong is as much a regulatory-relationship problem as a technical one, which is why it belongs in the same conversation as the utility and regulator stakeholder work described in our guide to navigating utility and regulator stakeholders.
  • Hardware-reported limits, not nameplate specs. Battery management system (BMS) firmware often reports a derated capability that differs from the datasheet, and it updates on its own release cadence. Dispatch software has to respect whatever the hardware is reporting today, which is exactly the mismatch our piece on the hardware-software two-clock problem describes — the optimization layer moves fast, the firmware layer doesn't, and the product has to be built assuming that gap is permanent, not a bug to be fixed away.

Every one of these constraints should be visible, editable, and versioned per asset — a fleet operator running fifteen sites under twelve different warranty terms and eight different interconnection agreements cannot be served by one global constant.

Explainability: Why Operators Demand to See the "Why"

Asset managers who are financially accountable for a battery's economic life will not accept a dispatch signal they cannot interrogate — explainability is a retention feature for storage optimization software, not a nice-to-have analytics add-on. Every dispatch decision needs a traceable answer to "why did it do that."

The underlying job the operator is hiring the software to do isn't just "dispatch the battery" — framed through a jobs-to-be-done lens, as detailed in our complete guide to Jobs to Be Done, the deeper job is "protect the asset's economic value while capturing the revenue that's actually available." A dispatch signal that can't explain itself fails that deeper job even when its math is correct, because the operator has no way to verify it without independently re-deriving the decision.

What an explainability layer actually needs to output

  1. Per-interval attribution. Which objective-function term — arbitrage, regulation, capacity reservation — drove this specific charge, discharge, or hold decision.
  2. The binding constraint, if any. If the engine wanted to discharge further but a SoC floor or a reserved-capacity commitment stopped it, that constraint should be named, not silently enforced.
  3. A degradation-cost estimate per decision, not just an aggregate monthly figure, so an asset manager can see the cost side of the trade the software just made.
  4. An audit trail lenders and insurers can read, since project-finance covenants and insurance underwriting increasingly expect evidence that an asset was operated within its warranted envelope.

Trust in the dispatch signal doesn't erode gradually — it drops sharply the first time an unexplained decision costs money. That's exactly the kind of moment mapped in an operator's emotional experience over the customer journey framework: confidence climbs steadily through onboarding and normal operation, then craters at the first opaque, expensive surprise, and rarely fully recovers without a visible explanation mechanism put in place afterward.

DARPA's Explainable AI (XAI) program, launched specifically because operators of consequential automated systems refused to act on recommendations they couldn't interrogate, found something similar in a very different domain. A correct-but-opaque recommendation gets overridden or ignored far more often than an explainable one, even when the explainable version is measurably less optimal on paper.

Building the Dispatch Roadmap: From Objective Function to Shipped Feature

A dispatch roadmap sequences from making the objective function explicit, to modeling the trade-offs it implies, to instrumenting explainability, and only then to giving operators configurable weighting — building the solver before doing this groundwork produces a fast, confident, and quietly dangerous product.

In practice, that sequence looks like this for most BESS optimization teams:

  1. Run a cross-functional objective-function session before writing a single user story — asset management, commercial/origination, and whoever owns the warranty relationship all need to agree on what terms belong in the objective function and which constraints are non-negotiable.
  2. Model the tension explicitly, before engineering encodes it. This is where a causal-loop view earns its keep: mapping how chasing arbitrage revenue reinforces short-term cash flow while simultaneously feeding a balancing loop that accelerates degradation and erodes long-term asset value is exactly the kind of reinforcing-versus-balancing dynamic Prodinja's Systems Engineering causal-loop tool is built to surface, so the trade-off is explicit and agreed on paper before an engineer ever encodes a solver.
  3. Write acceptance criteria as constraints, not adjectives. Replace "the system dispatches optimally" with testable statements: "never exceeds N cycles per day," "always reserves X% SoC during the capacity-commitment window," "produces a decision-attribution record for every dispatch interval."
  4. Roll out in phases, revenue-max backtest first (to establish a ceiling estimate), then the constrained multi-objective version, then the explainability layer, then operator-configurable weighting between arbitrage aggressiveness and degradation conservatism.

