The software company that created FT Mercati Go to ftmercati.com

Service

AI assistants that answer on company know-how

An AI assistant that knows your company's procedures, documents and history and reads them alongside market data: answers with sources, role-based access.

Updated on

In brief

An assistant on company know-how is a bot that answers using the company's documents, procedures and data, not the generic knowledge of the internet. FT Support builds it by connecting the internal sources and, where needed, FT Mercati's market data: decision makers find in the same conversation what the company knows and what the market is doing.

  • It answers from the company's sources and always shows where the information comes from.
  • It respects permissions: everyone sees only what they have access to.
  • It can also read FT Mercati prices, indexes and fundamental data.
  • The sources are updated without redoing the project.

The problem it solves

In every company the knowledge that matters is scattered: in procedures, in contracts, in the quotes of the last ten years, in the ERP (the company’s management software), in the heads of three people. Whoever has to decide pieces it together by hand every time, phoning around and searching through folders.

An assistant on company know-how puts that knowledge a question away: “What terms have we granted this customer over the last three years?”, “What does the procedure say about an out-of-warranty return?”, “How did the margin on brass orders do when copper went up?”.

How it works, without jargon

  1. The sources are chosen. Documents, manuals, procedures, meeting minutes, but also ERP tables and reports. You start with a few reliable sources, not with everything.
  2. They are prepared. The documents are split up, indexed and linked to the existing permissions. Each fragment remembers which document it comes from.
  3. Answers are based on the sources. For each question the system retrieves the relevant passages and the model builds its answer from those, pointing to them. If the sources are not enough, it says so.
  4. You measure and improve. The questions left unanswered are the list of what is missing from the documentation: this is often the most useful result of the first months.

This technique is called RAG (retrieval-augmented generation). The practical advantage is that the know-how stays in your archives: it does not end up inside a model, it is updated when you update the documents, and access to it can be revoked.

Combining it with market data

Important decisions are made by bringing two things together: what the company knows about itself and what is happening outside. This is where the link with FT Mercati comes in: every day it distributes LME prices, Milan Indexes, scrap prices, exchange rates and millions of fundamental data series.

An assistant built by FT Support can read both sources. Asked “which expiring contracts have a price out of line with today’s copper?”, it can put on the table the company’s contracts and price lists, the reference prices and the indexation clauses. The decision stays with whoever has to take it, but it is taken with the whole picture in view.

This is the idea behind FT CEO, our assistant for senior management.

Who it helps most

  • Senior management: getting documented answers without waiting for someone to prepare a report.
  • Sales and purchasing: the history of terms, prices and negotiations, a question away.
  • Support and quality: procedures, manuals and cases already solved, searchable in natural language.
  • New staff: learning how the company works by asking, instead of interrupting colleagues.

What you need to get started

A limited set of reliable sources, a person who can say which ones are up to date, and a group of pilot users willing to ask real questions for a few weeks. The path is that of our method.

Frequently asked questions

Know-how assistants: the most common questions

What does it mean to train a bot on company know-how?

In most cases the model is not retrained: it is given controlled access to the company's documents, procedures and data. For each question the system retrieves the relevant passages and the model answers by citing them. It is faster, cheaper and safer than actual training, and the sources are updated without redoing anything.

Do confidential documents stay confidential?

Yes. The assistant respects permissions: each user gets answers based only on the sources they have access to. The models are used through business APIs, under contract terms that exclude the use of data for training, and where needed the archive stays on the company's servers.

Can the assistant make up answers?

The risk is greatly reduced by having it answer only on the basis of the sources retrieved, always showing where the information comes from, and having it say that it does not know when the sources are not enough. Decisions stay with people.

Can market data be connected to internal information?

Yes, and this is the point where FT Support and FT Mercati meet: prices, indexes, exchange rates and fundamental data can enter the same conversation alongside the company's data, for example to check a price list or an indexation clause against the reference prices.

Do you have a process you would like to see working better?

Tell us about it in a first conversation, with no obligation: together we work out whether and where AI brings a measurable benefit, and which step to start with.