Battery energy storage systems are dispatched by optimisation engines that chase the best combination of arbitrage, frequency response, and capacity payments across a forecast horizon. Almost all of these engines share a quiet assumption: that a battery degrades in a smooth, predictable way that can be captured by a single curve plotted against cycle count. That assumption is increasingly the thing standing between operators and the revenue their assets are theoretically capable of earning.
Real lithium-ion degradation is not one process but two running in parallel. Calendar ageing eats away at capacity simply from time spent at a given state of charge and temperature, regardless of whether the cell ever cycles. Cycle ageing depends heavily on depth of discharge, C-rate, and the state-of-charge window a cell sits in while it cycles — a battery worked hard between 20% and 80% SOC ages differently to one worked the same number of cycles between 45% and 55%. The two processes also interact: a cell held at high SOC in hot weather calendar-ages faster, and that faster fade changes how it responds to the next cycle. None of this is linear, and very little of it is captured by the single “cycles to 80% SOH” curve that shows up in a data sheet.
Most commercial dispatch optimisers still treat degradation as a static cost adder — a fixed £/MWh throughput penalty baked in to make the optimiser reluctant to over-cycle. That’s a reasonable simplification for a first pass, but it produces two failure modes in practice. First, the optimiser can under-value cycling early in life, when the battery is genuinely cheap to cycle, and over-value it later, when the same throughput is doing disproportionate damage — leaving money on the table in year one and accelerating fade in year four. Second, because the model doesn’t distinguish between a shallow high-frequency cycle for FFR and a deep cycle for arbitrage, it can’t tell the difference between a MWh that costs the asset almost nothing in lifetime terms and one that materially shortens it. The optimiser ends up solving the wrong problem precisely.
The commercial consequence is a widening gap between forecast and actual state of health. Revenue models built on schedule assumptions from an inaccurate degradation curve misprice the asset’s remaining useful life, which matters directly for augmentation timing, warranty compliance (most OEM warranties are throughput- and DoD-conditional, not just calendar-based), and how the project is valued at refinancing. An operator can be perfectly compliant with the optimiser’s own limits and still find the battery has aged faster than the model assumed, because the model never distinguished the kind of cycling the asset actually did.
The fix gaining traction across the industry is to replace the static penalty with a physics-informed or semi-empirical degradation model — often derived from electrochemical models like the Doyle-Fuller-Newman framework, or from OEM cell test data reduced into a lookup surface across SOC window, C-rate, and temperature — embedded directly inside the dispatch optimiser rather than bolted on afterwards. This turns degradation from a fixed constraint into a decision variable the optimiser actively trades off against revenue, cycle by cycle. Early deployments report meaningfully better alignment between forecast and realised SOH, and dispatch schedules that shift load toward gentler SOC windows during low-price periods and reserve harder cycling for the highest-value dispatch windows.
It’s a less visible upgrade than a new revenue stack or a bigger inverter, but for operators watching realised degradation drift away from the business case, it’s often the highest-leverage fix available — because it corrects the assumption everything else in the optimisation stack is quietly built on.
