Call · 30 min
·Services

AI agents for business, one per task. Next to each one, how it's measured.

In brief

Fifteen AI agents and two services: each one does a single task, on top of the tools the business already uses, and hands anything that commits the business to a person. Next to each one is its measurement, labelled with the kind of measurement it is: bench means a test system, then there's work done for a client. For the others, the measurement is taken on your own data in the first month.

This page starts from the problems: six costs a business pays without seeing them, and under each one the systems we've already built to deal with it. No business takes them all, and none should. You start where the loss is most obvious, often with just one: if that piece earns back more than it cost, we look at the next; if not, you stop there. Which one comes first is decided by the method, after we've looked at the business.

As pieces are added, they come together in one place where you can see everything: the Cruscotto.

Below is a summary of each piece. Each one has its own page explaining how it works step by step, what runs on its own and what waits for a person, and which number shows whether it's paying off.

  1. Hours spent doing repetitive work by hand
  2. Enquiries and contacts lost across channels and languages
  3. Missed bookings and customers who don't come back
  4. Money you're owed that arrives late, or not at all
  5. Customers who are out there but don't find you
  6. Deadlines and numbers nobody checks in time

AI agents, task by task.

The agentWhat it doesThe measurement
AI agent for replying to customer emails and messagesTime recovery

Reply to customer enquiries arriving by email, WhatsApp, website form and phone.

7.290 s from enquiry to draft, across 100 enquiries. Bench, 4 September 2026

AI agent for invoice processing: invoices, orders and delivery notes into ExcelTime recovery

Get the data from invoices, orders, delivery notes and incoming documents into Excel or your business software, without retyping it.

98.7% of fields correct across 30 documents. Bench, 4 September 2026

AI agent for translating and replying in the customer's languageOperations

Reply to customers in their own language, with emails, product sheets and price lists already translated.

Right language 97 times out of 100, in 5 languages. Bench, 4 September 2026

AI agent that keeps your CRM up to dateOperations

Keep the CRM up to date: customers, enquiries, orders and availability in one place.

4.262 s from WhatsApp message to queued contact. August 2026

AI agent for appointment remindersRevenue recovery

Send reminders for appointments and bookings, and free up cancelled slots in time.

Measured on your own data in the first month: no-shows

AI agent for winning back inactive customersRevenue recovery

Find the customers who have gone quiet and draft the message to bring them back.

Measured on your own data in the first month: orders from customers who had gone quiet

AI agent for invoice payment remindersCost recovery

Send invoice payment reminders and chasers, from the first to the last.

Measured on your own data in the first month: average days late on payments

AI agent for finding the right grantsCost recovery

Find the grants and incentives the business is genuinely eligible for.

Measured on your own data in the first month: eligible grants flagged in time

AI agent for requesting and replying to Google reviewsNew demand

Ask for a Google review after every purchase and draft a reply to every review.

Measured on your own data in the first month: new reviews and replies given

AI agent to find new clientsNew demand

Find new clients who match your criteria, and rule out the rest.

389 profiles, the right 10 in an hour. For a client, 11 September 2026

AI agent on WhatsApp and your website, answering from your documentsNew demand

Answer customers on WhatsApp and on your website with information from your documents.

Measured on your own data in the first month: correct answers to fixed questions

AI agent for B2B cold emailNew demand

Prepare B2B cold emails that land in the inbox and talk about the company receiving them.

From 428 names to 10 ready emails. Bench, 4 September 2026

AI agent for product sheets and sales copyNew demand

Write product sheets, web pages and sales materials that all give the same figures.

Measured on your own data in the first month: from draft to publication

Visibility in search engines and AI assistantsNew demand · service

Get found on Google and cited by AI assistants when someone describes the problem you solve.

52 pages indexed. On itria.io, August 2026

Integrated digital storefrontNew demand · service

Build the company website connected to internal systems, with enquiries arriving ready to act on.

Measured on your own data in the first month: useful enquiries from the website

AI agent for your business deadline trackerInternal processes

Check contracts, renewals, certificates and routine inspections, and send an alert when action is needed.

Measured on your own data in the first month: deadlines flagged in time

AI agent for the monthly report and management dashboardInternal processes

Prepare the monthly report and the management dashboard on a fixed date.

Measured on your own data in the first month: reports delivered on the agreed date

Each name opens its own page: what it takes in, what it does, who approves, where it writes and how it's measured, in six lines, with how it works step by step underneath.

Hours spent doing repetitive work by hand

Time recovery
Inbox AI

AI agent for replying to customer emails and messages

We measure: average time to first response.
How it works

Enquiries from WhatsApp, email or the website are gathered in one place and sorted in real time. Recurring questions get an answer the business has approved; everything else goes to a person, with the context already gathered.

Expected resultsReplying to a lead within 5 minutes makes you up to 100 times more likely to reach them than waiting 30 minutes. Source: InsideSales.com/MIT study, 2007

What that figure really says, and what it doesn't, is covered in the five minutes that decide an enquiry. Where automated enquiry handling is heading, and the caveats, are in Gartner's forecast on customer service.

