Here’s The Main Reason Why Most AI Projects Fail

How’s your AI project coming along? If training data challenges are getting in the way of your goals, it may be because your training data provider is not stepping back and looking at the Big Picture. In short: your provider may not be owning the full-training data lifecycle.

Granted, it’s a complex, fine line that takes skill and experience to walk: how do you scale a data annotation project without compromising on data annotation quality?

“If you are going to have an annotation platform, you’ll need one with robust annotation capabilities,” says Sveta Kostinsky, director of sales engineering at Samasource. “This allows you to address a variety of annotation requirements and edge cases. For a platform like this to be successful, it has to be fueled by machine-learning algorithms, increasing annotation efficiency and the ability to reduce annotation time. Quality checks have to be automated and embedded into the platform to ensure high quality annotation achieved from the very beginning of the project launch.”

Easier said than done, but Samasource has been providing high-quality annotated training data to 25 percent of Fortune 50 companies since its founding in 2008. It also has the distinction of being named the first Bcorp-certified AI company, meeting the highest standards of verified social and environmental performance, public transparency, and legal accountability to balance profit and purpose.

One other non-negotiable must-have for AI project success: the human element.

“Your platform must set up the Humans in the Loop for success,” Sveta says, referring to the globally trusted provider of ethical data annotation and model validation services. Goals include avoiding bias through better dataset collection and annotation, assuring that AI models work with the same accuracy everywhere. She says, “This means providing an instant feedback loop to agents, and allowing the project to quickly adapt to changes.”

To make this happen, Sveta and her team provide each client with a project manager (yes, an actual human being) who reviews and considers the best annotation approaches, recommending the best one suited specifically to your project.

“We believe in quality transparency,” Sveta says. “Prior to your project launch, we will design with you a quality rubric and ensure that we deliver on that quality service-level agreement [SLA]. Our experienced agents and full-time employees allow us to easily ramp even the most complex of projects, and be agile when we introduce instruction changes for ongoing projects.”

If your AI project does not adopt this strategy at the very beginning, is all lost?

“We find that clients come to us in very different stages of AI adoption,” Sveta says. “Some clients come when they have a mature data pipeline, others when they are still designing the project. We discuss different aspects of the projects starting from assembling a non-biased data set and reviewing and recommending different annotation methodologies and approaches. It’s about designing a quality rubric accounting for edge cases to ensure they receive high quality annotated data from the very first delivery we make.”

One other must-have in any successful AI project: ironclad security. Samasource implements security best practices on a physical level; for example, background checks on all employees, and IT/infrastructure and technological security through encryption devices and ISO certification.

“We also offer anonymization technology blurring and altering appearance to visual PII data,” Sveta adds. “Overall, we take a multi-layered security approach.”

Using these time-tested strategies, new Samasource projects include innovating solutions for Orbisk, a company aiming to reduce food waste through AI, and helping Orthogrid provide AI-driven solutions to digitally transform orthopedic surgery.

“It’s really wonderful to see how computer vision and natural language processing [NLP] technologies are adopted by and revolutionizing new industries,” Sveta says.

Assistance from Samasource can often change the course and fortunes of a company venturing into AI projects. In some instances, its work can even help and improve our environment.

For instance, in 2019, the UN reported that nature is declining globally at rates unprecedented in human history.  This devastating impact on global biodiversity affects nearly one million animal and plant species threatened by extinction.

Vulcan, a company dedicated to improving ocean health, combating climate change and supporting conservation, asked Samasource for help with strengthening its AI efforts.

“Our expert annotators applied our technology to advance AI algorithms, powering one of the leading wildlife conservation efforts,” Sveta says.

Through its work with Vulcan, the Samasource team was able to address the current threat of wildlife trafficking while collecting data that will inform long-term strategies to protect endangered species and ensure stable and thriving generations.

As a result, Vulcan has significantly improved its turnaround times to process training data, allowing algorithms to thrive without compromising on quality.

To date, Samasource has labeled nearly one million images for Vulcan, achieving above industry-standard quality SLA of 95 percent in support of their efforts. Together, Samasource and Vulcan are combating large-scale poaching that contributes to extinction, while supporting a more sustainable ecosystem through image processing solutions.

Another result of this work: Samasource has won the 2020 AI Breakthrough Award for Best Image Processing Solution.

This clearly would have thrilled Samasource’s founder, the late Leila Janah, an activist who firmly believed that providing meaningful, dignified, living wage work was the best way to permanently lift people out of poverty. She founded a movement — Give Work (“give work, not aid”) — to bring her vision to life, which has since been named after her to honor her legacy. The foundation supports entrepreneurs in Kenya and Uganda and provides funding and mentorship to founders whose business models further the Give Work movement.


Bottom line:

To assure the success of your AI project, find a training-data partner that can help you:

  • Use Machine Learning powered annotation tools that enable rapid labeling.
  • iterate on instructions
  • optimize your workflow configuration.
  • quickly adapt to ramp-ups, focus shifts, and edge cases.
  • master API-centric architecture that is built to scale using extensible stateless microservices, ensuring scalability and development velocity.
  • bring delivery sites that are ISO certified and employ biometric security.
  • deliver on throughput within schedule. Turnaround time should be as low as two days from your new workflow request to production delivery.
  • achieve anything from implementing a robust quality rubric to raising edge cases.

Find out more about our content partner, Samasource here.

Recent Posts

View All Posts