The best yes is a faster no: why AI's first use case in investment was ruling out a project

Miguel Galatas, who introduced himself as nTeaser's business development director, opened the investment panel at UsefulAI 2026, in Madrid on 3 June, with the inevitable question: what was the biggest revolution AI was bringing to their products? Arnau Tibau Puig, CTO of Tetrax, answered with a specific case. A German investor, a project in Spain, and a report delivered in hours instead of the weeks of data room work it used to take.

> "In that case it was a decision not to invest, but what would previously have taken literally a few weeks of an adviser analysing a data room, we were able to produce a report in a few hours that supported that decision."

The first use case a supplier chooses to tell you about is the one that has worked best for it. And what worked best for Tetrax was backing up a no.

You pay for due diligence in full even when the project falls through

Here are the sums almost nobody does when someone is selling you productivity. The technical report, the lawyers' hours, the site visits and the committee time are already paid for by the time the no arrives. And the later it comes, the more that spending hurts.

Tibau Puig explained why the no takes so long, using a figure he learned from his first pilot projects:

> "2,000 documents in an order that somebody has thought appropriate, from a project that may have been in development for 6-7 years, that has even been through several hands, through different administrative milestones, through different regulatory phases or even different laws in force. And understanding what is going on is a circus."

He ruled out the shortcut you are thinking of in his very next sentence: "If you take this and give it to ChatGPT, to Claude, any of them, you won't be able to draw any conclusion."

That messy file is what sets the timetable. Juanma Presa, director of growth and digital tools strategy at ONE Hub Energy, put it in months. The average M&A deal in the sector (buying or selling an asset) takes between six and nine, and with AI "we are already seeing reductions to under three months". That is his own estimate of the market, so treat it as one. The mechanism, though, you can check against your own P&L: six months of process is six months of an expensive team working on an asset that may come to nothing.

More options, same project

Pablo Martínez, country manager for Iberia at Modo Energy, had framed it as a multiplier: "where before you and your team could get to two or three projects, you'll be able to start covering ten projects". Later in the panel he put a number on it himself: ten times the projects with the same team. That is an easy case to make to any investment committee.

Irene Negro, lead principal consultant at Rated Power (part of Enverus), qualified it:

> "The idea really isn't so much to increase the pipeline by 10, which if you can, well, great […] but above all to be able to analyse all the possible options and all the possible scenarios within a single project."

Then she got down to the operational detail: "for a plot, give me all the possible options, and with that option I either move forward with the project or decide to drop it at an early stage".

In budget terms, AI lets you work through every option on a plot in an afternoon and drop it before you spend the first euro on advisers. For Negro, that is exactly where the value lies. She had put it earlier as the question every committee will recognise: "is it worth putting capital in here, or dropping the project right from the start?"

On his own patch, Martínez added that "the fundamental models we use and the way we forecast future prices has barely changed". His methodology, he said, is essentially no different from what it was a few years ago. If forecasting has stood still, the new value is in how fast you can rule things out.

What a late no costs in Spain

If you are making decisions in Spain right now, the market has already put numbers on it. On 21 September UNEF, Spain's solar PV association, said the list of lenders financing storage in Spain was empty. It also estimated that between 9,000 and 10,000 MW of solar capacity had entered the market with no PPA, no regulated remuneration and no auction, exposed to zero prices. Its director general, José Donoso, is asking the ICO, Spain's state-owned lender, for a credit line of about €315 million a year to cover debt service. The association places the way out in 2028-2029.

Behind those megawatts sits a distribution of hours, and that is where you get paid or you don't. In the first half of 2026, Spanish solar PV's capture price came in at around 46% of the average daily market price. Close to 1,300 hours were priced at €5/MWh or less.

In storage the concentration is tighter still. Josefin Berg of S&P Global said at an AleaSoft seminar on 17 September that around 10% of days deliver 60% of annual revenue from balancing services in Spain, the services the system operator pays for to keep supply and demand matched. She made the international comparison in dollars. A four-hour battery would have earned close to $700,000 per MW in Spain in 2025, compared with less than $200,000 in Great Britain.

When value is concentrated like that, the gap between a good asset and a mediocre one comes down to a handful of hours and days you have to get right. Sweeping every option on a plot in an afternoon tells you whether any of them works at those numbers. When none does, you know before you commission the first report.

A fast no is only worth something if you can defend it

That no rests on a layer beneath the model. Tibau Puig pointed to where his company's budget goes: "the added value is in all the pre-processing you have to do to be able to give the right information and context to that LLM". Structuring 2,000 messy documents so that an agent can work on them effectively is the part you will not see in any demo.

When he was asked how you guarantee that what your AI produces will hold up in front of a bank, a fund or a judge, Tibau Puig changed the yardstick:

> "what is important to bear in mind is that the benchmark is not perfection […] today an adviser's legal reports have errors and they aren't reproducible either. You ask two different advisers for this and you'll get two different reports."

Straight away he supplied the other half. AI has to be held to a higher standard, "since there isn't a person and there isn't any accountability involved there".

For a no that decides where the money does not go, that standard is called traceability. Negro described the starting point. Investors commit capital on the strength of figures whose origin they sometimes do not know: "whether it's the layout, the yield […] you have to take it on trust, you invest or you don't". With traceability, the investor, or the lender's technical adviser, can follow the thread back to the source data, and that is what turns your fast no into one that survives a committee.

Asked what AI will not be able to do in the sector over the next five years, Tibau Puig said the need for human judgement does not go away, and that "this last mile of judgement, of context, of expertise, will be even more relevant than before". For a rejection, that means the machine reaches the no sooner, but someone still signs it. None of the four promised that the machine would close the deal. Negro drew the line in one sentence: "AI will be able to optimise processes, but I don't think it is going to sign the final decision."

A no in hours gives you back capital, people and time for the yes that is really out there. With the banks out of storage, according to UNEF, and between 9,000 and 10,000 MW of solar exposed to the market, it is the saving most badly needed this year. Clearing out the portfolio before it eats up the due diligence budget is, for now, the use case with the clearest sums.

"More capital, less friction" panel, RENMAD UsefulAI 2026, Madrid, 3 June 2026. Moderator: Miguel Galatas (nTeaser). Panellists: Arnau Tibau Puig (Tetrax), Irene Negro (Rated Power, part of Enverus), Pablo Martínez (Modo Energy) and Juanma Presa (ONE Hub Energy).