How we leveraged AI as a framework for continuous user-feedback.

Just like every product company with a mobile application, we have App Store and Play Store reviews flowing in every day. As our userbase increased, the daily count of these reviews, on both stores followed suit and it often bothered us that we were not doing much with this data.

This was what birthed the first version of Lamurudu which was pretty much a bot that did a daily poll of reviews from both stores and posted them to Slack. 

The goal at the time was to make the reviews more accessible by posting them on Slack, where the team resides… and yes, it worked! Every day by 10:38pm, Lamurudu, the review bot would post all the reviews in the last 24 hours from both Play Store and App Store. Here’s a sample:

But then, it didn’t take long before the itch crept up again. A lot of data was coming from the reviews and yes, we could see them, but were we really getting the best value from this? Was there perhaps a way we could make this more efficient? Of course, yes! So, time to go back to the drawing board.

The answer was obvious and it was staring us in the face: not many people were going to keep reading those reviews every day. We needed some form of comprehensive summary, we needed some context and relevance, we needed the reviews to speak to us. So, we went the route of generating sentiment analysis via AI.

I fed the reviews to Sisi, our in-house Gemini wrapper and it did wonders! Not only was Sisi able to render a clear sentiment analysis based on weekly data, Sisi was also able to provide actionable feedback based on the reviews. To crown it all, Sisi could also tag relevant teams for each feedback. 

Amazing right? In practical terms, here’s how Lamurudu V2 works:

• Runs every Friday at 10 AM via a scheduled job.
• Pulls reviews from Play Store and App Store over the past week.
• Groups the reviews (by rating) and sends them to a dedicated Slack channel for visibility.
• Performs sentiment analysis to extract common trends, issues and feedback.
• Posts a structured summary (essentially a TL;DR) of user sentiment—to wrap it up. It also tags different teams wherever relevant.

That’s it.

Here’s a view of how 10 AM looks like on our app-review channel on Slack:

The results were almost instant:

• Individual teams were now being tagged directly on specific, well-defined issues, so we saw a significant rise in the rate at which issues that came from user reviews were resolved.
• We started harvesting awesome product enhancement ideas from some of the reviews.
• We felt more connected to our users than we had ever been. It’s a distributed effort now. We actively celebrate the positive reviews and collectively work on the not-so-positive ones.

What’s next?

Quite frankly, the possibilities are endless. Perhaps extending the bot’s capabilities to creating linear tickets for bugs? Auto-generated replies to common reviews? There are no limits!

Stay tuned. We are cooking. 😉🧑‍🍳