Joserra's phone rang on a Thursday at eleven. It was an important client, one of the big-order-every-quarter kind, asking something specific: in the contract they'd signed three years ago, what had been agreed about the returns period? Fifteen days, the client said. Thirty, Joserra thought.
—Let me check and I'll call you back in five minutes.
He didn't call back in five minutes. He called back the next afternoon.
Because the contract was there. That wasn't the problem. It was paid for, it was signed, it was filed. The problem was where. Joserra started with the client's folder on the server. Not there. He tried an old folder called "Contracts 2023", but that year's contracts were half there and half in the inbox of someone who no longer worked at the company. He called the accountant, who had a copy, but the accountant was travelling. In the end an admin found it, in a scanned PDF, inside a subfolder called "Misc signed def FINAL".
It was thirty days. The client was half right and so was Joserra. But the answer had cost a day and a half, a call to the accountant, and the feeling, once again, that in his own company information played hide and seek.
What the company knew and couldn't remember
Joserra's company, a distributor of catering equipment in Valencia, knew a great deal. In its folders were thirteen years of contracts, terms agreed with every supplier, product manuals, warranties, emails with agreements, accepted quotes. All the knowledge of how the business worked was written down somewhere.
The trouble is that "somewhere" isn't a place. It was spread across thousands of documents, named by twenty different people over thirteen years, each with their own logic, or none. "Contract_final_v2", "this is the good one", "send to Pepe". Folders inside folders. Scanned PDFs you couldn't even search by text, because they were photos of paper.
The company, really, didn't have a memory problem. It had a recall problem. It knew everything. It couldn't bring anything to mind at will. And every time it needed to recall something specific — a clause, an agreed price, a condition — it sent a person to search, sometimes for hours, for what it already had but couldn't find.
What he asked for and what he needed
Joserra called me asking to "tidy up the folders". He wanted someone to get into the server, rename the files and put them in their place. A months-long job, hugely expensive, that would last until the first person saved a document in a hurry with a bad name. That is, until the following week.
I proposed something else. Not tidying the documents, but putting on top of them a search that understands what's inside. Something you could ask, in your own words, "what returns period did we agree with Hostelería del Turia in 2023?", and that would answer with the answer and, below it, the exact document it took it from, so you can check.
—Does that exist?
—That's one of the few things today's artificial intelligence does genuinely well.
How it works, no smoke
The idea is simpler than it sounds. You take all the company's documents — the contracts, the manuals, the important emails, even the scanned PDFs, which are first turned into text — and you give a system the ability to read and understand them. Not to memorise them: to understand them, which is different.
When you ask a question, the system doesn't look for loose words like your computer's search, which demands you remember the exact file name. It searches by meaning. If you ask about "returns period", it finds the clause even if the document says "period for the restitution of goods", because it understands they're the same thing. And when it finds it, it gives you the answer in a sentence and shows you the exact passage of the original document, with its name and date, so you can see it didn't make it up.
That last part is the important one, and it's what separates a serious tool from a toy. The system doesn't replace the document or ask you to trust it blindly. It takes you to the document. It does what a colleague would do if they knew thirteen years of the archive by heart: "look, this was agreed here, page four, take a look."
What changed at Joserra's company
We set the search to read the whole archive. It took a couple of weeks to be ready, mainly because of the scanned PDFs, which had to be converted to text one by one with a tool. When it was ready, I showed Joserra how to use it with his own Thursday question.
He typed: "returns period Hostelería del Turia contract". In four seconds: thirty days, clause six, contract signed 14 March 2023, and the link to the PDF. What had cost him a day and a half now cost the time it took to type the question.
But the good part came afterwards, with uses neither Joserra nor I had foreseen. The new sales rep, two months in and knowing nobody's terms, stopped interrupting the veterans: he asked the search. The purchasing person found in seconds what each supplier had promised them. And when someone left the company, their knowledge no longer left with them, because it was in the documents and the documents could now be queried.
The knowledge that stops walking out the door
There's something Joserra told me a couple of months later that sums up well what this is for.
—What surprised me most isn't how fast I find things. It's that I'm no longer scared of Andrés retiring.
Andrés had been at the company twenty-eight years and knew everything: what was agreed with whom, why they changed supplier in 2015, which client had a special deal and since when. All of that, until then, lived in Andrés's head and in documents only Andrés knew how to find. The day Andrés left, the company was going to lose twenty-eight years of memory.
We haven't put Andrés into a machine, nor do we need to. But we've made it so the documents Andrés knew how to find can be found by anyone, by asking the way you ask a person. The company's memory has stopped depending on Andrés being there. And that, for a company thirteen years old, is worth more than tidying a thousand folders.
Because companies almost never have a problem of not knowing. They have a problem of not being able to recall, in time, what they already know. And that, today, has a solution.
Search by meaning, not by file name
Your computer's search forces you to remember what the file is called. If someone else saved it, with a different logic, five years ago, you won't find it. That's how a company can have everything and find nothing.
An AI search over your own documents looks for what things mean, not what they're called. You ask in your own words and it returns the answer alongside the exact document it took it from, so you can check. There's no need to reorganise thirteen years of folders: you put something on top that knows how to read them.
Does your company know everything but find nothing?
If your team loses hours searching for contracts, terms or documents it knows exist, an AI search can sit on top of what you already have. We can look at it together.
Let's talk