Black box, open books: why banks now audit the algorithm before the battery

The question that shaped the trading panel at RENMAD UsefulAI 2026 did not come from the speakers. Someone in the audience asked it on 2 June: how accurate is the forecasting model against what it predicts, and "how much do the banks believe it". Then they put it in the language a deal is signed in. I have a battery, an optimiser runs it for me, I say I am going to earn this much: "the bank, how much of that does it buy".

Jesús Mora of enspired started with the technical side. In his view, "comparing what the perfect path would have been with the one you have taken is impossible to evaluate". Asked whether the buying decisions could be explained, he did not dodge the question. "The thing is, it's a bit of a black box, because the magnitude, the scale of the decisions that are taken, one by one, are perhaps understandable, but taken together it escapes us," he replied. Rubén Martínez of the Instituto de Ingeniería del Conocimiento got to the same place: "in the end it's a black box and all those decisions it takes, well, you have to believe them".

Two of the three speakers put the same black box on the table, in public, and admitted that their product is not audited decision by decision. The third, Manuel Moliner of BM2Solar, says the opposite: he does open the machine up for the banks. If you finance storage in Spain, that conversation is your problem, because the capacity auctions now have rules.

€9 billion for showing up: the part of the revenue a bank can actually read

Brussels cleared Spain's capacity mechanism under state-aid rules on 29 May 2026: up to €9 billion over ten years, with combined-cycle plants, pumped hydro and storage competing in the same auctions. Order TED/966/2026 of 15 September, the ministerial order that creates the capacity market, was published in the BOE, Spain's official gazette, on the 17th and came into force on 18 September.

The capacity payment rewards available firm capacity. It pays you to be there when the power system calls on you and penalises you if you are not, without asking whether you got it right. Everything else in the revenue stack depends on decisions taken every quarter of an hour: the day-ahead and intraday markets, and the balancing services a Spanish battery earns its money from. Nobody on the panel said they could explain those decisions one at a time. The capacity market hands the bank a slice of revenue that can be contracted and, in doing so, leaves the rest hanging on those decisions.

And the rest is the bigger share. Running a battery with an algorithm rather than by hand can be worth 30% more revenue, according to Mora. Note what that figure measures. It compares two ways of operating the same battery, not two batteries, and it comes from a supplier that sells exactly that. That difference is what the lender has to underwrite without being able to read the trading that produces it.

Banks ask for the manual and walk away with the track record

Banks do ask, and Moliner describes it from the receiving end. "We talk to a lot of banks and they are also very interested in understanding how we do it […] This isn't an act of faith, explain to me how the algorithm works."

No decision-by-decision answer to that question exists. Yet a credit committee still has to put a figure in the financial model and defend it to its risk department, so it looks for something it can check and file. That already exists, as a record: who has operated how many batteries, for how long and with what result, verified by a third party. Mora was talking about the banks. "They're also professionalising in a certain way in order to understand, to be able as well to separate the wheat from the chaff. And now to demand, just as it demands of you that the EPC contractor has I don't know how much track record, well, the optimiser too."

The EPC contractor designs, procures and builds the plant, and its track record has sat in the credit file for decades. The optimiser, the company that operates the battery in the markets, is now measured by the same yardstick as the builder. One step earlier in the chain sits the off-taker, who buys the energy and carries the market risk. Some off-takers only want to work with optimisers that have a proven track record, Mora said, "because in the end it's the one that puts its money on the line". When the party risking its own money screens on track record, the bank is quick to copy the test.

Moliner brought the discussion round to the result: "this isn't about you beating the machine on one trade, on one in particular but rather after a year who is the one bringing you more money".

Mora cited 2,200 transactions in a single day for a single battery, and cautioned that it had been "a pretty good day". Moliner counted 96 periods a day, seven or eight markets and, by his reckoning, not thousands of variables but a hundred or so, updated every fifteen minutes or less. He summed it up: "all of this is a hell of a mess", and it "is beyond any human mind". Nobody goes through a log like that by hand. The year-end account, by contrast, can be audited.

The missing piece, which for now you have to find outside Spain

This is where opacity turns into a gap in the market. If the optimiser's verifiable track record is the new focus of a lender's due diligence, someone has to measure it and certify it. Today that piece is split in two, and neither half is signed by a Spanish firm.

The audited half belongs to enspired, which publishes the actual results of its German battery portfolio, verified by KPMG Austria, and says on its own website that no other optimiser has published third-party-confirmed revenues. It is the company's own claim and should be read as such, though the figures on that website are euros already earned and reviewed by an auditor.

The other half looks at Spain and does the opposite. Modo Energy has launched a benchmark index for the Spanish market, the ME BESS ES. It simulates a 50 MW, four-hour battery with the median profile of Spain's connection queue, run against actual prices since January 2024 and, according to its published methodology, with no ex-post calibration factor. That tells you what the Iberian market would have paid over a given period. It says nothing about what anyone actually earned, and the firm that publishes the index is not Spanish either.

Put the two halves together and you have the figure a Spanish credit committee does not see today: your audited euros divided by the benchmark battery's euros, over the same period and with a third party's stamp on it. That is a ratio, and a ratio can be priced.

And it cannot be imported. Moliner ruled that out: "the algorithm we have in Spain is totally different from the algorithm in Austria […] but the beast inside is different". He described a peninsula unlike the Central European markets, with thinner liquidity in the continuous intraday market and a longer gap between gate closure and delivery, where "you have to stick your neck out an hour before". A track record reaches only as far as the algorithm that produced it, and an audited record in the German market is not enough to finance a Spanish battery. The gap is Iberian, and it is open now, with the capacity mechanism already in force.

What you can demand on Monday

If you are signing an optimisation contract, or looking at one from the credit side, the track record is negotiable and it pays to get it in writing.

• Ask for actual portfolio results audited by a third party, because a backtest on past prices is written by the seller and an audit is signed by someone accountable for it
• Name the benchmark index in the contract and fix the comparison period, so that a good result has a denominator
• Document the rules for human intervention, because every manual correction the machine is not told about leaves a mark on its retraining

That last point came from Martínez. If an operator overrides one of the battery's decisions and nobody tells the machine, "the algorithmic model is going to think that that decision-making was done badly. So that retraining can have a negative impact, in this case, on the efficiency of the battery". Wire the safety net wrong and it degrades the asset it is there to protect. And a track record that does not log those overrides cannot tell the algorithm's own work from the corrections made to it.

The first optimiser to walk into a Spanish credit committee with an audit report under its arm, rather than a backtest, will be carrying something that committee can actually file. It will no longer be asked for an act of faith. It will be asked to open its books, and to show that they add up.

Trading panel with presentations, RENMAD UsefulAI 2026, Madrid, 2 June 2026. Manuel Moliner, CCO of BM2Solar; Rubén Martínez, of the Instituto de Ingeniería del Conocimiento; Jesús Mora, business development at enspired. Moderated by Cintia Hernández, of RENMAD.