AI recommendation
Brand new feature in the robot training environment that accurately selects data on websites, helping users reduce human errors and save time in robot training.
OVERVIEW
While Browse AI has always provided a seamless robot training process, users still struggled with training the robot on certain data due to technical limitations. This gap in functionality often led to manual selection errors, slow training processes, and increased reliance on Customer Support.
To solve this, we introduced an AI feature that simplifies the training process by automatically selecting accurate data, labeling datasets, and defining pagination. By doing so, we reduced the common errors caused by manual setup and brought our product’s value of ‘ease of use’ to the forefront.
By leveraging AI technology effectively, we’ve been able to boost task completion rates, reduce the time users spend on training, and minimize human errors - ultimately cutting down the demand on Customer Support and enhancing overall efficiency.
MY ROLE
Lead Product Designer - Brainstorm Session, Research, Wireframe, User Flows, High fidelity design, Prepare assets, Dev Handoff, A/B testing
Collaborated with 1 Product Manager and 3 Engineers.
TIMELINE & STATUS
2.5 months, Launched in May 2024
IMPACT
Increased positive accuracy ratings by 7%
Improved robot studio completion rates by 7%
Raised user satisfaction by 8%
The easiest way to extract and monitor data from any website.
This is one of the key value Browse AI have. In order for us to continue reach that point, we were on a journey to continue iterating the current experience in the robot studio where we emphasize ‘ease of use’ with increasing robot completion rate with higher success.
👋 Platform Context (Visit Website)
Robot Studio is a robot training environment within Browse AI where the user can easily train the robot to extract specific data and monitor changes. It offers step by step guidance for the user to easily select data and label the data for the robot to track those data in the future.
Sneak peek
Instead of manually selecting data sets, users now can receive AI recommended datasets to help them reduce time of data selection and labelling datasets.
RECOMMENDED DATASETS
AUTOMATED PAGINATION
By automatically selecting the pagination and also enhancing the visualization of the each options, it reduces the complexity of choosing the pagination which decrease the overall error states and CS human support work.
🛑 Data is so important but users struggle to train the robot to extract that data.
Problem Space
From interviews with users, one insight stood out to us. That data accuracy was critical. As long as they get the data accurately, they are willing to learn how to train the robot.
So we analyzed users' behaviours on Robot Studio where users are training the robot to extract the data later on. By watching the user recordings and testing users in Robot Studio, we uncovered a key challenges that impacted data accuracy and user experience as well.
👆 Struggling to train the robot comes from technical limitations.
Users are unable to train robots for specific datasets they need on webpages because certain elements can’t be selected with the user's cursor due to technical limitations.
THE STRUGGLE
USERS FEEL
Let’s leverage AI to guide users in training robots, enhancing accuracy and improving ease of use.
The process
1. Guiding principles
We began by establishing guiding principles for designing AI, recognizing it as the key to solving our problem. Drawing from user interviews and team discussions, we prioritized two key aspects below. These guiding principles helped us stay on track with our goals in mind.
Transparency & communication
User empowerment
2. Diverging with rapid sketches
I needed to quickly get up to speed on the project, having just joined the company three weeks prior. To accelerate my onboarding, I hosted a brainstorming workshop with a fellow designer who had been with the company for over two years. This helped me tap into their expertise and gain valuable insights to inform the design process.
3. Translating into pixels and grids
Building on the brainstorming session sketches, I explored various ways to showcase AI recommendations and account for edge cases. I worked closely with engineers to understand technical limitations, ensuring the designs were both feasible and effective.
4. Iterations
Through design feedback and ongoing collaboration with engineers, I iterated on various designs to introduce AI recommendations at the right moment.
Our goal was to create a ‘whoa’ moment for users when they received their AI-generated results. Since the loading state played a crucial role in keeping users engaged and preventing drop-off, I dedicated time to exploring ways to make the wait feel intentional and seamless.
after multiple discussions and iterations…
Final experience
BEFORE
TRANSPARENCY & COMMUNICATION
AFTER
USER EMPOWERMENT
Flexible editing with use autonomy
In addition to displaying AI-recommended datasets, it was essential to give users control over their choices. They could review and flexibly customize the AI suggestions or opt for a manual flow, selecting specific datasets to extract based on their needs.
Edit data column
Select data manually
Visual pagination method
After design
For launch, we implemented A/B testing over several weeks to identify the optimal timing for introducing AI recommendations. Our goal was to seamlessly integrate recommendations—minimizing disruptions to the user flow while ensuring easy access to reliable datasets at the right moment.
A/B testing
Through testing, we discovered that allowing users to select the list they wanted to extract—while AI recommended datasets within that list—led to greater accuracy and faster task completion. Based on these insights, we made a final pivot before launch, rolling out this approach to 100% of users.
More A/B testing
The impact
After launch, we saw a positive impact on data extraction, including higher user satisfaction, increased task completion rates, and improved data accuracy. User feedback on Hotjar also highlighted positive experiences with AI recommendations in Robot Studio.
Based on past experiments, task completion rate is the strongest indicator of subscription growth and user retention. Given these results, we believe this feature has brought us closer to our business vision.
SUCCESS METRICS
Looking back
Looking back, I realize the importance of maintaining an ongoing relationship with our users throughout the process. If I could go back, I would prioritize continuous engagement - regularly checking in, gathering feedback, and understanding their evolving needs. This would not only ensure that we’re building alongside them but also provide valuable insights into user sentiment, helping us refine and improve the experience in a more user-centric way.
BUILDING STRONGER USER RELATIONSHIP
One of my biggest takeaways from this project was the critical role of resilience in design. Adapting to challenges, experimenting with different approaches, and iterating quickly were key to refining the user experience. By working collaboratively as a team, we tested multiple flows, learned from our failures, and made necessary adjustments. This iterative approach ultimately allowed us to create a more seamless experience for users while driving meaningful impact for the business.
POWER OF RESILIENCE AND ITERATION