The three maturity stages, compared

Each stage of that rollout looks different on every axis an asset owner actually cares about, not just on headline revenue:

Dispatch strategyRevenue captureWarranty / degradation riskOperator trust
Revenue-max (naive)Highest on paper, short horizonHigh — no cap on cycling against warranty termsLow — no attribution, decisions feel arbitrary
Constrained multi-objectiveSlightly lower, more durableLow — bounded by explicit SoC and cycle constraintsModerate — safer, but still opaque
Explainable, configurable multi-objectiveComparable, tunable per assetLow, and auditable against warranty termsHigh — every decision is attributable

The naive strategy only wins on the one metric measured in isolation this quarter. Every other axis that determines whether an asset owner renews a software contract moves in the wrong direction.

This kind of asset — expensive, long-lived, and deeply exposed to physical wear from software decisions — sits squarely inside the broader discipline our energy and climate complete guide covers: hardware-adjacent product management where the software's mistakes show up as capital losses, not just bad reviews.

Key Takeaways

  • Dispatch is multi-objective, not single-objective — arbitrage revenue, degradation cost, and grid-service commitments pull in different directions, and treating them as one "maximize revenue" function hides the conflict instead of resolving it.
  • Name the objective function's terms explicitly in product requirements — arbitrage, service payments, and a real degradation cost curve — so trade-offs get decided deliberately instead of by whatever a default solver configuration happens to produce.
  • Revenue-max dispatch can quietly violate warranty terms by favoring frequent, shallow, high-SoC cycling that looks fine in aggregate EFC throughput but is disproportionately damaging according to NREL, Sandia, and EPRI degradation research.
  • Constraints — SoC bands, cycle budgets, reserved capacity, hardware-reported limits — are product decisions, configurable per asset and per contract, not engineering defaults or firmware constants.
  • Explainability is a trust feature, not a reporting feature — operators need per-interval attribution and a named binding constraint, because an unexplained costly decision breaks trust faster than it can be rebuilt.
  • Sequence the roadmap deliberately: objective function first, trade-off modeling second, explainability third, configurable weighting last — never build the solver before the groundwork.

Frequently Asked Questions

What is dispatch optimization in battery energy storage?

Dispatch optimization is the software layer that decides, interval by interval, whether a battery charges, discharges, or holds — balancing arbitrage revenue, grid-service commitments, and degradation cost against real operating constraints. It's the core decision engine of any battery storage software product, not a single pricing algorithm.

How does battery degradation affect storage revenue?

Degradation reduces a battery's usable capacity and power over time, directly shrinking future arbitrage and capacity revenue while also risking warranty exclusions if the battery was cycled outside its covered operating envelope. A dispatch strategy that ignores degradation cost is borrowing revenue from the asset's future, not creating it.

Can BESS dispatch software violate battery warranties?

Yes — most commercial storage warranties cap annual cycle throughput and exclude fade caused by operating outside specified state-of-charge or temperature bounds, and an unconstrained revenue-maximizing dispatch engine can exceed those limits while still looking financially optimal against short-term price signals.

What's the difference between arbitrage and ancillary-service revenue for storage?

Arbitrage revenue comes from buying low and selling high against volatile energy prices and carries the highest degradation exposure, while ancillary-service and capacity revenue comes from being reliably available and typically carries much lower cycling stress. A well-designed dispatch objective function weighs these differently rather than treating a dollar from each source as interchangeable.

How do product managers make AI dispatch decisions explainable to operators?

PMs make dispatch explainable by requiring the software to output, for every interval, which objective-function term drove the decision and which constraint (if any) was binding — turning a black-box signal into an auditable trail that asset managers, lenders, and insurers can independently verify.