Elasticsearch The Power Behind Spotify and Bol.com

You used a search engine to find this article. You used one this morning without thinking about it, the way you breathe. Search is the most invisible technology in your life, which is precisely why almost nobody can tell you how it actually works. So at Money 20/20 Europe I went looking for someone who could, and I found Arno van de Velde standing at the Elastic booth, ready to explain the machinery behind a verb you use fifty times a day.

Van de Velde is a Principal Solutions Architect at Elastic, the company behind Elasticsearch. His opening line reframed the whole conversation: “We’re a Dutch company that was founded as a search company.” That is not a small footnote at an Amsterdam event. The engine quietly running underneath Spotify, eBay and, as he was quick to point out, Bol.com, was born here. What followed was the clearest explanation of search, security and artificial intelligence I have heard at this show, and it ended on two words that should matter to anyone paying an AI bill: fewer hallucinations, fewer tokens.

The Dutch engine hiding under your favourite apps

Elastic does not have a consumer brand the way Bol.com or Spotify does, and that is the point. Van de Velde described a company that powers other companies, mostly out of sight. Its public list of users reads like a tour of the modern internet, and the firm itself was founded in Amsterdam in 2012 before going public on the New York Stock Exchange in 2018. When he reached for an example a Dutch reader would feel in their bones, he picked the obvious one.

“I don’t know anybody living in the Netherlands that doesn’t use Bol at least two or three times per week. You’ve been using their technology without even knowing it.”

Spotify is another he names, and the claim holds up in public: Spotify’s own engineers have written about building natural-language podcast search on top of Elasticsearch. The pattern repeats across the web. You interact with the result. You never see the engine.

How you search a petabyte without melting the building

The first thing I wanted to understand was scale. Handling a retailer the size of Bol is no small trick, so how does one engine search that much data and stay fast? Van de Velde’s answer was refreshingly concrete, and he kept reaching for a library.

Elasticsearch was built distributed from the ground up, he explained, so it spreads across many machines and scales horizontally into the petabytes rather than hitting the ceiling a traditional database eventually hits. But raw speed is the easy half. The hard half is relevance, returning a meaningful answer rather than just a fast one.

Here the library analogy did real work. You do not walk into a library and scan every book; you use the index card. Elasticsearch uses the same idea, an “inverted index”, to jump straight to where an answer lives, and a scoring method called BM25 to decide which words in a sentence actually carry meaning. Rare words tend to be the meaningful ones, so the engine weights them. Underneath, every shard of stored data is a small instance of the open-source Apache Lucene library running its own miniature search, and a node that holds nothing relevant simply drops out of the race. Power gets spent only where the data actually is.

It is an elegant idea dressed up in jargon, and it is the same idea that lets the platform do something far less friendly than helping you find a podcast.

The same logic that finds a song finds the attacker

This was the turn in the conversation I did not expect. Searching for a product and hunting for a hacker, van de Velde argued, are the same problem wearing different clothes. You are sifting millions of records for the meaningful few, fast.

After Elastic acquired the endpoint-security firm Endgame around 2019, it built a full security-analytics layer on the same engine. Pour in network data, firewall logs, endpoint telemetry, and you can ask hard questions across all of it at speed. He gave a number that stuck with me. When the Log4Shell vulnerability detonated across the industry, Elastic scoured about five petabytes of data in roughly seven minutes to ask the only question that mattered in a breach: did this attack touch us, and do we see the pattern anywhere else?

There was a commercial tell in how he framed the pricing, and it is worth noting because it explains the strategy. Elastic does not charge by the number of endpoints you protect. “We don’t particularly care” about that count, he said; the company charges for what you do with the data, which is why a solid tier of endpoint protection is free and open. You can read more on the Elastic Security pages. The free tier is the hook. The scale is where it earns.

Search AI, and the two words that pay for themselves

Then we got to artificial intelligence, and van de Velde said something I have not heard a vendor admit so plainly. For a while, the industry assumed large language models would make search obsolete. The opposite happened. The models turned out to be brilliant at language and bad at knowing things, and the thing they lack is context.

That gap is the entire pitch for what Elastic now calls Search AI. A model on its own guesses. A model handed the right documents by a fast, factual search answers. Pair the two and, in his words, you can “almost diminish the amount of hallucinations to next to nothing.” This is the same retrieval-augmented approach now spreading through every serious AI deployment, and Elastic’s bet is that the search layer is the part that makes the model trustworthy.

The part that should grab anyone running AI at scale was the economics. I pushed him on it, because it sounded too good, and he doubled down:

“You don’t want to spend time, tokens and effort on remembering. We’re talking about milliseconds of searches. It is cost effective, it doesn’t make you burn more tokens, and it is factual. We all know the hallucinatory effect that happens at large scale. What if you could dampen that?”

Two promises, and he knew exactly which two would land: fewer hallucinations and a smaller bill. If you are spending one or two million tokens a week feeding an AI agent, a millisecond search that hands the model real context instead of forcing it to reason from memory is not a nice-to-have. It is the difference between an agent that is reliable and affordable and one that is neither. Elastic has been shipping the lower-memory vector search and inference tooling to back the claim, and it is the reason a search company has a booth at a money show in the first place.

Why this matters for the Dutch and European reader

There are two takeaways here, and the first is a quiet point of pride. One of the engines that organises the modern internet is Dutch, built in Amsterdam, and most people who rely on it every day have no idea. In a tech conversation that usually flows one direction across the Atlantic, that is worth saying out loud.

The second is more practical, and it cuts across the whole AI gold rush. Everyone is racing to bolt a large language model onto their product. Far fewer are asking the unglamorous question that decides whether that model is safe to ship: where does it get its facts, and what does each answer cost? Van de Velde’s argument is that the boring layer, search, is precisely what turns an impressive demo into a system you can actually trust with money and security. For European fintechs building under tightening regulation, where a hallucinated answer can become a compliance problem, that is not a technical detail. It is the whole game.

This interview was recorded live at Money 20/20 Europe 2026 at the Amsterdam RAI. For more conversations with the people building the future of finance and technology, watch the full Cryptofocus series, and read our companion piece on where the billion-dollar crypto deals get made with Royal Park Partners.