Chatbot de Copilot

Set up your own virtual assistant from scratch with Microsoft Copilot Studio

Microsoft Copilot Studio is an innovative tool that allows us to create and modify our own copilots or virtual assistants. Thanks to this, we can have several copilots for different situations and they can be fed from different data sources. Importantly, they can always interact with other systems. Want to know more?

In the following, we will discuss the key points to be able to generate our first copilot successfully.

MICROSOFT COPILOT DATA SOURCE AND INSTRUCTIONS

One of the most important points when generating a new copilot is the capacity we have to obtain answers related to a specific data source. As we have seen in other Itequia articles, Microsoft provides us with copilot versions in which we have access to all the content of the web or to the content of our tenant’s documentation. These copilots that we can generate have a clear characteristic and it is that we can limit the level of the information on which it searches:

  • Websites
  • SharePoint and OneDrive sites (Only to your documentation)
  • Files we upload manually

Here we realise a big change and that is that we can define exactly what we want it to talk to us about.

IA generativa y Microsoft Copilot

In addition to this data source we provide, we will be able to give you instructions so that the co-pilot will be able to respond in the most appropriate way.

Instrucciones Microsoft Copilot

MICROSOFT COPILOT ENTITIES AND TOPICS

Another important point for the understanding of Microsoft Copilot is the entities and themes. When we generate a new copilot it comes with default entities and themes which we can take advantage of.

Regarding entities, we must understand that they are a series of rules that allow us to classify words or groups of words when we converse with the virtual assistant.

Entidades y temas de Microsoft Copilot

In the image above we can see a series of predefined entities. So our language will be able to extract them. However, we can generate new ones using closed lists of words or regular expressions.

Entidades de Microsoft Copilot

In terms of issues, we can highlight 3 ways of understanding them. The first one is achieved when we do not use artificial intelligence at all. If we do not activate the dynamic chaining triggers, the copilot will not be able to direct us only towards a topic. Let’s look at the following example.

Chat Microsoft Copilot

Even if we want it to direct us to a topic, the copilot is not able to go automatically. It needs concrete information. However, let’s look at another copilot where the dynamic chaining capability is used.

Chat Microsoft Copilot

We see here how it has already been able to go to the topic without having to make a pre-selection.

Finally, there is a last functionality called conversational add-ons where we can start a topic by explaining a situation to it in natural language.

Desencadenador Microsoft Copilot

The usefulness of this type of topic trigger is that we can use the person’s phrase or text for the next actions in the topic.

ACTIONS AND THEIR USEFULNESS IN THE ISSUES

In the previous point we talked about how the virtual assistant detects whether we are talking about one topic or another. However, we have not talked about what a topic itself is. Within them and from the trigger, a series of actions will be used where copilot can ask us or use the information we provide to generate answers or perform actions on other systems.

Mensaje Microsoft Copilot

We can go from the easiest, such as sending us a message or asking us for a piece of data and then using it as in the image above; to setting a series of conditions to branch the flow as in the image below.

Mensajes Microsoft Copilot

Or even the information collected can be used to make integrations with other external systems through what are called actions.

Acciones Microsoft Copilot

In this image we can see how we can integrate with a system external to Microsoft such as Jira.

IA GENERATIVE IN THE RESPONSES

A very interesting configuration to use in the robot’s responses is to use generative AI responses. This can be done by ticking this checkbox in the Generative AI section.

Once this option has been checked, we can use the action to generate AI responses. Let’s see how it works:

Respuestas Microsoft Copilot

In the Input we will mark what we want to generate, but we can also set up several things within the configuration of the action.

Chatbot Microsoft Copilot

On the one hand, we can moderate their content and set guidelines for their behaviour.

Datos Microsoft Copilot

Also, we can add up to four Microsoft tenant websites and sites so that it responds only with the internal information we provide.

Servicios Azure OpenAi

In addition, we will be able to access custom data sources in the form of a table that we will generate ourselves during the execution of the theme or connect directly to Azure Open AI.

Mensaje Microsoft Copilot

Finally, we will see how we can make the copilot answer directly or get the answer in a variable to treat it as we see fit.

Microsoft Copilot Studio provides us, thanks to a simple interface, with the ability to have customised copilots within our company and targeted at different user groups.

CONCLUSIONS

The ability to use different data sources and the possibility to use AI-generated action responses gives us the possibility to harness the full power of AI. In a simple way, it will be possible to access information from our systems and use co-pilots for a multitude of scenarios. The AI theme continues to grow and with it all the functionalities that will be offered with this fantastic product.

Adrián Cano – Lead Developer at Itequia