How UiPath uses AI to Automate Product Feedback Collection
Creating products that customers really want is the goal of any company’s product and engineering teams. With the right tools to understand and meet customer needs, companies can get higher Net Promoter Scores (NPS), resulting in higher new customer conversion rates and retention rates. Research shows loyalty leaders grow revenues 2.5 times as fast as their peers and generate two to five times shareholder returns.
We’ve explored different tools to process product feedback until we finally decided to use our own products—UiPath AI Center and other UiPath products. We chose AI Center not only because it automates feedback classification, but because it is easy to use.
Our unprocessed user feedback dropped by 70% within two months. This solution will save me and my team 3,000 hours of manual work (per year) in processing user feedback.
With a liberal arts background, I have to admit I wouldn’t have guessed I would be writing a blog post about how to use artificial intelligence (AI) to improve how a company collects product feedback. I wouldn’t have guessed I would be using AI at all. That’s part of why I’m writing this blog post—not only can you use automation to do manual tasks for you, having a technical background isn’t a requirement to learn and use automation.
A little bit more about me: I joined UiPath Forum in June 2018 to learn how to build UiPath Robots and to help other users whenever I would see an opportunity to share my newly acquired knowledge. My contributions were eventually rewarded with a full-time position as a UiPath Forum Community Manager, which in time allowed me to start building robots that help UiPath efficiently gather user feedback from our Forum. Nowadays, I enjoy being able to build several automations that help us integrate our Forum with a variety of internal tools and facilitate the discussion about UiPath products with our user community.
Now, let’s get back to improving how product feedback is collected.
UiPath chose Productboard as the central platform that helps us collect and process user feedback. It allows us to gather entries from different sources, and then process and prioritize them accordingly. Until now, it was a manual effort to get the right product manager to see the feedback relevant to their product. For example, each new Productboard entry needs to be tagged with the correct product tag, which is the first important step in getting the feedback properly processed. Because manually tagging user feedback became a burden and took time away from processing feedback, we decided to look into ways to improve this.
We provided Productboard with our feedback, but a big ask like that would also mean a long implementation time. Therefore, we decided to use our own toolset to see if we can make this part of the process more efficient.
In the end, we used the UiPath Platform and the result allowed us to save a lot of time. The manual task of tagging user feedback became a job for our UiPath Assistant attended automation, fueled by English Text Classification model, one of our out-of-the-box (OOTB) UiPath AI Center machine learning (ML) models retrained to categorize user feedback. We chose AI Center because it was the most viable way to automatically classify the amount of feedback that we receive. Before, someone had to manually read over each new feedback entry and assign a product tag that matched the content. Now, all they have to do is to run the AI-enabled automation that does it for them, and eventually check the execution report which is conveniently saved as a Microsoft Excel file in UiPath Data Service.
To make all this happen, we presented the business case for this automation internally through UiPath Automation Hub. Everything starts as an idea, but you need something that will allow you to prioritize the development of these ideas based on several factors. In this case, we submitted three automations that would work together to help us reduce our Productboard workload. Eventually, we got green light to develop these processes.
Having a list of the basic requirements was one thing but knowing how to address them was another. Especially when one does not have much experience with ML, which was true in my case. Therefore, addressing the most important requirement first seemed like a good starting point.
Thankfully, UiPath AI Center provides some OOTB pre-built ML models, which can be retrained with one’s custom data. In our case, it came down to extracting Productboard feedback that was already properly tagged and then using it to retrain the model that would be able to categorize user feedback for us. After a couple of days of work for us and 16 hours by AI Center (to learn based on our input data), we had our first working ML model that could translate user feedback into specific product tags. It felt magical, especially because of how easy it was to achieve given my lack of experience with the matter. The end result was an automation that was able to convert Productboard data into a single input file for AI Center to learn on.
The next step was building an automation that would use these newly acquired ML capabilities to tag user feedback in Productboard. The process was developed using UiPath Studio, but it felt natural to combine multiple functionalities of UiPath Platform to make it as convenient to deploy and to use as possible. As such, we developed an attended process which is now able to perform the required task. These processes are deployed to UiPath Assistant with our Automation Cloud™ Orchestrator, which is a simple and efficient way to distribute them.
The process can be started at any point from UiPath Assistant. When launched, it will find the pieces of feedback that are without a product tag and it will then use the custom trained ML model to figure out the proper tag for each of them. When finished, it produces a simple output Excel file and uploads it to a Data Service entity in Automation Cloud. This makes it easy to double check the process output without having to dig through individual process execution logs.
The overall benefit of this process was immediately seen when it became available to our product managers. Our unprocessed user feedback dropped by 70% within two months. This solution will save me and my team 3,000 hours of manual work (per year) in processing user feedback.
In this blog, I’ve shared my experience of classifying user feedback using the English Text Classification model. AI Center also offers classification models to work with other languages, including French and Japanese. The classification models can be used to solve a variety of business problems including processing emails, classifying resumes, customer surveys, and IT service tickets. Please go to the English Text Classification documentation page to learn how to use the model, and AI Center product page to discover more ML models.
To learn more about AI Center, please visit the product page.
I also encourage you to register for the UiPath AI Summit. The free, virtual event takes place over several weeks, with different AI sessions every week. Whether you're new to AI or an experienced practitioner, there are sessions for you.