Case Study: ServiceNow turns hand-drawn ideas into workflows with SuperAnnotate

ServiceNow turns hand-drawn sketches into workflows with SuperAnnotate, reaching 95.5% accuracy

ServiceNow, a cloud software company, aimed to launch a new feature powered by a vision-language model that could convert hand-drawn workflow sketches into executable automations. Their main challenge was obtaining a large, high-quality, domain-specific dataset for training, as past outsourcing efforts had resulted in poor quality annotations and delays. They partnered with SuperAnnotate and leveraged its AI data platform and AI Operations Services to overcome this hurdle.

SuperAnnotate implemented a structured, three-stage quality control annotation process, resulting in a high-fidelity training corpus of over 18,000 images. This data was used to fine-tune a model called StarFlow, which achieved 95.5% accuracy on internal validation, outperforming GPT-4o. In real-world use, the model delivered a 40-50% improvement in enterprise workflow interpretation, dramatically accelerating creation and reducing manual effort for ServiceNow's users.

ServiceNow