Microsoft AI Builder
Senior Designer | Paris, FR
AI Builder is a simple, point-and-click tool within the Microsoft Power Platform that helps non-technical users bring machine learning into their business and use AI to automate processes and predict outcomes.
The challenge
Since most AI Builder users lack a data science background, the UX needed to be intuitive and approachable to build user trust in the technology. Regular user studies and customer interviews were essential for keeping our designs aligned with these goals.
User flows & wireframes
Designing new AI Builder features required extensive exploration before finalizing designs. I often led brainstorm sessions with our broader team to generate innovative ideas. From these sessions I developed user flows and wireframes that guided the team toward the most effective design directions.
Conducting a brainstorm session for a new AI template.
Early wireframe explorations for discovering and creating an AI model.
Research
AI Builder benefits from a dedicated UX research team, which conducts user studies, creates benchmark reports, and collects customer feedback. Product research always shapes and informs the direction of my design work.
Persona maps paint a clear picture of the journey our users go through.
The launch
AI Builder launched in June 2019 to an enthusiastic crowd of Power Platform users. The excitement and positive feedback motivated us to continue refining and expanding its capabilities. AI Builder has since delivered measurable value to the Power Platform business, extending the potential of Power Apps and Power Automate.
AI template modal window with instructions for creating and using the model
CASE STUDY
Form Processing AI template
The Form Processing AI template is a custom model that speeds up data entry by extracting and organizing data from documents such as invoices and tax forms.
1.Understanding the problem
To successfully create a working Form Processing model, users needed to complete four basic steps. However, telemetry data showed a 42% user drop off at step two.
Problem: Auto detected fields doesn't always capture the correct text
Challenge: Design a new UX that lets users select and edit both auto-detected and manually chosen fields
1. Import multiple document samples for the model to use
2. Select which auto detected text fields the model should extract
❗️ 42% user drop off
3. Train your model using the selected text fields
4. Publish your model and use it in apps and flows
2.Brainstorming
I led a brainstorming session to generate a wide range of ideas. After voting, we selected two promising directions to test with users. Although design is rarely a linear process, these steps illustrate how I approach creating a design solution that solves real customer problems.
3.Sketching & Whiteboarding
My initial sketches focused on educating users about the new feature and guiding them through the required selection steps. A whiteboarding session helped refine the chosen direction and fill gaps in the user flow.
4.Prototyping
After finalizing the flow, our team built the concept in Figma, collaborating to create all required screens for a complete design. We developed a working Figma prototype to test the experience with customers.
5.User interviews
Within a day, the prototype was ready for testing and our research team presented to users to gather feedback. The reception was very positive and with minor adjustments, we released our update.
6.Updated release
The improved version of Form Processing launched in spring 2020. Customers welcomed the update, and the success rate for completing form processing models increased by ↑ 30%.
