Interview with Pierre-Eric Marchandet, CTO at DialOnce, Specialist in AI for Customer Relations
- How does an effective chatbot work for customer relations? How does AI analyze and interpret the question?
1. A good chatbot for customer relations knows how to recognize contact patterns and differentiate :
Patterns that will call upon the generative model to provide precise information from your knowledge bases (such as product descriptions, general conditions, opening hours, tutorials, etc.).
Critical language patterns for which a static response should take over to control the response and enable omnichannel capabilities (the customer can be guided to another resolution channel, according to predefined rules).
2. It is based on a large business dataset as training data.
Yes, a good customer relations chatbot is trained on a very extensive business dataset. By this, I mean that if you rely on generative AI and your knowledge base to respond to your customers, you risk having an unsatisfactory comprehension rate. The unique repository of customer relations intentions and solutions that we have built over time allows the chatbot to be more efficient and, above all, to master and control the response provided.
3. It can qualify the request with precision.
Indeed, because this chatbot is trained for customer relations and has a clear view of the typologies of intentions and associated solutions, it can accurately qualify the user’s request. It can therefore ask the user for additional information to remove any ambiguity from the request.
4. It has a rich response content database to fuel generative responses.
In addition to the customer relations data ingested into our customer relations repository, the customer relations chatbot must be able to refer to a large company-specific database (website, contracts, general conditions, FAQs, customer space, etc.). It must be able to understand the data at its disposal and sift through the data that may be useful to the user’s expectations, distinguishing them from outdated or inappropriate data.
5. It does not hallucinate during responses and admits when it doesn’t know.
Of course, hallucinations are controlled, and responses are monitored. A good customer relations bot must be finely tuned on a quality database to achieve good results.
6. The cost of the query is controlled through prompt configuration.
This is a point that is often overlooked but essential! Indeed, if you want to control the operating costs of the chatbot, it is better to have, beforehand:
Carefully selected all the documents you make available to it.
Optimized the prompt to limit the number of queries to OpenAI.
It’s a matter of expertise, and at DialOnce, we know how to assist our clients in configuring and optimizing their customer relations chatbots.
7. The prompt is adapted to the request or the company’s style.
One last point; consider the chatbot’s ability to adapt to the tone requested. It can respond as if it were a knowledgeable customer service advisor, for example, according to your company’s guidelines (informal or formal language, formal or familiar language), but it can also adapt to the user’s language and translate responses instantly.
Thank you for these valuable insights, Pierre-Eric !
DialOnce is at your disposal to present its solutions and introduce you to your future playground !
