If X happens, do Y
Fixed rules, perfect for the steps that never change: a sync, a reminder, an invoice that goes out by itself.
A payment arrives → the back office updates.
AI agents and automation
Where there’s data, an agent works.
Custom AI agents: they read your data, act in your tools and hand over to a person when needed. To sell, support, make sense of the numbers and keep the CRM in order.
Four families of agents
What an AI agent is
It doesn’t just answer. It works.
An AI agent is software that reads your company’s data, understands what is going on, decides the next step and carries it out in your tools. When a case isn’t its to handle, it calls a person.
Answers questions
It converses, but it usually stops there: it can’t see your CRM and doesn’t act on your systems. An answer, not a job done.
Opening hours, prices, ready-made replies.
Has a goal and does the work
It reads the context, picks the steps, uses the tools — CRM, email, WhatsApp, databases — and hands over to a person when needed.
Qualifies a lead, logs it in the CRM, books the appointment and alerts the salesperson.
They aren’t alternatives. Automations move the data; agents make the decisions that used to wait for a person. Together, a process runs by itself.
Where there’s data
Every piece of data is work waiting to happen.
Every company sits on data nobody has time to read: leads, conversations, contracts, orders, spreadsheets, documents. An agent reads all of it, every day, and does what today waits for someone to find five minutes.
Your data
- CRM
- ERP · Back office
- SQL database
- Website and forms
- Spreadsheets and documents
- E-commerce
- Calendars
The agent
- Reads
- Understands
- Decides
- Acts
- Hands over
The work, done
- Lead qualified, appointment booked
- Request solved, or ticket already prepared
- Reports and alerts that arrive on their own
- Customer record updated and clean
- Draft reply ready to review
And where the data is something else — administration, contracts, stock, HR — the principle is the same: if a process reads and writes data, there is almost always an agent that can help.
What we build
Four agents. One goal: less work that doesn’t need doing.
Each can live on its own or work with the others, on the same data and under the same rules.
Sales
Sales agents
No lead goes unanswered. No quote goes without follow-up.
- Reply at once to anyone who writes from the website, WhatsApp or a form, at any hour
- Qualify the lead with the right questions and log it in the CRM, with no duplicates
- Offer the trial, the visit or the appointment and put it in the calendar
- Follow up on quotes and stalled requests, and reawaken dormant leads
- Hand the salesperson the hot lead, with the context already summarised
Connects to
- Website
- CRM
- Calendar
An exampleA lead writes at 11:10 pm. The agent replies, understands what they want, offers two time slots and creates the record in the CRM. The next morning the salesperson finds an appointment, not a message to call back.
Support
Customer support agents
The right answers, where people already write.
- Answer from your company’s own information, not from what “sounds” right: if the answer isn’t there, they say so
- Handle the requests that keep coming back: opening hours, bookings, cancellations, documents, the status of a request
- Work on the website, in the app and on WhatsApp, in several languages
- Open the ticket already filled in and pass it to the right person with the conversation attached
- Recognise a complaint or a delicate case and hand over to a person straight away
Connects to
- Website
- App
- Tickets
An exampleA customer asks how to cancel. The agent checks the rule, explains it in two lines and says where to do it. If the case is personal, it opens the ticket with all the context and alerts the team.
Data analysis
Data analysis agents
Ask the question. The number comes with its explanation.
- Query your database in plain language: “how many leads became customers this month, by channel?”
- Prepare, every week, the report someone builds by hand today, and send it to whoever needs it
- Watch the data continuously and warn when something drifts: a drop in conversions, a missed payment, a duplicate
- Reconcile and reclassify data that comes from different sources
- Explain what they see and which numbers it rests on, so nobody has to take it on trust
Connects to
- SQL database
- ERP
- CRM
- Spreadsheets
- Dashboards
An exampleEvery Monday at 8 a message arrives: where the funnel stalled, which channels paid off, what changed since last week. With the real data and a link to dig deeper.
CRM
CRM support agents
A CRM that fills itself in and tells you what to do next.
- Update records and pipeline from emails, calls and messages: less typing by hand
- Clean up the database: duplicates, missing fields, records to merge
- Summarise a customer’s history before a call or a meeting
- Prepare the draft reply to an email: a person reads it, and sends it
- Suggest the next action and remind you of deadlines, renewals and promises made
Connects to
- CRM
- Calendar
- Tickets
An exampleAn email comes in. The agent recognises the sender, finds their history, writes a draft in the thread and leaves it there. No reply goes out without a person having seen it.
