AI agents that answer, understand and get the work done.
A custom AI agent answers customers and colleagues on your website, WhatsApp, email and phone, looks up information in your systems and takes action: logs the lead, books the appointment, opens the case. When a case is delicate, it hands it to a person.
- Connected to your systems
- EU servers
- A person always in the loop
Requests arrive at all hours. Your team doesn't.
Messages, emails and calls pile up. Every late reply is a customer cooling off.
"I'll call you back tomorrow."
Requests come in the evening and at weekends. By the time someone replies, the customer has gone with whoever answered first.
"Always the same questions."
Opening hours, prices, availability, order status: the team spends hours repeating the same answers instead of handling the cases that matter.
"I'll put it in the CRM later."
After every conversation someone has to copy data, book appointments and update the tools. When the day is busy, nobody does.
What an AI agent is, in plain words
Not a chatbot with canned answers: an assistant that understands the request and gets the job done.
A traditional chatbot follows a fixed script: if the question falls outside the expected answers, it gets stuck. ChatGPT understands any request, but it doesn't know your company and can't do anything in your systems.
An AI agent combines the two. It understands written or spoken requests the way a person would, looks up information in your tools (CRM, management system, calendar, documents) and carries out the actions you've authorised: logging a lead, booking an appointment, opening a ticket, preparing a draft quote.
You decide what it can do on its own and what needs approval. When a request is delicate or outside its scope, the agent hands it to a person with a summary of the conversation ready.
The differences, side by side
| Capability | Traditional chatbot | Generic ChatGPT | Custom AI agent |
|---|---|---|---|
| Understands freely worded requests | No, follows a script | Yes | Yes |
| Knows products, prices and procedures | Only the answers entered | No | Yes, from your systems |
| Acts in your tools | No | No | Yes, with the permissions you set |
| Works on WhatsApp, email and phone | Usually website only | No | Yes |
| Hands over to a person when needed | Sometimes | No | Yes, with a summary |
| Logs every action | Rarely | No | Yes |
What an AI agent can do
The most common actions. Each agent only does the ones you decide.
Answer right away
Replies within seconds, at any hour, using your company's information: products, prices, opening hours, procedures.
Qualify leads
Asks the right questions to understand needs, budget and timing, and passes only genuinely interested leads to sales.
Book appointments
Checks real availability in the calendar, offers free slots and confirms the appointment to both sides.
Update your tools
Logs leads and requests in the CRM, opens tickets, updates the status of an order or a case in the management system.
Prepare draft documents
Prepares quotes, summaries or email replies for approval: a person checks and sends with one click.
Hand over to a person
When a case is delicate or outside its scope, it brings in the right person and passes on the conversation summary.
Where the agent works
On the channels your customers and team already use. The same agent can cover several channels.
Website
A chat on your website that answers, qualifies and books while the visitor is still on the page.
WhatsApp Business
Replies to WhatsApp messages from the company's official account, including attached photos and documents.
Reads incoming emails, understands what they ask, replies or routes them to the right person with a draft answer.
Telegram, Teams and Slack
For customers on Telegram or for your internal team on Teams and Slack: questions on procedures, data and project status.
Phone
A voice agent answers calls, understands the request, books appointments or transfers the call to a colleague.
On every channel the agent always states that it is an AI assistant, as required by European rules on artificial intelligence.
On the phone too
- The customer calls the company number.
- The agent answers, introduces itself as an AI assistant and understands the request.
- It checks availability and books the appointment, or answers the question.
- If a person is needed, it transfers the call with the summary ready.
Where it works best
Typical situations where an AI agent saves time from day one, each with an example request.
Lead qualification
Today
Website requests land in a shared inbox and get read the next day, after the customer has contacted a competitor.
Example request
“How much would solar panels cost for an 800 m² warehouse?”
What changes
Instant reply, qualified lead in the CRM and a call booked with the right salesperson.
Customer support
Today
Most requests are about the same topics, and the team spends the day repeating the same answers.
Example request
“My order 4587 hasn't arrived yet, where is it?”
What changes
A reply with the real order status; complex cases go to a person with a summary.
Bookings and appointments
Today
Bookings come in by phone and message, and someone has to juggle calendars and availability by hand.
Example request
“Do you have a slot for a visit on Tuesday afternoon?”
What changes
Availability checked in the real calendar and the appointment confirmed, even out of hours.
Internal team support
Today
Colleagues keep asking the same experts where to find a figure, a procedure or the status of a project.
Example request
“Where are we with the delivery for the Rossi account?”
What changes
An answer from company systems, without interrupting anyone.
Quotes and technical requests
Today
Every quote request means collecting details the customer didn't send, with endless back-and-forth by email.
Example request
“I need a quote for 200 metres of fencing.”
What changes
The agent collects every detail needed and prepares a draft quote for approval.
E-commerce and after-sales
Today
Questions about sizes, availability and returns arrive at all hours, and after the purchase contact with the customer stops.
