Optimizing machine learning models for gen AI workflows

The Problem

A multinational technology company was developing a new product that required high-quality AI capabilities. To manage operational costs and quickly expand AI workflows, the organization sought to outsource AI testing. The risks associated with the product’s output, however, included ensuring cultural relevance and compliance with specific geographical regulations.

The Solution

Highspring implemented an incubation workflow with 60 onshore human reviewers to enhance the machine learning model, help ensure high-quality AI-generated ads and establish foundational policies that reduce risks and protect the brand. The team developed guidelines, analyzed AI-curated content and established a policy feedback loop with the company’s internal stakeholders to manage policy changes and reduce subjectivity. The solution also involved expanding the training model for search relevancy and accuracy by demographic and product categorization. Maintaining the quality of the user experience and retaining third-party advertisers were key priorities during development.

Our Impact

Within six weeks, Highspring achieved and maintained a quality rating of 99.5%. The training model expanded the ad engine’s content ranking and recommendations while ensuring compliance with existing integrity measures. This resulted in more efficient modifications and enabled more informed product positioning, store adjustments and clustering. Phase 1 of the project focused on solutions in English, leading to Highspring being selected for Phase 2 and expanding the project to include an additional 14 languages.