Dumb chatbots aren’t such a dumb idea…
There is now a multitude of chatbots available to the recruitment community. However, as a community, we do not seem to be over the ‘Natural Language’ honeymoon. There is a desire for a recruitment chatbot to have full-blown conversations with the user (read: candidate), needing the candidate to type every single answer. At most conferences and exhibitions, you will hear the inevitable sales pitch by a chatbot provider — “our chatbot is an AI platform that uses natural language processing and machine learning to provide truly engaging conversations to your candidates. Other chatbots that don’t employ AI are just dumb and another version of filling in a form.” Are they really that much of a dumb idea though?
When selecting a chatbot supplier, or even designing a conversation of your own, it is imperative that you think about the user experience — not the perception of what makes you look good. There is an en vogue ‘fashion’ around implementing Machine Learning and Natural Language Processing (or as some would call — AI) purely for the sake of implementing Machine Learning and Natural Language Processing. I would say this — you don’t need AI to have a good conversation.
When selecting or designing a chatbot user experience, consider every button presented to the candidate as a choice. Therefore, when you consider a chatbot for your recruitment process, look at the reasoning behind one that gives the user a button vs. one that requires a user to type words. Consider the fact that the keyboard is a big panel of buttons.
Therefore, what you are really doing is deciding whether to ask the candidate to pick from a small number of buttons or a great many buttons (pressed in a particular sequence) which almost always is asking substantially more of the user.
If you do choose the latter, please think carefully about making sure that it’s really worth the extra effort by the candidate. There are interface elements such as buttons and date pickers that provide a more positive user experience.
In one example of a chatbot implementation, no matter how much time was taken to ‘train’ the chatbot and perfect the language interface, users were dropping away from the desired conversation flow (at an alarming rate) — which usually ended with candidates getting frustrated and commands being typed such as “I want to speak to a HUMAN” (note the capitals) and “is there a real person I can speak to”.
So we went back to the drawing board — did we need this ‘sexy’ technology to provide a solution to a problem in recruitment? Ultimately, we wanted to implement conversational computing and automation to help recruiters to (re)engage with their candidates and transform their CRM/ATS into an active database or constantly updated candidate records that could be resourced from — en masse and at speed.
Did we need to embark on a conversation with a candidate that required them to type out every answer (even yes/no answers)? No, but we could provide option buttons to select in the majority of cases.
Did we need to employ natural language processing and machine learning, which requires months of implementation and ‘training’?
To highlight this, current performance stats for a non-NLP chatbot implementation are:
Average conversation completion rate (engagement rate): 88%
Average conversation completion time: 80 seconds
Speed of response from conversation invite: 10 minutes (40%), first hour (60%), first 24 hours (97%)
Recruiter time saved: 43–110 minutes per day
Not such ‘dumb’ results…
As a recruiter and a technologist, I urge you to think about the candidate experience rather than implementing AI for the sake of implementing AI.