How Hugo Helped A Leading Innovator Confront Inclusion Gaps

Our client stands at the forefront of technological innovation. For over two decades, they have pushed boundaries in search, AI, and software – bettering lives through empowering information and data-driven solutions.

The Need for Inclusive Data

This trailblazing company shoulders great responsibility. How do we ensure all groups benefit from life-changing technologies as capabilities advance? How can diverse needs and preferences be consistently understood?

Our partner set ambitious inclusion goals – wanting products that resonate across their global user base. However, achieving equitable AI requires expansive datasets…and representation gaps persist.

In this particular instance, the client needed to enhance visual recognition for its Pixel photo software through a more diverse data set of facial imagery. Data for users with darker skin tones proved notably lacking. The client partnered with Hugo on a mission to fill this critical gap – collecting 50,000 photo ID videos of people with undertrained skin tones each month.

Why Hugo: Unrivaled Geographic Reach

With operations centers spanning strategic African regions, Hugo maintains unparalleled on-the-ground access to a rich diversity of black communities and offers teams that innately grasp intricacies around language, culture, logistics, and demographics.

This powerful localized presence enabled swift, tailored recruitment; within 14 days, Hugo had developed hyper-targeted campaigns and tapped willing participants across five countries to match the client’s criteria.

As an added advantage, having teams with a grasp of local contexts allowed for nimble adaptation to challenges. When pandemic impacts threatened timelines, our distributed model pivoted quickly across locations, minimizing delays through coordinated troubleshooting.

Seasoned Data Practices Aligned to ML Rigor

As veterans managing large-scale global data projects, Hugo brought battle-hardened practices to uphold quality:

• Multi-layer QA defenses prevent errors, with automated checks supplemented by manual reviews of issues like motion blur.
• Sensor arrays and capture methods purpose-built for computer vision ensure consistent ML-grade inputs.
• Specialist privacy/consent teams handle identification ethically across regions, supporting inclusion with cultural awareness.

At Hugo, protocols are tailored to each client’s parameters. For this client, the team created five custom environments (outdoors, kitchen, living room, bedroom, office) and mapped out the different equipment specifications for each to ensure the IDs collected aligned with the model’s guardrails.

The Outcome: Progressing Inclusion

In just three months, Hugo delivered the entire project scope:

• 600,000 tailored photo IDs collected (against a 12-month target for the same number of IDs)
• 97% participating acceptance rate
• 3-4% opportunity cost, far below 10% goal

But more than metrics is needed to capture the full impact. With new voices represented, facial recognition technology will be better equipped to see diverse users; experiences will feel more welcoming, and friction will fade.

Our partner is propelling this progress across products. As computer vision continues advancing, segments once left behind are now factored into the future. “Hugo’s commitment to inclusion and execution were game-changing,” noted Michelle Z., the client’s Machine Learning Lead. “This data unlocks innovation that serves and empowers more people.”