Stacking revenue on a grid battery: what actually works in 2026.

When we say "stacked revenue" we mean four specific revenue streams, in the UK context, with direct analogues in ERCOT, CAISO and the Iberian markets: energy arbitrage (wholesale day-ahead and within-day, plus the balancing mechanism); frequency response (Dynamic Containment-Low and -High, Dynamic Regulation, Firm Frequency Response — the contracts that pay the battery to be available to respond to grid frequency excursions); capacity market obligations (a longer-dated contract obliging the asset to be available at peak periods); and ancillary services like reactive support, black-start commitments, and locally-contracted constraint-management products.

Each of these revenue streams pays differently. Each constrains the others. The energy arbitrage stack pays per-MWh delivered; frequency response pays per-MW available; capacity market pays a fixed-availability premium with an in-kind delivery penalty; reactive support is a niche but consistent revenue line in network-constrained zones. The naive assumption is that they can be summed — that the operator can stack all four and collect the sum. That assumption is wrong, and being wrong about it is how operators leave a quarter to two-fifths of available revenue on the table.

If you commit to frequency response, your state-of-charge envelope shrinks. A typical DC-L commitment requires the battery to be sat at 50% state of charge ±20%, with enough headroom in both directions to ride a one-second grid event for a contracted duration. That window is not available for arbitrage. If the day-ahead price profile has a £180/MWh evening peak, the operator who has committed the full capacity to DC-L cannot capture it — they get the FR fee instead, which is materially smaller. Naive operators discover this at settlement, having committed the wrong capacity to the wrong stack three days too early.

Capacity market obligations are harder. They are binary: in any settlement period flagged as a "stress event," the asset must be available and must deliver. Failure to deliver incurs a penalty calculated as a multiple of the clearing-period revenue. An operator who has the asset at 5% state of charge because it spent the morning doing energy arbitrage will find out at 17:43 that they are about to lose more in capacity-market penalty than they made all day in wholesale.

Reactive support and constraint management add a fourth class of conflict — locational. These products are contracted by the DNO or the system operator against specific feeders or grid supply points. A battery that physically lives behind a constrained feeder may earn three to five times the normal locational rate for being available in the right way, but the "right way" couples the SOC envelope to grid voltage in ways that conventional ESS optimisers do not understand. Operators who treat reactive support as an afterthought either turn it off (and lose the premium) or breach it (and lose the contract).

FATHOM Dispatch solves the joint problem every five minutes with explicit, operator-readable conflict-resolution rules. Capacity-market obligations are hard: the optimiser will not produce a plan that breaches them, even at the cost of every other revenue line on the day. Frequency-response commitments are soft-hard: the optimiser will not breach them without operator approval, but it can recommend a buy-out from the market when the joint value of taking a different action exceeds the FR revenue plus the buy-out cost. Wholesale arbitrage is soft: the optimiser will give it up when the joint value of holding capacity is higher. Reactive support and constraint management sit between, with per-portfolio rules.

The rules are not buried in code. They are in the customer's deployment configuration, in plain YAML, and the customer can modify them — within bounds we negotiate at contract signature — without redeploying. We did this because in our first six months we discovered that every operator wanted the rule weights tuned slightly differently, and the only way to keep velocity was to expose the dial rather than fight over it.

Across eight live UK assets we measure a mean revenue uplift of +14.2% versus a strong rules-based baseline, sustained over twelve months. The baseline is not a strawman — it is the incumbent stack that the operator was running before FATHOM, with conservative but production-grade rules for stack arbitration. The uplift figure is calibrated against two independent counterfactuals: a control window (months where the lane was deliberately switched off for one asset at a time and the incumbent stack was reinstated), and a counterfactual model that replays the period with the incumbent's stack on the actual market and forecast inputs FATHOM saw. The two methods give numbers within 0.6% of each other.

Per-asset, the figure ranges from +9.4% (an asset on a heavily constraint-managed feeder where the locational premium dominates and there is less arbitrage opportunity) to +21.7% (an asset with a flexible capacity-market position and good wholesale exposure where the joint optimisation has the most to work with). The range is not a function of asset size; it is a function of how many of the four stacks were materially active.

No optimiser can save a poorly sited battery. Siting decisions still dominate lifetime economics. If the asset is on a feeder with no locational premium, in a zone with weak capacity-market clearing prices, and far from any constrained transmission node, the joint optimisation is working on a smaller pie than it would be elsewhere. We say this in every pilot conversation: if the question is "should we build this battery at all," the optimiser is not the right tool to ask. If the question is "given that this battery is built, what is the best we can do with it," it is.

No optimiser can act on forecasts that do not exist. The joint optimisation is solved against the calibrated quantile bands from LANE-1. If those bands are not calibrated — if the forecast vendor is shipping point estimates with no associated probability distribution — the joint optimisation degrades to a point-based optimisation, and the uplift figure falls toward the industry baseline. We always integrate with FATHOM Forecast for this reason; we will integrate with a customer's existing probabilistic forecaster if it can emit P5–P95 bands, but we will not pretend that a single best-guess forecast is enough.

No optimiser can survive bad market data. The lane assumes that the BMRS, EPEX and ENTSO-E feeds are accurate to within their published latency. When they are not — and we have seen multi-hour delays during balancing-mechanism disputes — the lane falls back to a conservative envelope that gives up some revenue in exchange for not making decisions on stale data. The operator gets a notification on the structured event stream the moment the fallback engages.