Your complete guide to chatbots in recruitment – part 2

Have you read the previous article about Chatbots? Here’s my continuation from the first bit – with more of a technical focus on the design and functioning of chatbots.

First, know your process

“The question we need to answer first is why should the user be willing to use a chatbot? What’s their benefit in using that kind of tech? What features do we want to integrate and how do they help the user?” says Luc Dudler from jobpal.

Chatbots are far from just universal conversational tools – they rely on pre-defined processes and models. With that in mind, the first step in the chatbot’s implementation is a detailed analysis of the entire recruitment process. What are the different steps and situations? What’s the ideal interaction with the candidate? How do you train a chatbot to be ready for this?

Source: Chatbotsmagazine

“Two different chatbot’s can both answer 80% of incoming questions but the questions are 100% different because the target audiences are different. If you build a recruiting chatbot for business consultancy, it’s going to be a different chatbot then for the hospitality sector,” explains Luc.

If you want to implement a chatbot, be ready to spend a lot of time with the chatbot vendor on analyzing your recruitment process. “What’s also important is the use of small touchpoints that are not so much about technology, but more for the candidate. It’s daring, rather than just automating,” says Luc. Companies must be also ready to adjust existing workflow to better fit the chatbot experience.

Select a supplier that has experience with HR and recruitment. Designing a customer support bot is quite different to a recruitment chatbot. Read this great article about design framework for chatbots.

Chatbot copywriting

After you design logical models, trees and flows, the conversation needs to be designed. Chatbot conversations are back-and-forth interactions of 1–2 line messages within a short timeframe.

It’s gotta be natural, intuitive, and in-line with the tone of voice of your company as well with your target group. And that’s the job for a chatbot copywriter or a conversation designer.

Shitty copywriting can screw everything up, but trying to sound too human isn’t good either. Transparency is ranked as the highest priority for candidates, so trying to pass off a chatbot as a human recruiter—no matter how well-intentioned—can damage the relationship.

Serious user frustration. Source: Chatbotsmagazine

“We’re very upfront that Wendy is a chatbot. When you first meet Wendy and throughout each chat, there’s space for jokes where Wendy can be self-referential. So if someone asks a question that’s outside of Wendy’s knowledge base, she might say something like, “Sorry I didn’t catch that. I’m just a chatbot after all.”

That’s just one example of how voice design can frame the user’s experience. When people have really clear expectations about an AI’s limitations they’re likely to be more empathetic when it doesn’t understand something,” explains Tali Kuhel, Copywriter and Conversation Designer at Wade & Wendy,

Max Armbruster from Talkpush highlights that its customers can personalize the way the chatbot sounds. Another functionality is a hybrid conversation when human recruiters can at any time step in and take over the conversation.

Read more about Designing a Chatbot’s Personality

Communication platforms

A usual question is: “How does the candidate engage with the chatbot for the first time?” There are many options.

Chatbots usually live in a common messaging service, such as Facebook Messenger, Slack, Line or Viber. “Candidates don’t have to download any new software into their already overcrowded phones — a huge plus! — because 99% of them already have at least one messaging app installed. Unlike an app, a chatbot living within Messenger doesn’t burden candidates with a constant stream of required updates, that eat away at their phone’s storage capacity,” explains Lukas Bierfreund, Product Manager @TalkPush.

Another huge plus is that a conversation in Messenger can continue on any device, which is not the case of an app or web-based chat.

Here are most common scenarios on how an interaction can start:

  1. A job post is distributed as sponsored content on Messenger. After clicking on it, the interaction with a chatbot begins.
  2. Candidates find the post in Facebook (either paid, organic or Facebook jobs) and by clicking on “Apply now” the conversation in Messenger begins.
  3. Candidate can be taken to messenger from the career website.
  4. Candidate can get an email (e.g. after applying) with a link to the bot.
  5. Conversation can be engaged from a QR code or a messenger code.

A digital agency in Prague has invited candidates to engage with their career chatbot via a code printed on t-shirts during a career fair.

Communication doesn’t have to be limited to text. Candidates can be invited to record answers through voice, or even video. Actually, voice control in connection with home assistants such as Alexa can bring in a different level of experience when it comes to chatbots.

What’s interesting is that at the end of the conversation with the bot, many people humanize the experience by ending the conversation with “thanks for listening” or “thanks for hearing about my experiences” or “thank you for answering my questions” even if they know it is a bot. Compared to handing them a form, no one is going to thank you at the end of the form, explains Drew Austin from Wade and Wendy.

Integrations

To get the maximum out of chatbots and to deliver a superior candidate experience, chatbots must be well integrated with existing systems, such as Workday, SuccessFactors, Greenhouse, etc.

First, there must be the transfer of an applicant’s data. If the chatbot collected answers to qualifying questions, they should be sent to the ATS, where they are recorded with the candidate and then a recruiter can make the right judgment.

Chatbots should get the applicant’s status from the ATS to launch related actions. After the application takes place, the chatbot is able to inform candidates about their application status and go with them through qualifying questions and pre-screening. So far, that candidate who hasn’t spoken to a human yet receives much better experience, service and a level of responsiveness that no company without a chatbot could deliver.

When it comes to integration, the chatbot must be able to manage its own scheduling system. For example, what would the next step in the interview process be?

Luc explains: “My favorite use case is interview preparation – it doesn’t require technology to do it, but 24 or 48 hours before the interview, the chatbot reaches out to you and says “Hi Alice, I am really looking forward to meeting you in person. Here’s a few things you should prepare for the interview.” And it will share with you the location, who you will meet, and on a 2018 level: the links to the LinkedIn profile of who you will meet, and a few reasons why the company is hiring, etc.

Allowing the chatbot to schedule the meeting in line with the interviewer’s availability is another great use case for both the candidate and recruiter.

Another common integration feature is to offer search and apply functionalities on job boards. For example, TalkPush integrates with Facebook Jobs.

How does a chatbot understand?

Chatbots need a workflow management to know what to do and when. But it also needs a system to understand simply what the user is telling him.

Human language tends to be messy, and the same intention can be expressed in many different ways, not to mention the probability of spelling errors. The first step is to integrate the Natural Language Processing (NLP) which translates the language into something more structured. Here are some of the methods used:

Normalizer – it’s a library that catches common spelling errors, expands contractions and abbreviations, and converts UK English to US English;

Tokenization – breaking sentences into words and removing punctuations;

Named entity recognition – looking for words in predefined categories such as addresses or product names;

Dependency parsing identifies phrases, subjects, and objects.

The next step is Natural Language Understanding to catch the gist and intent in the sentence. For example “Hi there”, “How are you doing?”, and “Good morning” has the intention of greeting and it might trigger a certain action. The simplest way would be to convert the sentence into a vector of numbers. The corresponding intents are also given ‘codes’ to identify them numerically. This input is fed to a training algorithm which learns how to classify these sentences. This classification, aka understanding, is based on statistical models, which indicates what the highest probability is for matching a certain combination of words with a specific intent.

To train the algorithm, the system needs to get feedback from the user. There are several ways of doing it. One is to provide a menu with possible answers or to ask the user whether the answer was helpful or not. These inputs are able to adjust the statistical model in the chatbot’s algorithm.

During the stage of early implementation, you must be ready to invest time into feeding the bot with the right answers, confirm the flow, evaluate feedback from users and make sure you get what you want: greater efficiency and a better candidate experience.

P.S. Wanna know more about chatbots? Interested in how to select a chatbot vendor? Part 3 is coming soon.