Frequently asked questions about artificial intelligence

Athics is an Italian artificial intelligence company based in Reggio Emilia, at via Meuccio Ruini 10. We design AI agents, chatbots and voicebots for businesses, build tailor-made solutions and connect them to the systems already in place. Two products are ours as well: Crafter.ai, the platform for building AI agents, and Portrait, our psychometric profiling technology.

It starts with a discovery call to work out which activity weighs most on the team. Then comes the review of available content and data, the build of the agent on one narrow use case, testing with real people and the release into production. New use cases follow, once the first results are on the table.

Yes, and the method is the same: what changes is the size of the scope. A large organisation usually starts with a pilot department and expands from there, a small company starts with a single well-defined use case. Either way the first release covers one process, not the whole of customer service.

There is no one-size price list, because the figure depends on how many use cases the agent covers, how many integrations are needed and what shape your existing content is in. After the first call you get a quote with the scope written down: what is included, what is left out and what happens after go-live.

An AI agent is software that completes a task by holding a conversation. It understands the written or spoken request, checks the data you gave it access to, decides what to do and acts: it opens a ticket, books an appointment, updates a record in the CRM. At the end it tells the person what it did and what is still open.

Ordinary automation runs a fixed sequence: if A happens, do B. An AI agent gets a goal and picks the steps itself, so it also copes with requests nobody mapped out in advance. That is why it pays off where questions arrive in a thousand different wordings.

It says so. A well configured agent admits the limit, offers the closest alternative and hands the request to a person together with the full conversation history, so whoever picks it up does not start from scratch. Unanswered questions go into a report, and that report is next month's work list.

By the numbers of the process it works on. We look at the share of requests closed without a human, first response time, how many conversations are handed to an operator and how users rate the exchange. The comparison has to be against the months before launch, otherwise it stays an impression.

The chatbot goes where your audience already is: website, WhatsApp, Messenger, Telegram, Microsoft Teams, Slack, email. The knowledge base stays single and serves every channel, so one correction to the information shows up in every conversation instead of being done three times over.

Through RAG, short for retrieval augmented generation. Your manuals, price lists, pages and procedures are indexed; before answering, the chatbot retrieves the relevant passages and writes the answer from those. The practical gain is twofold: the answer stays anchored to your documents, and it changes when the documents change.

The chatbot detects the language of the person writing and replies in the same one, and the main European languages need no extra work. A knowledge base written in Italian can serve requests in English or German too. Where exact terminology matters, as in technical documentation, it is worth translating the key content.

Rule-based bots follow a hand-written answer tree and understand nothing outside it. An AI chatbot reads a freely worded question, copes with typos and half sentences and keeps track across several messages. For the user the difference is one thing: they write the way they speak, instead of hunting for the right button.

Like a conversation, not like a touch-tone menu. The voicebot speaks with a natural synthetic voice, listens while the caller talks and copes with interruptions, so nobody has to wait for the message to end before speaking. At the start of the call it states that it is an automated system, as European rules require.

Yes, it connects to the number you already use, through your telecoms provider or your cloud PBX, with no change of number and no rewiring of the internal network. You decide how much work it takes on: every call, only the out-of-hours ones, or only those queuing beyond a set waiting time.

Yes. On outbound it handles appointment reminders, delivery confirmations, renewals and follow-ups on contacts left hanging, and it writes the outcome of each call into your business system. One caveat: promotional calls to consumers need valid consent, so the number list is prepared beforehand with whoever handles privacy on your side.

On short exchanges and structured data, such as an order code or a date, transcription holds up well even with a strong regional accent. Noise from a shop floor or a busy street lowers accuracy: in those cases the voicebot asks the caller to confirm the important details and, if the second attempt fails, passes the call to an operator.

Crafter.ai is the AI Agents Studio by Athics: the no-code platform for building, publishing and running conversational AI agents. The dialogue is designed visually, the knowledge base is loaded from company documents, and the agent connects to the channels and systems the business already uses.

Crafter.ai is an Athics product, built by our own team. Companies that prefer to do it themselves work directly on the platform; those who want support hand us the analysis, the configuration and the integrations. In tailor-made projects Crafter.ai is often the technical base we build the solution on.

Crafter.ai combines large language models with RAG over your documents and with the business rules you set. Models are interchangeable, so the agent can move to a newer model without rebuilding the project, and the points where it must not improvise are pinned to controlled answers.

Yes. The quickest route is a guided demo on one of your own use cases: we load a few of your documents and look together at how the agent answers. Write to us from the contact page and we will set up the session. From there a pilot on one department or one channel is the usual next step.

That is settled during the project with your IT lead: the agent can sit on the platform infrastructure or on the systems you already run, with retention periods agreed in writing. The rule we follow is to keep the minimum for as long as needed, and not to duplicate information that already lives in your business system.

The default setting of an Athics project is that your content serves to answer questions, not to train models shared with anyone else: the agent's knowledge base stays tied to your project. Where third-party models come into play, the data processing terms are checked with whoever handles privacy on your side before go-live.

Compliance covers data processing as a whole rather than a single piece of software, so the useful question is how the agent fits your privacy model. We design agents to support GDPR compliance: access limited to the data actually needed, no incidental collection, roles and retention periods agreed with you before release.

The AI Act, European regulation 2024/1689, requires transparency towards anyone interacting with an artificial intelligence system: the person has to know they are talking to software. In our projects that means an explicit statement at the start of the conversation and a route to a human operator that stays open.