🇩🇪 Maximilian Domann - Product Manager,
🇻🇳 Linh Nguyen - AI Engineer,
🇲🇽 Ricardo Escobar - Software Engineer,
🇮🇳 Yash Chhillar - Software Engineer
An AI plugin in your ERP system that can gathers and analyzes external data for the users to help them make more precise and data-driven decisions without relying on the intuition when planning inventory.
Reduce inventory planning time by
Gather and summary external data that is related to your business
We launched the ai assistant, Popsicle, from 0 to 1 to help an ice cream business in Mexico to better plan their inventory and make data-driven decision based on external data gather by AI technologies.
Different factors influence decision making but they can be complex, unreliable, and time-consuming. The reality is that people usually rely on their gut feeling or intuition to make decisions. The requirement from SAP is that the proposed product accelerate intelligent decision-making for business owners with Generative AI.
interviews with people who work in the field of data analysis or AI technologies (eg. supply chain manager)
interviews with people who work on SAP analytics products (eg. Product manager from SAP Analytic Cloud)
We conducted market research and 11 rounds of interviews to understand users' behaviors when using data analysis platform/software and point of view regarding how ai affect their daily task.
From the market research, we found that only 29% of the employees use data to back up their decisions. Also, from the interview, we learned that despite the vast amount of available data the majority of decisions are still made based on “gut feelings”.
One supply chain manager mentioned that he relies on external data to make decision, however, finding the information he need in a dashboard or external resources is time consuming.
People sometimes don't trust AI technology because it is too complicated for them to use or they don't like what AI tells them how to analyze data since they are already the expert in the field.
Based on our research, we observed a common need to make more data-driven business decisions by reducing the time of gathering and analyzing external data, especially for small-to-medium-size business. We also found a need to build more trust between users and AI technologies. These opportunities inspired us to define the ideation scope in focusing on the user journey from the AI gathering data to analyzing data to the users make decisions.
User Journey Map with 3 high-lighted design target and the HMW questions
We decided to develop an AI plugin in the ERP system that can gather external data such as weather, relevant news, and currency, analyze data, and provide predictions and recommendations for users. With Popsicle, business owners can make more precise and data-driven decisions without relying on intuition during the inventory planning process.
Retrieve relevant and trending information based on a pre-defined prompt using Google's Search API.
Create a summary and title of five different sources (articles) using Open's AI API.
Create a new prompt based on user's score on previously formed prompts using Open's AI API.
Find the relevance and interpretation of the business using Open's AI API. Providing a specific and detailed business context for finding the relevance.
Iterations of prototype
user testing session
We worked collaboratively within an agile framework and conducted product testing on a weekly basis. By adapting the feedback early in the design process, we can iterate our product more efficiently.
In the testing stage, we focused on realizing the usability of the product features and user flows, and users' preferences of visual styles. We got some positive feedbacks for the product concept. Yet, we also found some issues in the design. For that reason, we made some iterations based on our observation.
In the first exploration, we design a search function and news section that invited the user to type the key words that are related to their business and received the external new. We hope the external news can help them better make their decisions.
User don't know what should they type in the search bar and not sure about the connection between the news and their internal data.
In the second exploration, we design a notification function that brings the user's attention to the dashboard they should look at. At the same time, the external section will slide in and provide the ai interpretation for the data and suggested actions.
The way the notification show is not obvious enough and the connection between the chart and external data section is not clear.
In the third exploration, we design a linear information flow to make it easier for users to follow. Users can also hover over the exclamation icon on the page to see the change in the dashboard above.
Users like the linear method of the design, but they mentions there is too much information on the page which makes it too overwhelming to digest.
In the fourth exploration, we expand three sections to full screen and only show the most important information in the beginning. If the users want to learn more, they can expand the card to learn more about the ai interpretation and the connection with their internal data/chat.
Users still think all the information is a bit overwhelming and the information lack hierarchy.
In our final design, we keep the linear information flow that guides the users from top to bottom, from left to right.
In the top left section, we provide the external data that the business owner refers to the most when making decisions, which is the weather, relevant news, and currency.
In the relevant news section, we create a summary and title of five different sources (articles) using Open's AI API. We also allow users to give feedback on whether the news is helpful for them in order to train our ai model.
In the bottom left section, we provide interpretation from the ai to facilitate the analysis process for the users.
We highlight the most important information on the dashboard and provide access to detail by clicking the "more" button to expand the model window.
In the right section, we highlight the recommended actions for users.
User can clearly see what action they can take and expand to see the detail. There is also a question mark icon that allows users to understand where the number comes from in order to increase the credibility of AI.
When presenting the product to our stakeholders, we also create a product roadmap that describes the vision for development. Currently, I already accomplished step 1 and step 2.
In the beginning, I was worried about not doing enough research before designing. But through the agile working methods, I embrace the incomplete product and learn to improve it through product testing in the early stage of the design process.
In a cross-functional team, it is important to be able to quickly get diverse input and align directions from different perspectives. I learned how to provide insights when needed as well as ask for clarification during technical discussions.
When designing, I once struggled with some issues in Figma. By proactively asking for help from another designer in the program, I learn new techniques in a more efficient and effective way.
I found the daily meeting and GitHub backlog really helpful when working in a cross-functional team. They keep each team member engaged with the project and aware of what everyone is working on. In this way, we can work collaboratively to achieve the team’s goals.
