Insights AI based assistant 

For the latter consideration, the stakeholders looked for external help. A Proof of Concept (POC) within a product-led approach was suggested the best way to proceed, and the stakeholders agreed. From a technical perspective, the project involved delivering an application with human language capacities that connected to the Carbery CRM data, providing answers back in human language throughout an integration to Microsoft Teams.

After researching Copilot as a solution, the engineers concluded that it wasn't mature enough for its use (at that time) for this occasion. After exploring the integration of the chatbot with Microsoft Teams, the user experience in handling connection delays was suboptimal. Answers taking more than five seconds to arrive couldn't be placed in the question thread, leading to a workaround of posting answers in a new thread. However, this workaround introduced additional delays and created confusion in the channel. The issue was unrelated to the AI, it was only related to the interface. The approach to this problem was to develop a custom chatbot, which will allow the customer to compare performance between the custom chatbot and the Teams-based interface.

I started collaborating with the cross-functional team of developers, Data Scientist and Product Manager at this point in the Product Discovery Phase. Now that it was known that a custom chatbot was needed, I had to identify the requirements to be met according to the project scope, needs and objectives from a UX perspective:

List of requirements aligned with the project's scope, needs, and objectives.

So, let's find out an answer to all our questions to better define the problem and start wireframing.

Following the POC premise of accessing Carbery's data in a simple way, it is important to have a simple interface that gathers all the minimum elements that can make the chatbot not only functional but reaching stakeholders minimum feasible requirements (ask questions against their CRM in human language, and get accurate answers). In this sense, and as starting point, three screens are the minimum needed:

• Empty state.

• Processing the request's (waiting) state.

• Answer provided.

Let's see if we will need more or not!

Visually and functionally speaking, the design will remain simple as it is influenced by the limits established for the POC:

• The user interaction will be considered at a basic level: One question - one answer. Users ask a question and get an answer. It is suggested to improve the users' interaction for future stages, but not for the POC. Some recommendations made for next steps were rephrasing the users' prompt or giving examples of different answers so the user can choose the more accurate to help train the system.

• The different variations of errors won't be considered for the POC. Therefore, visually speaking there is no need to find a solution on how to help the user in case of getting back a false answer, or no answers due to a technical issue.

• Something similar will happen with the verification feature, that will be defined for future implementation but not for the POC. At this point there is no need to include any element in the design that can allow the user to verify the accuracy of the answer provided by the chatbot.

• The user journey is as well out of the POC picture, it is not relevant how or when or why or what the user does until it has to ask something to the chatbot or what is he going to do after. It is only relevant to ask and get an accurate answer. 

• All these limitations also eliminate the need of including historical chats or a navigation menu.

• Nevertheless, the time and date elements are needed to identify how long it takes the system to retrieve some information. Along the same idea, it is crucial to allow the user know when the system is working/thinking or not.

• For the POC there is no need to include an exit option for the chatbot. Since the prototype runs in the browser, the navigation bar will handle this function.

Following this, the design will be very austere. It will only gather the basic elements, components and screens:

• The elements for the main screen will be only two: Search field, and send icon. 

• Subsequent screens also will include: text field (where the user's prompt will appear), text field (where the chatbot's answer will appear), and another text field (where the status of the query will appear). These text fields will have to change their appearance based on state and user interactions.

For the selection of the appropriate Design System, I took into consideration different parameters:

• That the predefined components did not have to be modified/adapted excessively, meaning that the components were compatible with elements that were going to be used for the design.

• That the components could be modified, if necessary.

• That the developers had previous experience with that Design System.

• That the languages the developers were going to use were compatible with the selected Design System.

According to these parameters and after checking different Design Systems and discussing with the developers, Material Design was the selected Design System.

The use of human language was a basic requirement given that the questions were going to be asked by the workers themselves. In turn, its use will give rise to the interface that will be used, a natural language interface, as this one allows users to interact with a system in a conversational manner.

But before I start sketching, let’s take some time to better understand and get to know the users of the design and to identify the main persona in which the POC will be focused. This will also help in understanding what they expect to see and how they expect to use it.

So, let’s highlight the different teams/people who used/ wanted to use data. This info was taken to group the users in a task-based form:

• Finance

• Commercial Ops

• Commercial Director

• Accounts Managers

Prospective data users: the different teams/people who used/wanted to use data.

A workshop was recommended to pinpoint user problems, needs, and expectations, with the aim of gaining valuable insights for the persona development. This workshop provided a deeper understanding of various potential-user profiles and their information requirements, with a focus on uncovering frustrations and pain points related to their information needs.

Sorting and analysing all the information from the workshop also helped me translate similar and outstanding insights, identifying different personal profiles as Leadership, Account Manager, Sales and Marketing Manager.