Tweet Hunter Case Study

Tweet Hunter built a Twitter growth tool, continuing the experiment of launching a new product every week from its early founding days. For a distribution channel, it leveraged the 2,000 Twitter followers it had already secured, and whether a product generated revenue was the standard for validation.

The key partner was AI21. The team moved on from legacy GPT-3, choosing fine-tuning flexibility and customized pricing, and AI21 lowered the cost of scaling by pricing usage of custom fine-tuned models the same as foundation models. The fact that customization with other LLMs cost roughly 6x more also made a difference.

Generative AI for social media was tricky, requiring adjustments for character limits, hashtags, and even tone. Tweet Hunter refined its features by training on its own data using AI21's 3-click custom model training.

The features were divided into four pillars.

• Thread idea generator
• Hook generator
• Tweet writer
• Tweet extender

This flow connected naturally, completing the product experience, and the service secured over 5,000 paying customers.

The biggest challenge was training data quality. Data gathered through freelancers and spreadsheets lacked consistency, but once diverse tweet datasets were trained using AI21 Studio's features, recommendation quality improved quickly. Model quality could also be verified before launch, and as a result, Thibault Louis-Lucas's personal account grew from 2,000 to 60,000 followers.

Tweet Hunter achieved 1M ARR and an eight-figure exit within a year, and was subsequently acquired by lempire. Going forward, the company plans to expand AI21 to other product lines as well, including personalized cold emails, automated follow-up messages, and multi-channel lead management.