What a battery's efficiency number leaves out

Round-trip efficiency is the number that decides whether a storage project makes money. It's the fraction of the energy you put in that you get back out, so an arbitrage business plan multiplies straight through it. A point or two moves millions of dollars over a project's life.

Which is why it's strange how casually the number gets quoted. A datasheet says 95 percent. A pitch deck says "greater than 90 percent round-trip." Nobody asks the question a measurement scientist asks first, which is not "how efficient is it" but "efficient between which two points, measured over what."

I've written here about a battery that returns about half of what you feed it and is still the right choice, and about how the job you give a pack decides how fast it wears out. This one is narrower and more practical: what the efficiency number on the page actually contains.

The definition already tells you

The Department of Energy's own characterization work is unusually clear about this. In the Pacific Northwest National Laboratory report that the storage industry uses as a reference for technology and cost assumptions, round-trip efficiency is defined as "the ratio of net energy that is discharged to the grid (after removing auxiliary load consumption) to the net energy used to charge the battery (after including the auxiliary load consumption)."

Read that twice. The auxiliary load is subtracted from what you get out and added to what you put in. It gets counted against you at both ends, because that is honest: it's energy the site consumed. And the same report lists what auxiliary loads are, in plain language: "heating, ventilation, and air-conditioning (HVAC), battery management systems (BMSs), PCS controls, and pumps (for flow batteries)."

So the standard definition includes the air conditioner. The problem is that a great many quoted numbers do not, and the report shows this happening in its own literature tables. In the lithium-ion cycle-and-efficiency table, one industry source contributes a range of 90 to 98 percent, and the note beside it reads: "Not including auxiliary loads." A few lines up, PNNL's own grid-scale testing at utilities lists 87 to 91 percent, with an AC-to-AC round trip of 83 to 87 percent. Same technology, same table, roughly ten points apart, and the entire difference is bookkeeping.

The flow-battery section makes the point even more sharply. One source puts generic flow batteries at 70 to 75 percent "when battery system auxiliary load is included," while PNNL's testing of a vanadium flow system "has shown an all-inclusive RTE of 65 percent at the 4 h rate." The word doing the work in both sentences is "included."

The same battery, measured every way

The best illustration I know of is a field test rather than a literature table. PNNL instrumented a 2-megawatt, 4.4-megawatt-hour lithium-ion system installed near Puget Sound Energy's Glacier substation and ran it through baseline tests and real duty cycles: arbitrage, frequency regulation, capacity, load following.

Across all the tests performed, counting every loss source, the measured round-trip efficiency ranged from 37.0 to 87.6 percent.

That is one battery. Not a technology class, not a range across vendors. One pack, one site, one test campaign, and a spread of fifty percentage points depending on what it was asked to do and what you counted.

Then the line that should be printed on the inside of every storage pro forma. Reporting the capacity-test results, the authors note: "For perspective, the RTE when just considering PCS losses is 90 percent for the C/6 and C/4 rates and 93 percent for the C/2 rate." So if you measure only across the power conversion system, ignoring everything the site consumed to keep the batteries alive, the very same equipment reports 90 to 93 percent. That is the number that ends up on a slide.

"This supports the conclusion that there is no single RTE that represents BESS performance. Assuming an artificially high RTE makes the BESS more attractive than it actually is."

Where the missing energy goes

Two mechanisms account for most of the gap, and neither is exotic.

The first is idle time. A battery doing energy arbitrage spends most of its life doing nothing, waiting for a price spread, and it is not free to sit there. In the arbitrage tests, rest accounted for 65 to 90 percent of the total test duration, averaging 81 percent across four runs. Measured with that rest included, arbitrage round-trip efficiency came in at 61 to 78 percent. Take the rest back out of the calculation and it rises to 85 percent. Roughly twenty points of efficiency were consumed by a battery sitting still.

The report is specific about what the sitting costs. During rest, the DC battery was powering both the auxiliary load and the power conversion system, "with average power for each being 10 kW," which worked out to a state-of-charge drop of 0.25 percent per hour for auxiliary load and another 0.25 percent per hour for the inverter switching. At times the draw during rest hit 100 kW, dropping charge at 2.3 percent per hour. A pack left connected and idle for a long weekend is quietly eating its own stored energy.

The second mechanism is power level. Auxiliary load is close to a fixed overhead, so the less energy you move, the larger a share of it that overhead takes. The Glacier data shows this cleanly: efficiency rose with C rate, and the frequency-regulation tests, which move small amounts of energy at low average power, returned 54 to 75 percent at average powers of 130 to 220 kW, against 80 to 82 percent for the capacity test at around 600 kW. The authors identify a sweet spot in the C/4 to C/2 range that "balances electrochemical losses vs. losses due to auxiliary load."

There's a seasonal fingerprint too, which is the part that surprises people. Thermal management ran off a 22°C setpoint with heat pump units outside each container, and the auxiliary load tracked temperature well enough to measure. At the C/6 rate, the gap between efficiency with and without auxiliary consumption was about 2 percentage points for a baseline test run in January, but only 0.5 points for a test run at the end of September. In this climate, on this system, the slope for heating was 2.5 to 4 times that for cooling. The same battery is a different battery in winter.

What this does not mean

I want to be careful here, because a 37 percent figure is exactly the kind of number that gets torn out of context and waved around.

These are 2019 reports on a system commissioned years earlier, using that era's controls, thermostat-based container cooling, and an inverter that sat in a switching state during rest when it did not need to. Modern systems are meaningfully better: liquid cooling instead of ducted air, smarter idle behavior, better balancing, and the LFP chemistry that took over the grid is more tolerant of the conditions that drove some of these losses. Nobody should read "37 percent" as a current fleet number, and I'm not offering it as one. The low end of that range came from the least favorable combination of low power and long rest.

The durable finding is not the value. It's the structure. Auxiliary load is a real, measurable, seasonally varying energy consumer that sits outside the cells, and whether it appears in your efficiency number is a choice someone made before handing you the number. The lab said it plainly: there is no single round-trip efficiency that represents a battery's performance. That was true in 2019 and it is true of a container shipping today, because it follows from where the meters are, not from the chemistry.

This kind of thing, storage measured and scheduled against what a site actually does rather than what a datasheet promises, is the work I do on the energy side, and I'll be exact about my role: I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator. I don't own that company and earn nothing from this link; I flag it because it's a field I build in, not just write about. Full policy here.

The signal

A spec is a measurement taken under specific conditions, not a promise, which is the same argument I made about reading a spectrum. Efficiency is just a spec with money attached.

So when someone quotes you a round-trip efficiency, three questions recover most of the truth. Between which points was it measured, DC to DC across the battery or AC to AC at the meter where you actually get billed? Does it include auxiliary loads, the HVAC and controls and pumps, or was that energy left outside the boundary? And at what power and how much idle time, since a battery that sits still most of the year is paying an overhead your spreadsheet probably assumed was zero?

If the answer to any of those is a shrug, the number isn't wrong. It just isn't about your project.