Turning chaotic social chatter into warm, on-brand conversations automatically

It's 2 AM, and Maya, Head of Growth at ACME Skincare, wakes up to a notification. A defaming comment has been sitting at the top of their highest-performing Instagram ad for three hours. By morning, their ad engagement has dropped 30%, and potential customers are seeing negativity before they even learn about the product.

This wasn't unique to Maya. Across mid-sized eCommerce brands pulling in $2M+ annually, social teams were drowning. They were manually scanning hundreds of comments daily, deciding what deserved responses, crafting replies that stayed on brand, and doing it all fast enough to protect conversion rates.

E-commerce teams are expected to manage multiple social channels, respond to customer comments, and keep engagement high, all while running the rest of their business. This often means juggling dozens of messages at once and trying not to miss anything important. Keeping up can be exhausting, especially for businesses without a dedicated social media team. Missed comments can lead to lost sales, frustrated customers, or inconsistent brand messaging. This is why we designed Blueberry Social, a tool that brings every interaction into one clear inbox and surfaces the most important messages first. With AI assistance, teams can respond faster, maintain their brand voice, and turn everyday conversations into sales opportunities.

Halfway through the project, everything changed. We had spent weeks building a social listening tool for startup founders. Then the co-founders came back with new data: mid-sized eCommerce brands were the real opportunity.

Our entire ideal customer profile had flipped. Instead of solo entrepreneurs, we were now designing for growth teams at brands doing millions in revenue.

We made a deliberate decision to move fast. We used AI tools like V0 to spin up rapid prototypes, ran weekly testing sessions, and iterated based on real feedback. The goal wasn't perfection. It was reducing ambiguity fast enough to build something valuable.

We talked to growth leads and social media managers at eCommerce brands. Through testing sessions, we discovered their real pain points. Below are some key points we have gathered through interviews, supported with quotes from our participants.

After talking to our participants, we brainstormed on solutions to the problem. This brings us to our first how-might-we statement: How might we deliver immediate value while gathering enough information to make AI responses feel genuinely on-brand?

From that HMW statement, we branched off into several explorations...

During our first test, Participant 1 was immediately confused: "Inbox? Is he looking at the DMs?" The terminology we thought was obvious was completely foreign to users whose mental model came from email, not social platforms.

The emoji problem: Participant 6 called out something critical: "Some of these emojis are things that wouldn't be used likely. Less emojis the better." The AI responses were clearly AI-generated because they were too cheerful, too generic.

When Participant 6 regenerated responses, he said, "I find it much better and easier to understand. These are good responses." Users could spot inauthenticity instantly. Our AI had to match their actual brand voice, not what an AI thinks customer service sounds like.

What we learned about filters: Participant 1 wanted to "crank through it as fast as possible" and filter by post. Participant 6 reinforced this: "Subsets for data (filters): ADS first. Bring those to the top."

But users didn't want endless filtering options. They wanted smart defaults that surfaced what mattered most, with the option to dig deeper if needed. Clarity over choice.

Participant 6 gave this approach a 6.5/7: "Easy to navigate. Each one has good buttons around each section. It would be nice to have a couple of presets to help out customers."

The preset insight: Multiple users mentioned wanting preset brand voices. Not everyone would want to start from scratch. Participant 5 asked, "Can you save multiple brand voices?" She imagined having different tones for different contexts.

The scariest feature for users was full automation. Testing revealed nuanced feelings. Users were open to it, but only after the AI proved itself. We added intentional friction with confirmation prompts. Transparency builds trust, not assumptions.

After weeks working with Shadcn components, we hit a wall. Blueberry looked like every other B2B SaaS tool. Clean, functional, forgettable.

The founders wanted Blueberry to feel friendly and approachable, not corporate. We explored references together: Duolingo's playfulness, Shopify's polish, Discord's casual vibe. We didn't want to go full cartoonish, but we also didn't want buttoned-up corporate.

What emerged: bright, inviting colors that made the interface feel warm. Rounded edges everywhere to soften the experience. And most importantly, our mascot Bloo and blueberry iconography threaded throughout to reinforce personality.

Blueberry taught me that designing AI products isn't about making the technology impressive. It's about making users feel confident and in control.

The best AI tools don't replace humans—they amplify them. Blueberry gave social teams superpowers: respond to everyone, protect their brand, capture sales opportunities, and still have time to focus on strategy.

That shift from "I don't trust AI" to "I could actually see myself using this" was everything.