Freelancers lose leads while they sleep or work on existing orders. Research showed the first 3 minutes are critical for closing a sale, and sellers spend 2-3 hours a day just on pre-order communication. The goal was a Personal AI Assistant that works 24/7 on behalf of the freelancer. But creators at the time saw AI more as a threat to their copyrights and the value of their work than as an opportunity. So beyond building a useful tool, I needed to lower the adoption barrier, give freelancers a sense of control, and give buyers transparency. There was no established playbook for building agentic interaction experiences.

Existing solutions didn't address the creator's perspective. Upwork took a search/match-based approach, B2B chatbots like Intercom and Drift had no precedent in a creative context, and freelancer automation startups focused on communication management and lead generation, acting as a manager, not a co-creator. Fiverr Go's differentiator was clear: exclusive training on the freelancer's own work, under their control, with intellectual rights owned by the creator. But the real problem was trust. Creators saw AI as something that takes from them. Competitors' models were built mostly for support and communication automation, not for representing a creator's voice. So I wasn't building a chatbot. I was building an extension of the freelancer — their tone, their portfolio, their rules — while making it transparent enough that buyers knew they were talking to AI. Calibrating how far the assistant should go before handing over was the hardest part. Too early felt broken. Too late felt risky for the creator's reputation.

I designed the conversation flow for both sides: buyer side from first message to purchase or handover, seller side from setup to preview mode. Every element of the assistant had to feel like it came from the freelancer, not from the platform. The AI was infrastructure. The creator's identity was what the buyer should see. Sellers needed real control. I gave them three availability modes (Always on, When offline, Custom schedule) along with the ability to set the assistant's name, tone, and handover rules. Without this, creators wouldn't trust the tool enough to turn it on. On the buyer side, I designed the widget system so the chat wouldn't feel dry. Gig cards, portfolio previews, quick replies, action buttons. These gave the buyer a sense of working with the agent, not just reading responses. Widgets became activation points for the next step, turning conversation into collaboration. Early on, I considered a fully autonomous flow with no handover option, but creators needed an exit ramp, a way to step in when the conversation mattered most.

Through the Model Creation flow and AI Assistant, I helped establish a platform identity that emphasised the humanity of artistic collaboration, supplemented by AI and its ability to extend the creative capacity of freelancers and business owners on the platform. Within the first quarter, 6,000+ sellers activated the platform and 200,000+ buyers interacted with the Fiverr Go Assistant. Creators using the Personal Assistant saw a 50% increase in client conversion rates, which drove Seller Plus adoption up 20% year-over-year. Services segment revenue reached $29.5M in Q1 2025, up 94% year-over-year.