Data Ingestion

AI agent for invoice processing: invoices, orders and delivery notes into Excel

We measure: documents processed with no manual input.
How it works

Emails in foreign languages, PDF orders, delivery notes and invoices are read and turned into structured data, with nothing retyped into the business software. The same mechanism covers bookkeeping prep (recording invoices and transactions before they go to the accountant) and any kind of recurring data entry that currently takes hours by hand.

Expected resultsItalian invoices have been arriving in structured form since 1 January 2019, and in 2025 some 2.4 billion of them passed through the national e-invoicing exchange (Sistema di Interscambio): the manual work that's left is orders, transport documents and whatever comes in from abroad. The time saving is measured on your own batch of documents, before any quote. Source: Agenzia delle Entrate, I risultati 2025

Enquiries and contacts lost across channels and languages

Operations
Multilingual Draft

AI agent for translating and replying in the customer's language

We measure: preparation time for a multilingual message.
How it works

Messages and documents going abroad are drafted in the recipient's language, and always reviewed by a person before they're sent.

Expected results76% of consumers prefer to buy products with information in their own language; 40% never buy from a website in another language. Source: CSA Research · Can’t Read, Won’t Buy, 8,709 consumers in 29 countries
Data Infrastructure

AI agent that keeps your CRM up to date

We measure: leads and follow-ups no longer lost.
How it works

No more scattered Excel files: one central database, synced in real time across sales, stock and leads, instead of spreadsheets nobody updates at the same pace.

Expected resultsEstimates of the return on CRM projects vary widely depending on how far the system is actually used: the same research firm measured 8.71 dollars for every dollar invested in one case analysis, and much lower figures in later surveys. It's a range, not a promise. Source: Nucleus Research

Missed bookings and customers who don't come back

Revenue recovery
Booking Recovery

AI agent for appointment reminders

We measure: no-shows prevented.
How it works

A reminder goes out before the appointment, and anyone who doesn't confirm is contacted again. A slot that would otherwise sit empty can be filled in time.

Expected resultsIn healthcare, where measurement is most rigorous, a systematic review of 21 studies covering more than 16,000 patients found that people who get a reminder are 25% less likely to miss their appointment (15% versus 21%). Source: BMJ Open, systematic review and meta-analysis · In restaurants, an industry study of 212,000 bookings measured a 32% drop with automatic reminders. The actual effect still has to be measured on site.
Customer Pipeline & Reactivation

AI agent for winning back inactive customers

We measure: time to follow up, and the share of leads followed up.
How it works

Every contact is sorted by priority and status, with a single history. An automatic sequence gets back in touch with dormant customers and tracks the replies and the orders won back.

Expected resultsA contact who hasn't said no stays open, and almost always without a next step assigned. This system shows who has stalled, for how long, and whose move it is, which is exactly where a deal fizzles out without anyone deciding to let it go.

Money you're owed that arrives late, or not at all

Cost recovery
Reminders

AI agent for invoice payment reminders

We measure: average days to get paid.
How it works

A three-stage sequence of reminders follows up overdue invoices, instead of relying on the mood of the person who was meant to do it that morning.

Expected resultsEuropean businesses lose an average of 73 working days a year chasing late payments, and about 11% of revenue is paid late. Source: Intrum, European Payment Report
Grants Matcher

AI agent for finding the right grants

We measure: opportunities flagged in time to act.
How it works

Grants and incentives relevant to the sector are spotted and flagged as they come up, instead of depending on one person's memory.

Expected resultsNo more deadlines missed because nobody happened to check that week.

Customers who are out there but don't find you

New demand
Reviews

AI agent for requesting and replying to Google reviews

We measure: average time to reply to a review.
How it works

After the visit, the system gathers reviews and drafts a reply, ready for the owner to approve before it's published.

Expected results97% of consumers read reviews of local businesses; 80% are more likely to choose a business that replies to all its reviews. Source: BrightLocal · Local Consumer Review Survey 2026, 1,002 US consumers
Lead generation

AI agent to find new clients

We measure: qualified leads per week, and the share rejected, with the reason.
How it works

The system looks for companies that match the buyer profile, checks the details against public sources, rejects those that don't fit and hands over the list with the reason for every pick. A person still does the sending.

Visibility

Visibility in search engines and AI assistants

We measure: indexed pages, impressions and average position in Search Console.
How it works

Content, structured data and guidance files written so that search engines and AI assistants understand what the business does, and cite it when someone asks.

Content

AI agent for product sheets and sales copy

We measure: time from draft to publication.
How it works

Spec sheets, web pages, emails and posts written in the buyer's language. AI writes the draft, a person reviews it, and the content says so.

Custom chatbot

AI agent on WhatsApp and your website, answering from your documents

We measure: answer accuracy against the actual knowledge base.
How it works

An assistant that answers from a knowledge base specific to the business (product sheets, policies, past FAQs), with or without a RAG system depending on how much knowledge there is to handle.