Example request
“Does this jacket run large? I'm usually a medium.”
What changes
Product advice during the purchase and personalised after-sales messages.
It works best where requests are frequent, similar to each other and require looking up or updating your systems. If that's not your case, we'll tell you during the Discovery.
From message to action, in 5 steps
What happens every time a request comes in.
The request arrives
From the website, WhatsApp, email or phone. If the sender is already in the CRM, the agent pulls up their details so it doesn't ask for the same things twice.
It understands what's needed
It interprets the request even when it's badly worded or incomplete, and asks clarifying questions instead of guessing.
It looks up the information
It checks your systems: catalogue, price lists, calendar, CRM, order status, documents. It only uses real, up-to-date data.
It acts
It replies and carries out the permitted actions: logs the lead, books the appointment, opens the ticket. Important actions wait for a person's approval.
It logs and hands over
Every conversation and action is logged. If the case falls outside its scope, the agent brings in the right person with the summary ready.
The AI model receives only the information needed for that single request, never the whole company archive.
You decide what it can do
An AI agent works for your company: you write the rules, and every action can be checked.
A defined scope
You set the topics, channels and permitted actions. Outside that scope the agent doesn't answer and hands over to a person.
Approval for important actions
Quotes, refunds, data changes: the agent prepares, a person approves with one click.
A complete log
Every conversation and action is logged, with date, channel and data used. You decide how long it's kept.
Servers in the European Union
Systems run on OVHcloud servers in European data centres, or in your private cloud or on your own servers.
The right model for each use
We choose the AI model based on language, complexity, confidentiality and budget. It can be swapped without rebuilding the agent.
Transparency with customers
The agent always states that it is an AI assistant. Customer data is not used to train models.
How we measure quality
Before it talks to your customers, the agent is tested on real conversations.
Test conversations
With your team we write 30-50 typical conversations, including the difficult ones, with the expected behaviour for each.
Three checks on every answer
Is the answer correct? Is the action taken the right one? When it shouldn't act, did it hand the request to a person?
A test on every change
New information, a different model, new rules: before publishing we rerun every test conversation.
Feedback from your team
Conversations handed to a person and answers corrected by hand become new test conversations and improvements to the agent.
You see the test results, and you decide when the agent is ready to talk to customers.
Agent Lead or a custom agent?
We have two routes. We choose together based on what you really need.
| Agent Lead | Custom agent | |
|---|---|---|
| What it does | Qualifies website visitors and hands you ready leads | Any task: support, bookings, quotes, internal help |
| Channels | Website | Website, WhatsApp, email, Telegram, Teams, Slack, phone |
| Connection to your systems | Standard integrations | CRM, management system, calendar and custom tools |
| Activation | In a few minutes, on your own | A phased project, starting with a free Discovery |
| Price | Monthly subscription, with a free trial | Fixed price per phase, set after the Discovery |
Start small. You decide whether to continue.
Four separate phases: at each step you have something concrete in hand and decide whether to go on.
1. Discovery
1-2 weeks · freeAnalysis of the requests you receive, the channels and the systems to connect. Together we decide what the agent can do on its own and what it can't.
What you get
Project plan, cost and time estimate, test conversations
2. Proof of concept
2-3 weeksA working agent on one channel and one use case, connected to your real data and tried by your team.
What you get
Working demo and a first quality measurement
3. Development
3-6 weeksAll planned channels and actions, approvals, logging, hand-over to a person, team training.
What you get
Live agent, documentation and training
4. Operations
ongoingConversation monitoring, information updates, improvements based on real cases.
What you get
Monthly report on conversations, quality and costs
Each phase is ordered on its own: after the proof of concept you decide whether to continue.
How pricing works
The cost of an agent depends on channels, systems to connect and conversation volumes. That's why we work like this.
The Discovery is free
We analyse requests, channels and systems and tell you whether an agent makes sense for you. No commitment.
Fixed price for each phase
You get the price of each phase before we start. It doesn't change along the way.
You decide phase by phase
We start with one use case on one channel. After each phase you decide whether to continue.
Transparent running costs
Hosting, maintenance and the phone or WhatsApp number come with a clear fee. Pay-per-use AI costs are passed through at cost.
When you don't need an AI agent
We'd rather tell you upfront than sell you a project you don't need.
Only a few requests a day
If you get few messages and the team replies in time, an agent costs more than it saves.
Decisions that need judgement
Diagnoses, consulting, delicate negotiations: the agent can collect the details, but the decision stays with a person.
Information that doesn't exist
If prices, procedures and availability aren't written down anywhere, the agent has nowhere to get them from. They need sorting out first.
You only need to qualify website visitors
In that case Agent Lead already does what's needed, activates in minutes and costs less than a custom project.
Frequently asked questions
Looking for something different?
How many requests go unanswered today?
In the free Discovery we analyse the requests you receive and tell you whether an AI agent makes sense for you. If it doesn't, we'll tell you.