AI for customer care: faster answers without losing the company’s tone

Customer care has a repetition problem

Opening hours, the status of a request, required documents, how a service works. A large share of customer requests revolve around a few questions, and every answer takes time. On top of that, quality varies: it depends on who replies and how busy they are at that moment.

This article is about existing customers and support. For prospects visiting the website, the perspective is different and we cover it separately.

Which requests can be automated and which cannot

The starting point is to split requests into two groups. These suit an assistant well:

• frequent questions with a known answer, such as hours, procedures and required documents;
• the status of a request, when the data is available in a company system;
• collecting the initial details of a request, as a conversation;
• pointers to where to find a document or form.

These should be left to a person:

• delicate complaints;
• requests with exceptions or that need negotiating;
• urgent or emotionally charged situations;
• decisions on refunds, terms or contracts.

How not to lose the company's tone

An assistant should sound like the company, not like a form. A few choices help a lot:

• A tone guide with examples: phrases to use and phrases to avoid.
• Answers already validated by the team as a starting point.
• Rules on register: how to greet, how to apologise, formal or informal address.
• A sample review of the answers before release.

Answers must also rest on approved sources, not on what the assistant finds elsewhere. On assistants and company data we wrote in Company data and AI: what you can share with an assistant and what you cannot.

When to hand over to a person

The most important rule is to decide in advance when the assistant stops. Some typical cases:

• the customer asks to speak to a person;
• the answer is not in the approved sources;
• the request concerns money, contracts or complaints;
• the tone of the message shows distress or urgency.

When the handover happens, the person should receive the whole conversation, so the customer does not have to repeat everything. In a custom project these rules are defined together with the team that answers today.

How to set up and check the assistant over time

1. Start from real questions. Take emails, requests and conversations from recent months and pick out the questions that come back most.
2. Write approved answers and sources. Valid answers become the base of the assistant, with a person responsible for keeping them up to date.
3. Define the cases to hand over to a person. Put the criteria in writing, starting from the list above.
4. Test on a small perimeter. One type of request or one group of customers, with human review of the answers.
5. Review conversations at regular intervals. Fix the sources when you find inaccurate answers and update the rules.

If the assistant must read the status of an order or request, it has to be connected to company systems. This is the topic of the page on AI integration and workflows (Italian page); a similar case, for events, is in Integrare la segreteria organizzativa con i sistemi aziendali: come farlo bene (Italian).

A concrete example in events is the webRay® AI assistant, which answers about programme, access and services using approved content. More examples are in Segreteria eventi 4.0: come usare l'IA (Italian) and IA negli eventi: come rivoluzionare l'organizzazione (Italian). If you do not yet know which task to start with, see Where to start with AI in business: how to choose the first process to improve.

Before switching on an assistant for customers

Frequently asked questions