Don’t see your case? Wherever a process reads and writes data, an agent can often help. Tell us the process: we’ll tell you whether it’s worth it.
Let’s talk→In production
Already at work, in a real club.
We built them inside Athlon Club, in Rome. Each agent started from a problem lived at the front desk, and today works every day beside the team.
Read the Athlon case→- An AI assistant on the website and app answers with the club’s information, points to the right form, and every request lands in the CRM already de-duplicated.
- An agent on WhatsApp answers, offers the trial and hands the conversation to the desk when a person is needed, in the team’s shared inbox.
- Incoming emails find a draft reply already in the thread: a person reads it and sends it.
- The rules can’t be bypassed: the agents don’t make things up, and in front of a complaint they call a person.
How we work
One process at a time. The result first, the technology second.
- 01
We pick one process
We start from a single flow: the one that costs the most time or loses the most leads. We measure how it is today, so we can say whether it works.
- 02
We connect the data
APIs, databases, webhooks: the agent reads from your systems and writes only where we told it to. The back office you have gets connected, not replaced.
- 03
We train it on your context
Instructions, company information, rules and tone of voice. We test it on real cases, the hard ones too, before putting it in front of customers.
- 04
We supervise it
A log of every conversation and every action, thresholds beyond which a person decides, and improvements on the cases it couldn’t solve.
Control and safety
A useful agent is an agent with limits.
It doesn’t make things up
It answers from the sources we point it to. If the information isn’t there, it says so and hands over to a person.
A person in the loop
Risky actions — a refund, a contract, an important message — wait for the approval of whoever is responsible for them.
Minimum permissions
Read-only where that’s enough. Each agent can do only what it needs, on only the data it needs.
Everything in the log
What it read, what it decided, what it did. It can be reviewed, corrected and explained.
Personal data is handled under the GDPR: consent split by purpose, only the data that’s needed, and the model and hosting chosen with you according to how sensitive the data is.
Frequently asked questions
What people ask us about AI agents.
What is an AI agent, and how is it different from a chatbot?
A chatbot answers questions. An AI agent has a goal: it reads your company’s data, decides the next step and carries it out in your tools — logs a lead in the CRM, books an appointment, prepares a report — and hands over to a person when needed.
How is it different from an automation?
An automation follows fixed rules: if X happens, do Y. It works very well when the process doesn’t change. An agent handles what rules don’t cover: free text, requests that are different every time, decisions that need context. We usually use them together: the automation moves the data, the agent decides.
Which processes is an AI agent worth it for?
Where there is a lot of data and repetitive work that still needs a bit of judgement: answering and qualifying leads, preparing support requests, updating the CRM, building reports, checking the data. Where the process is rare or high-risk, we start with the agent as a person’s assistant, not in their place.
Can an AI agent make mistakes? How do you keep it under control?
It can, like a person. That’s why we design it with limits: it answers only from the sources we point it to, it says so when it doesn’t know, risky actions wait for a person’s approval and everything goes into a log that can be reviewed. We test it on real cases before going live and improve it on the cases it doesn’t solve.
Does it work with my CRM or back-office software?
Almost always. We connect through APIs, webhooks or straight to the database — even a SQL Server on Windows — and we don’t ask you to replace what you use. If a system has no interfaces, we find the cleanest way to read its data.
Is my data safe? How does it work with the GDPR?
Personal data is handled under the GDPR: consent split by purpose, only the data that’s needed, minimum permissions for each agent and a log of what it does. We choose the model and hosting with you according to how sensitive the data is; health data is used only with explicit consent.
Do AI agents replace people?
No. They take away the repetitive work and leave people the part where they matter: the relationship, the exceptions, the decisions. When a situation is delicate, the agent hands over.
Which AI models do you use?
We pick the model by task, cost and how sensitive the data is. In our own projects we use, for example, Anthropic’s Claude models.
Where do we start, and what does it cost?
With a single process. We listen to how you work, look at the data and tell you, straight, whether an agent is worth it and what result to expect. The cost depends on the process and the systems to connect: we define it after the audit, before writing any code.