AI recruitment tool achieves world’s largest proprietary data asset built off structured interviews
Since 2018, Sapia.ai has been quietly building a unique proprietary data asset, language data based on over 12 million questions captured from 2.5 million users across more than 47 countries. This data set has been the fuel for its proprietary large language models, used in the delivery of smart chat: an AI interview trusted by the most trusted consumer brands in Australia and increasingly around the world to hire fast and fairly.
Smart chat is in a category of its own – hiring automation built on the science of a structured interview delivered via chat. From just a few hundred words, the AI discovers the various qualities that matter in talent management, especially soft skills that play a key role in decisions related to who to hire, move, and promote.
Everyone interviewed receives insights and a coaching tip to reflect on what the AI understood from the conversation. Being blind, untimed and asynchronous with algorithms built entirely off this 1st party data asset of language means it has delivered a near-complete removal of human bias from the candidate screening process and accelerated diversity outcomes. The impact of sapia.ai smart chat to increase applicant diversity is now validated by independent research.
Dr Buddhi Jayatilleke, the Chief Data Scientist of Sapia.ai, said, “When we started on this journey 4 years ago, we had a conviction that natural language had enough signal to truly understand people. After all, the structured interview is well established as a selection method with very high validity, if not the most. We have blended traditional assessment science with NLP and machine learning to disrupt the broken hiring practices, especially at the candidate screening stage.
“This proprietary data asset has fuelled our innovation journey from models that infer HEXACO based personality traits and multiple behavioural competencies to InterviewBERT, a fine-tuned instance of Google’s BERT model for context-sensitive interview response representation. The diversity and richness in our language dataset has allowed us to further increase the validity of the smart interview with innovations built to detect plagiarism, anomalous answers and most recently the ability to detect AI generated interview responses.”
“It is clear that recent innovations in foundational large language models (LLMs) are revolutionising language based AI systems. However the real value from these models can only be gained through extending them with domain specific data sets like structure interview data. Sapia.ai is well placed to bring the ‘last mile’ value from these LLM innovations to the HR stack through our large language data sets.”
Barb Hyman, Founder and CEO of Sapia.ai, said, “We chose to go against what every other company has and is doing. We resoundingly rejected resume data and social media data as a basis for our product. We don’t even use behavioural data or transcribed video data. Because we have been obsessed from the beginning to build AI tech that people trust. Resumés create bias that we cannot escape. ”
Sapiai.ai’s current focus is on selling to enterprise clients who see the largest gain from the transformative nature of the technology. They already work with many of the ASX 100 in Australia and New Zealand, including Qantas Group, Suncorp, and Spark NZ. Their largest customer, Woolworths Group, the largest private employer in Australia, invested in the company via their VC arm having seen the game-changing ROI from using sapia.ai for their volume hiring.
“Years before OpenAI released their GPT models we could see the power of technology which combined the human experience of chat and intelligence enabled by our innovations. Tailwinds are gathering pace for similar tech-chat is empowering, always on, always learning and smart. People are embracing AI for good, AI that gives people back their agency.”