Sama has introduced Bulk Annotation, a new feature aimed at addressing repetitive labour in AI data labelling processes. The company's latest advancement allows for groups of nearly identical items to be annotated collectively, rather than individually, addressing ongoing inefficiencies for enterprises handling large-scale training datasets.
Operational impact
The repetitive practice of labelling thousands of similar or duplicate items is a familiar challenge across sectors deploying AI. Manual annotation consumes significant resources and can lead to inconsistent datasets. Early pilots of Sama's Bulk Annotation indicate throughput improvements of up to 80% and a reduction in annotation inconsistencies by as much as 25%, according to the company.
Platform approach
The Bulk Annotation system uses embedded machine learning functionality to group duplicates, variants, and near-matches within a dataset. Annotators can then review and classify these groups in a single action, with one label applied across all related items. This model not only saves time but also improves consistency across datasets. Quality assurance processes are also changed, as reviewers are now able to validate at the group level rather than inspecting every item individually.
Sector applications
The solution is designed to be effective for a broad range of industries, including retail, generative AI developers, financial services, and healthcare. Retailers, for example, can annotate entire product families at once. Organisations managing document libraries, including those in compliance-heavy sectors, can benefit from quicker and more consistent data categorisation as their needs evolve. Sama has designed Bulk Annotation to accommodate both fixed and changing data structures, acknowledging the changing demands of enterprise AI programmes.
Workflow integration
The platform groups similar items automatically, without the need for clients to reorganise or pre-process their data. According to Sama, the integration of this tool with their in-house, managed workforce prevents fragmentation found with labour pool-based annotation models. The workflow optimisation from end to end is positioned as a distinguishing aspect of the offering.
Development background
The company developed Bulk Annotation based on input from both its annotation workforce and its clients. UX research and feedback gathered during the R&D process contributed to the platform's design and implementation.
"We created Bulk Annotation by listening to our workforce and clients," said Karan Vasdev, Product Manager, Sama.
Bulk Annotation is available to all Sama clients, with ongoing projects set to transition to the new system. The company continues to focus on supporting enterprise AI deployments by reducing the impact of poor data quality on project outcomes.
