{"id":3883,"date":"2025-11-13T16:23:02","date_gmt":"2025-11-13T14:23:02","guid":{"rendered":"https:\/\/www.puurdata.nl\/?p=3883"},"modified":"2025-11-17T10:20:44","modified_gmt":"2025-11-17T08:20:44","slug":"this-way-youll-find-what-youre-looking-for-in-unstructured-data","status":"publish","type":"post","link":"https:\/\/puurdata.nl\/uk\/zo-vind-je-wel-wat-je-zoekt-in-ongestructureerde-data\/","title":{"rendered":"Smart use of AI \u2013 This is how you actually find what you're looking for in unstructured data"},"content":{"rendered":"
<\/div><\/div><\/div>
\n\t
\n\t\t
\n\t\t\t
\n\t\t\t\t
<\/div> <\/div>
\n\t
\n\t\t
\n\t\t
\n\t\t\t

Smart with AI \u2013 How to actually find what you\u2019re looking for in unstructured data<\/h1>\n
\n\t
\n\t\t\t<\/div>\n<\/div>\n\n\n\n\n\t\t<\/div> \n\t<\/div>\n\t<\/div> \n\n\t
\n\t\t
\n\t\t
\n\t\t\t\n\t\t<\/div> \n\t<\/div>\n\t<\/div> \n<\/div><\/div>\n\t\t\t<\/div> \n\t\t<\/div>\n\t<\/div> \n<\/div><\/div>\n\t\t
<\/div><\/div><\/div>
\n\t
\n\t\t
\n\t\t\t
\n\t\t\t\t
<\/div> <\/div>
\n\t
\n\t\t
\n\t\t
\n\t\t\t\n
\n\t
\n\t\t

by Francesca Brzoskowski<\/em><\/p>\n

Imagine: a colleague in operations wants to know how your company resolved a customs delay in Germany last year. They know someone dealt with it and that emails were sent around. But as soon as they type \u2018Germany customs 2024\u2019 into the shared folder, 47 emails and 30 PDF reports appear, most of which have nothing to do with their question.<\/strong><\/p>\n

She asks three colleagues for help. One sends a report, another adds someone from logistics. An hour later, she has eleven browser tabs open, but still no clear answer.<\/p>\n

This happens in almost every organisation. Not because people work messily, but because the resources to search effectively are missing.<\/p>\n

Traditional search only works to a limited extent<\/h3>\n

Many organisations still rely on traditional search methods. These primarily look for exact word combinations: in emails, files, folders, or subfolders.<\/p>\n

If you type \u2018onboarding problems\u2019, you will only get documents that contain those exact words. One typo or a missing word, and important results will disappear from view.<\/p>\n

Slimmer searching with semantic and vector search<\/h3>\n

New search techniques work differently. Whether you use a modern search engine or an MCP server linked to your apps: these systems understand the meaning behind words.<\/p>\n

Text is converted into vectors, or numbers that represent the meaning of words. This allows a search engine to understand that a question about onboarding issues can also refer to documents about customer implementation, setup challenges, or delays with new accounts.<\/p>\n

An MCP server goes a step further. It searches all your data simultaneously \u2013 emails, documents, PDFs, shared drives \u2013 and automatically determines where and how to search.<\/p>\n

\"\"<\/p>\n

A practical example: from hours to seconds<\/h3>\n

A production company had fifteen years of maintenance reports, all saved as PDFs. Technicians knew that certain machines had recurring problems, but finding similar incidents normally meant: sifting through hundreds of reports.<\/p>\n

Following the introduction of semantic search with Elasticsearch, a technician could simply ask: \u201cWhat usually causes overheating in hydraulic systems?\u201d<\/p>\n

Within six seconds, he received six relevant incidents back, spread across different years, locations, and formats, but thematically related. Manual searching would take hours, if anyone would even start.<\/p>\n

When do you opt for which approach?<\/h3>\n

Which approach works best varies by organisation. These are the main considerations.<\/p>\n

Semantic search engine (e.g. Elasticsearch)<\/strong>
\nIf you\u2019re building your own semantic search engine, the data must first be prepared: splitting documents, embedding them, vectorising them, designing indexes and testing them with real search queries. It takes work, but it delivers predictable and easily verifiable search results. Ideal if you want concise, accurate answers.<\/p>\n

MCP servers<\/strong>
\nMCP servers allow you to skip that preparation. They work directly with what you already have: emails, tools, documents, shared folders. In return, you must ensure good authentication, permissions, and monitoring. MCP servers are quicker to implement, but harder to control without clear operational agreements.<\/p>\n

The way forward for organisations<\/h3>\n

Technology is no longer the limiting factor. Embedding models are accessible, and MCP servers are rapidly maturing. What counts most is where your organisation stands:<\/p>\n