Sergey Kravtsov, CEO and co-founder of ConnectiveOne, a startup that creates an omnichannel AI platform for communication, spoke about the results of testing Ivanna, an artificial intelligence assistant for recruiters. The developers of the AI recruiter sought to avoid dehumanizing her, but instead they had to fight the «uncanny valley» effect.
Sergey Kravtsov, CEO and co-founder of ConnectiveOne, a startup that creates an omnichannel AI platform for communication, spoke about the results of testing Ivanna, an artificial intelligence assistant for recruiters. The developers of the AI recruiter sought to avoid dehumanizing her, but instead they had to fight the «uncanny valley» effect.
According to Kravtsov, the idea to create its own AI recruiter appeared spontaneously after a wave of responses to one of the open vacancies. Due to the specific conditions (hybrid work format and mandatory office days in Kyiv), the company needed to quickly conduct a preliminary selection and find out basic things: whether the candidate had read the conditions, whether he was ready to discuss the details, and whether he could work in the office.
«All this could be entrusted to a recruiter — really, nothing complicated. But this is the time for correspondence according to scripts. And we are an AI company — we believe in automation and really love when AI takes over the routine. The task for @ConnectIvanna was quite simple: contact the candidate who sent a resume in the messenger → find out the willingness to talk → clarify the key conditions → pass it on», — explained the CEO of ConnectiveOne.
To communicate with Ivanna, we decided to choose Telegram — it allowed the AI recruiter to accept incoming requests and write to her first.
«Initially, we made the profile as close to a human as possible, but with a slight hint that it was not a person after all. The photo was generated by a neural network, a description was written, and even the date of birth was indicated. This was necessary to avoid dehumanization — the effect when a person sees that there is a bot in front of them and begins to purposefully break it: 'hey, stupid bot, give me your questions',» notes Serhiy Kravtsov.
However, in practice, the developers encountered the opposite — the «uncanny valley effect», when the object looks almost realistic, but something about it causes people to be repelled. During test contacts, some respondents suspected a scam and doubted that Ivanna was a real person. «This is definitely not a scam? You don’t look like a living person.»
ConnectiveOne responded: they replaced the photo with the company logo, and clearly indicated in the profile that Ivanna was an AI assistant. The developers conducted about 8 dialogues from her profile and based on them, they formed scenarios with the dialogue structure, processing logic, vacancy knowledge base, and instructions for the agent.
«A particularly important story is the co-pilot mode. A person monitors each dialogue. At any moment, he or she can pick up the conversation without the candidate noticing. In essence, it’s like an autopilot with a live recruiter at the helm, ready to take over at the right moment. This allows you to maintain quality and control without compromising automation,» notes the head of ConnectiveOne.
According to Kravtsov, with the help of Ivanna, more than 20 «full-fledged screenings in the machine» have already been conducted.
«Only two candidates asked if it was a live person. Both times we simply intercepted the dialogue and explained: these are the systems you will have to work with if you join us,» says the startup’s CEO.
According to him, the entire technical part of the project was implemented on the ConnectiveOne platform, using the Low-code builder and the FastlinePro system.
Earlier, Radhika Sharma, a 35-year-old Product Manager from New Delhi, India, shared her experience of interviewing an AI recruiter. She said it was «exciting and nerve-wracking,» and the AI was more attentive and meticulous than she had imagined.
«To my surprise, the AI assessed my technical knowledge, as well as my engagement, eye contact, facial expressions, posture, and clothing. I found my assessment to be remarkably accurate, but I underestimated how AI could assess me, so I didn’t think to wear a collared shirt,» the woman noted.
