In short
Every AI system that sees, hears or speaks first has to learn from labelled examples, and someone has to do that labelling well. Wendy Gonzalez, President and Interim CEO of Samasource, argues that this work can do two jobs at once: deliver the high-quality training data companies need, and give people living below the poverty line a first real job and a lasting career path. She backs the case with Samasource's own impact figures and the stories of workers in Kenya and Uganda.
Key takeaways
- Explain what training data is and why AI models cannot exist without it.
- Recognize why last-mile labelling quality matters for safety-critical systems such as self-driving cars.
- Describe an impact hiring model that recruits below the poverty line, trains people and pays living wages.
- Measure social impact by tracking former workers' earnings years after they leave.
- Frame diversity as a strength when seeking investors as a woman-led company.
Why there is no AI without training data
Gonzalez opens with a simple definition: before a machine can learn how to see, hear or speak, it has to be trained with labelled data. A self-driving car, for example, must be taught what a vehicle, a pedestrian, a lane and a road sign look like. She cites an IDC estimate that nearly $100 billion would be invested in AI by 2023, and adds, as she cited, that training data represents 80% of the time consumed in most AI and machine learning projects, with the majority of projects running into problems with data quality and labelling.
Her conclusion is blunt: "So at the end of the day, we can't have artificial intelligence models without training data." (1:27)
Giving work instead of aid
The second half of the story is why Samasource exists. Founded in 2008 by the late Leila Janah, the company starts from a belief Gonzalez states plainly: "And our belief inherently is that talent is distributed equally while opportunity is not." (1:52)
She illustrates it with Ken Kihara, a colleague from Mathare, one of the largest slums in Kenya, where she says youth and women unemployment hovers at 70% or more. Ken left high school with strong grades but had no degree, network or access, and ended up producing chang'a, a local moonshine. Samasource's hiring model deliberately ignores education and job history: the only criterion is living below the World Bank poverty line. New hires are trained in training data and AI, then employed full time with benefits and living wages. For Gonzalez, the result is not only more income but a career path, and a first real entry into the workforce where people build skills and professionalism.
Measuring whether it works
Gonzalez stresses that Samasource measures its impact rather than assuming it. As she cited, workers surveyed three years after leaving earn on average four times their previous earnings. "We have a well over 80% success rate of people earning more than they did at Samasource after they leave or going back to university." (4:22) Retention is high as well, with annualized attrition of roughly 10% or less, which lets teams build deep labelling expertise across hundreds of workflows in autonomous vehicles and augmented and virtual reality.
She brings the numbers to life with Martha, who joined after a long stretch of unemployment that Gonzalez attributes to her gender and where she was from, not her talent. Soon after joining, Martha moved out of the slum into formal housing, and today she is a leader in an online hospitality company in Nairobi. "Just absolutely incredible what the power of work can do in terms of transformation and how technology plays an incredibly important role in making that happen." (11:27)
Quality comes from humans and machines together
Why does labelling quality matter so much? Gonzalez gives an example any engineer can picture: "Imagine a self driving car that can detect a motorcycle, but not a bicycle or can read certain road signs, but not others." (5:18)
Samasource's answer is a vertically integrated workforce working on a machine-learning-assisted annotation platform. "So it allows our humans to do what is uniquely human and the more work we do on the platform, the smarter our systems get as we codify that knowledge of our workers." (5:40) The company covers the full training data lifecycle: identifying gaps in a client's data, pre-processing and transformation, labelling, quality assurance and model validation. She points to Walmart, whose huge and constantly changing product catalogue needs a search algorithm kept fresh so that customers can keep finding products. Delivery centers such as the one in Kampala, Uganda, combine secure facilities with sustainability, and the company has been approved for B Corp certification.
A woman-led mission
Gonzalez pays tribute to founder Leila Janah, who had passed away from a rare, aggressive cancer in the January before the talk. She describes Janah as a pioneering social entrepreneur with a deep belief that business can be a force for social change, and that technology leaders need to lead through action. That conviction shows up in hiring: "We hire 50% women." (8:52) Gonzalez adds that this balance runs from the labelling teams through management and up to the board.
She also shares a personal note on fundraising. In 2019 Samasource raised a $14.8 million Series A, with Gonzalez and Janah as two minority women executives seeking funding for a software company. "We were oftentimes the only women in the room, but it was really, really ultimately something that was greatly beneficial because we ended up finding investors that not only, you know, really loved our product and felt strongly about the promise of AI and had expertise in AI." (9:41) Those investors, she says, also saw the value in their diversity, a reminder that such backers exist. To date, Samasource reports moving over 50,000 people out of poverty, counting more than 11,000 workers and their dependents.
Watch the talk
Changing the World Through AI and Training Data, Wendy Gonzalez, 13 minutes: https://www.womentech.net/video/wendy-gonzalez-changing-world-through-ai-and-training-data
About the speaker
Wendy Gonzalez is President and Interim CEO of Samasource, where she is responsible for the company's overall vision and success, and she brings over 20 years of experience building technology and teams. Before Samasource, she led product and engineering at an IoT SaaS startup, held leadership positions at public companies in IT and spent over 10 years in management consulting.
Questions this talk answers
- What is AI training data?
- Why does the quality of training data matter so much?
- How does Samasource decide whom to hire?
- How does Samasource know its model works?
- What was it like to raise funding as minority women executives?
Drafted with AI from the talk transcript. Editor review pending.