[{"data":1,"prerenderedAt":196},["ShallowReactive",2],{"post-en-ai-documenti-aziendali":3},{"id":4,"title":5,"body":6,"date":182,"description":183,"extension":184,"image":185,"imageAlt":186,"imageNight":187,"meta":188,"navigation":189,"path":190,"published":189,"seo":191,"stem":192,"thumb":193,"thumbNight":194,"__hash__":195},"blog_en\u002Fblog\u002Fen\u002Fai-documenti-aziendali.md","AI on company documents: what really works",{"type":7,"value":8,"toc":173},"minimark",[9,18,23,30,33,37,65,68,72,104,108,111,132,136,139,162,165],[10,11,12,13,17],"p",{},"\"Can the AI answer using our documents?\" It is one of the questions we hear most often. The short answer is yes. The long answer is that ",[14,15,16],"strong",{},"it works very well for some questions and poorly for others",", and it pays to know which before you start.",[19,20,22],"h2",{"id":21},"how-it-works-without-jargon","How it works, without jargon",[10,24,25,26,29],{},"The assistant does not learn your documents by heart. When someone asks a question, the system searches the documents for the most relevant passages and gives them to the AI model together with the question. The model answers ",[14,27,28],{},"by reading those passages",", and it can show which document the answer comes from.",[10,31,32],{},"This technique is called RAG (Retrieval-Augmented Generation). The practical advantage is big: when you update a document, the assistant uses the new version straight away, with no training involved.",[19,34,36],{"id":35},"where-it-works-well","Where it works well",[38,39,40,47,53,59],"ul",{},[41,42,43,46],"li",{},[14,44,45],{},"Manuals and procedures."," \"How do I register a return?\", \"Which documents do I need for accreditation?\". The answers are written down somewhere, but nobody wants to look for them.",[41,48,49,52],{},[14,50,51],{},"Contracts and terms."," \"What is the notice period with this supplier?\". The assistant finds the clause and quotes it.",[41,54,55,58],{},[14,56,57],{},"Repetitive questions from customers or colleagues."," The same ten questions, every week. The assistant answers at once, and a person steps in only for new cases.",[41,60,61,64],{},[14,62,63],{},"Long documents."," Summarising an 80-page tender or comparing two versions of a set of rules.",[10,66,67],{},"In Media Accreditation, our product for press accreditation, the assistant answers journalists and staff from the event documents: times, rules, the press room map. The questions are always the same, and the answers are already written.",[19,69,71],{"id":70},"where-it-works-poorly","Where it works poorly",[38,73,74,86,92,98],{},[41,75,76,79,80,85],{},[14,77,78],{},"Sums and figures across lots of data."," \"How much did we invoice Lombardy customers in March?\" is not a question for documents: it is a question for your management system. That needs a direct connection to the data, for example through an ",[81,82,84],"a",{"href":83},"\u002Fen\u002Fblog\u002Fserver-mcp-gestionale","MCP server",".",[41,87,88,91],{},[14,89,90],{},"Old or contradictory documents."," If the folder holds three versions of the same price list, the assistant may pick the wrong one. Tidy up first, then connect.",[41,93,94,97],{},[14,95,96],{},"Poor-quality scans."," A crooked photo of a PDF is hard to read, for people and machines alike.",[41,99,100,103],{},[14,101,102],{},"Guaranteed zero errors."," A good assistant is rarely wrong and makes it clear when it is unsure, but it is not infallible. For important decisions, check the answer against the source it quotes.",[19,105,107],{"id":106},"privacy-first","Privacy first",[10,109,110],{},"Company documents often contain confidential data. Before starting, you need clear answers to three questions:",[112,113,114,120,126],"ol",{},[41,115,116,119],{},[14,117,118],{},"Where do the documents go?"," For Media Accreditation we use Claude models on AWS Bedrock in a European region: the data stays in Europe.",[41,121,122,125],{},[14,123,124],{},"Are they used to train the model?"," With professional services the answer must be no, written in the contract.",[41,127,128,131],{},[14,129,130],{},"Who sees what?"," The assistant must respect permissions: someone who cannot open a document must not receive its content through the AI either.",[19,133,135],{"id":134},"how-to-start-without-wasting-money","How to start without wasting money",[10,137,138],{},"You don't need to connect the whole company. Our method is simple:",[112,140,141,148,155],{},[41,142,143,144,147],{},"Choose ",[14,145,146],{},"one set of documents"," with a clear purpose, for example the support procedures.",[41,149,150,151,154],{},"Collect ",[14,152,153],{},"20 real questions",", the ones people actually ask.",[41,156,157,158,161],{},"We test the assistant on those questions and ",[14,159,160],{},"measure"," how many answers are right and complete.",[10,163,164],{},"Within a few weeks you know whether it works and how much time it saves, with your own numbers rather than a demo.",[10,166,167,168,172],{},"If you have a folder of documents people keep asking about, ",[81,169,171],{"href":170},"\u002Fen\u002Fcontact?topic=ai","tell us which one",": we'll tell you whether AI can help and where to start.",{"title":174,"searchDepth":175,"depth":175,"links":176},"",2,[177,178,179,180,181],{"id":21,"depth":175,"text":22},{"id":35,"depth":175,"text":36},{"id":70,"depth":175,"text":71},{"id":106,"depth":175,"text":107},{"id":134,"depth":175,"text":135},"2026-11-08","An assistant that answers from your manuals, contracts and procedures is a practical tool today. It works very well in some cases and poorly in others. Here is how to tell them apart.","md","\u002Fblog-ai-documenti-panoramica.png","In a round library, a small robot on a ladder takes a book and hands a person at the counter a glowing page, tied to the book by a thread of light","\u002Fblog-ai-documenti-panoramica-notte.png",{},true,"\u002Fblog\u002Fen\u002Fai-documenti-aziendali",{"title":5,"description":183},"blog\u002Fen\u002Fai-documenti-aziendali","\u002Fblog-ai-documenti.png","\u002Fblog-ai-documenti-notte.png","Vj3elF638C7Xly5mgnLSG0Z85r63_H9m_qGXzb-4Am0",1794240708131]