An assistant that talks to people is one of the systems that has to be disclosed as AI. What that actually means for the business using it is set out here.

Expected resultsTools built on retrieving answers from documents, and sold as hallucination-free, still get it wrong between 17% and 33% of the time: that's why every answer cites the document it came from, and outside the agreed scope the system says it doesn't know. Source: Stanford RegLab, preregistered evaluation
B2B Acquisition Engines

AI agent for B2B cold email

We measure: qualified meetings booked.
How it works

The infrastructure for cold outreach: domain setup, AI-assisted contact research and enrichment, and targeted sequences to the right companies.

Expected results82% of buyers accept meetings with sellers who reach out first, and it takes eight touches on average to get one. The buyers surveyed are mostly in the US, so treat it as an order of magnitude. Source: RAIN Group, 488 buyers and 489 sellers, 25 industries
Integrated Digital Storefront

Integrated digital storefront

We measure: website enquiries that actually qualify.
How it works

A website with editorial design, connected to the company's internal systems instead of sitting there as a static shop window cut off from everything else.

Expected resultsWhen more than 2,600 people explained why they did or didn't trust a website, its look was the most cited reason of all, mentioned in 46.1% of comments (Stanford Web Credibility Project); and 53% of mobile visits are abandoned if a page takes more than 3 seconds to load. Source: Google, The need for mobile speed

Deadlines and numbers nobody checks in time

Internal processes
Deadline Tracker

AI agent for your business deadline tracker

We measure: deadlines caught at least seven days in advance.
How it works

Every morning at 7:30 the system checks the coming seven days of deadlines (contracts, compliance filings, renewals) and flags only the ones that need action.

Expected resultsKnowledge workers spend a large part of their week looking for information and dealing with admin instead of high-value work: that's the time this system gives back.
Reports & Accountant

AI agent for the monthly report and management dashboard

We measure: the indicators we agree on together.
How it works

On the first of every month: a written narrative for the owner and a PDF report already in the format the accountant needs. The indicators are the ones set during the Analysis.

Expected resultsDuring the core hours of the day, people at work are interrupted by meetings, emails and messages on average every two minutes, 275 times a day: a report that arrives already digested saves one more interruption. Source: Microsoft Work Trend Index, 31,000 respondents in 31 countries

On their own they're automations. Together they're a system.

Each point is a piece we've already built, and under each name is the number it's judged by. Switch on the goals you care about and the lines show where they lead. Click a name to open its page.

Time recovery

Operations

Revenue recovery

Cost recovery

New demand

Internal processes

Questions about AI agents

How much does an AI agent cost?

According to the public price lists of five Italian providers, checked again on 25 September 2026, a custom agent costs 5,000 to 20,000 euros, and the monthly fee, where published, ranges from 500 to 5,000 euros a month. The price rises with the work: how many processes, how many systems to connect, how many languages and channels. The bands, compared with the cost of doing the task by hand, are on the page how much an AI agent costs an SME.

What's the difference between an AI agent and a chatbot?

A chatbot produces an answer and stops there. An agent reads from authorised sources, decides within rules written in advance, carries out an action in a real system, such as a row in the business software or a draft email, and stops when evidence is missing. That's why it has written permissions and a person approving the steps that commit the business: the difference, in full.

What are real examples of AI agents in an SME?

Three, each with its own kind of measurement. An agent that reads invoices, orders and delivery notes and writes the data into Excel: 98.7% of fields correct across 30 documents, measured on a test bench on 4 September 2026. An agent that files every WhatsApp message in the CRM: 4.262 seconds from message to queued contact, across 19 messages from August 2026. An agent that looks for new clients in public sources: 389 profiles narrowed to the right 10 in an hour, for a client, on 11 September 2026.

Which AI agent should you start with?

With the agent working on the task that costs most today, counted first: how many times a month it happens, how many minutes it takes, what an hour of the person doing it costs. It's often just one. If that piece returns more than it cost, you look at the next one; otherwise you stop there. You do the sum in six lines with what answering enquiries by hand costs, and the order is decided by the method, after looking at the business.

Does an AI agent decide on its own?

Within rules written in advance, yes; anything that commits the business, such as a price, a confirmation or a message to someone who doesn't know you, waits for a person's yes, and every sheet says so in the Approved by line. When it talks to a customer, the agent says it's a system, as required by Article 50 of the EU regulation on artificial intelligence, which applies from 2 August 2026: the information has to be given “at the latest at the time of the first interaction or exposure”.

·Where the pieces come together

The Cruscotto

Each piece on this page works on its own. The Cruscotto is where you see them together: built on how the business is actually organised, each morning it shows what needs a decision today and stays quiet about the rest.

Every Itria system is built the same way: one piece at a time, each one proven before the next. The method for deciding which piece comes first is the same for every business.

See the method

Want to see these pieces already assembled into a complete system? Real applications for hospitality and for food and agriculture.