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

Case study

A custom CRM with AI agents inside

The CRM we use every day at FT Mercati: an assistant that carries out operations, plain-language filters, AI checks on quotes and contracts.

In production Complete sales cycle: customers, quotes, contracts, invoicing Updated on

  • 9tools available to the agent, 5 of them read-only
  • 59screens, from customer records to management dashboards and trade fairs
  • 5languages covered by the AI translation of news, events and the service catalogue
  • 2roles with different permissions, which apply to the AI as well

In brief

This is the CRM FT Mercati uses every day to manage customers, quotes, contracts and invoicing. We custom-built it and put AI where it saves time: an assistant that knows the CRM and can carry out operations, filters written in plain language, intelligent checks on the data, automatic reading of orders in PDF. Every action the AI takes goes through rules, roles and confirmations.

Screenshots

The customer filter, written as a sentence in Italian and turned into a condition on the customer database
Filters in plain language: the sentence typed in the bar, here in Italian, becomes a real condition on the customer database, with the result in the grid straight away.
The AI assistant in the side panel has created a draft quote for a sample customer
The assistant that acts: from a request written in plain language, here in Italian, it creates the draft quote for the customer, summarises the data it used and says what is left to do by hand.
The draft quote created by the assistant appears in the customer's list of quotes
The draft created by the assistant appears among the customer's quotes: from here the operator adds the services from the catalogue and confirms.
The New Customer form with the data proposed from the VAT number
Customer records that fill themselves in: from the VAT number the CRM queries VIES and proposes the fields, which the operator checks before saving.

Screenshots from a demonstration environment with invented data: the companies, people and amounts are not real.

The starting point

A company that sells information services by subscription has a sales cycle all of its own: free trials, quotes built from a catalogue with different levels of exchange licence, contracts with renewals and cancellations, instalments, portal user accounts to activate and deadlines to keep an eye on. No standard CRM covered it without workarounds and without Excel sheets kept alongside.

So we wrote our own. And while we were at it, we designed it so that artificial intelligence (AI) would work inside it, not next to it.

What we built

A complete CRM: customer records with tax checks, quotes with multilingual PDFs, contracts generated from accepted quotes, invoicing with instalments and payments received, portal user accounts with their services and expiry dates, a sales planner, and dashboards on revenue, contracts, and the sales team’s targets and bonuses.

On top of this, an AI assistant that is always available in a side panel and knows which screen you are on and which role you have.

Where the AI works

  • An assistant that explains or acts. Asked “how do I record a renewal?”, it explains the procedure and takes you to the right screen. Asked to “create a quote for this customer”, it gathers the data, summarises it and, after confirmation, prepares the quote as a draft.
  • Filters in plain language. A sentence such as “Customers in Lombardy with a contract expiring within two months”, written in Italian in our case, becomes a real filter on the database. The filter can be saved, reused and shared. Every generated filter goes through an allowlist validator: only the documented tables and, for each role, only the ones that role is entitled to.
  • Intelligent checks. Before a customer, a quote or a contract is saved, an AI check flags the inconsistencies a fixed rule would miss, separating errors from warnings field by field.
  • Customer records that fill themselves in. From the VAT number, the CRM queries the European VIES service, searches the web for the company’s details and proposes the filled-in fields, which the operator checks before applying them. When an EU company is saved with incomplete VAT data, the server checks the VAT number with VIES, within eight seconds, and fills in the VAT regime; this also applies to customers created by the AI assistant.
  • Orders in PDF. The number and date of the purchase order and the CIG code (Italy’s identification code for public contracts) are extracted from the customer’s document; on confirmation, the CRM writes the notes that the e-invoicing program reads back.
  • Customer analysis for sales. A reasoned summary of how the customer uses the service, useful before a renewal or a phone call.
  • Renewals to watch. The salespeople’s notes are read to bring out the contracts at risk: announced cancellations, reminders, complaints.
  • Translations. News, events and the service catalogue go from Italian into English, French, Spanish and German.
  • Trade fairs. A complete module: from any exhibitor list published online, the scout agent collects the exhibitors, screens them for relevance and matches them to the customer records; then priorities, visits from a tablet with an offline queue, summaries of the notes. It is a case study in its own right, also available as a separate tool.

How we keep the AI under control

  1. A limited number of tools. The agent has nine tools: five read, three write, and one is deliberately blocked and sends the operator to the guided procedure.
  2. Confirmation before writing. The data gathered is repeated back to the user, who has to approve it.
  3. Everything starts as a draft. A quote created by the AI is a draft; a contract is awaiting confirmation.
  4. Hard rules in code. Duplicate VAT numbers, VAT groups, discount limits by role: these are decided by written rules, not by the model. Split payment (in Italy, public bodies and certain companies pay the VAT on an invoice directly to the State) is a rule too: every day the server aligns the VAT chargeability setting with the public lists of Italy’s Ministry of Economy and Finance and of the Indice delle pubbliche amministrazioni (Italy’s index of public bodies), with a cap on changes per run, a log and manual upload of the lists; an AI agent will soon be connected to these checks.
  5. Verified filters. The queries generated by the AI are checked before they run, and can only read.
  6. Log. Conversations, tools used and checks all remain on record.
  7. Contracts, not people. The salespeople’s notes are used to bring out the contracts at risk, not to assess the people who write them.

We use different models for different tasks: a faster one for conversation, checks and translations, a more powerful one for analyses and complex filters.

What it means for your company

This CRM is a real case, in production, used every day by the people who sell and the people who handle administration. It shows three things that hold true for any ERP, the company’s management software:

  • AI pays off when it knows the system it works in, screen by screen;
  • trust is built with clear limits, not with promises;
  • the results show up on repetitive tasks: searching, filling in, checking, translating.

If you have an ERP or a CRM and would like a digital colleague working alongside it, let’s talk.

Frequently asked questions

CRM with AI agents: questions and answers

What can an AI agent do inside a CRM?

It can find and summarise information on customers, quotes and contracts, build filters from a sentence in plain language, check that the data is consistent before it is saved, read documents such as purchase orders in PDF and prepare new records as drafts, which a person reviews and confirms.

Can the AI agent change data without supervision?

No. Most of the tools are read-only. Those that write ask for explicit confirmation, create records as drafts and respect the user's role. The most sensitive operations, such as creating login credentials, stay with people: the agent takes the user to the right screen and stops there.

Is this CRM for sale?

It was created for FT Mercati's sales cycle and is not an off-the-shelf product. It is, however, the foundation, in technology and in experience, that we start from when we build a custom CRM or ERP for another company.

Can an AI agent be added to a CRM that already exists?

Often, yes: if the CRM exposes APIs or an accessible database, the agent connects to those. The first thing to do is to choose one or two repetitive tasks on which to measure the result.

Other case studies

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.