Chatbots in HR

sgt-star-screenshot
This is a review of current literature on the application of chatbots to HR. It attempts to explore critical questions that HR leaders may ask when evaluating investments in HR chatbots. This is still a work-in-progress, and this blog post will be updated from time to time.

In what situations could chatbots be used in HR?

  • Carry out simple administrative tasks, freeing up HR’s time for more value added work
  • Customer service, where employees are the customers
  • Carry out repetitive or routine tasks / questions from employees, where answers can be automated, and reduce frustration for HR teams  (Bhaduri, 2016)
  • Provide 24/7 availability to answer employee queries
  • Improve speed of response to HR queries
  • Training  (Bhaduri, 2016) where individual (not group) learning is the focus

Aspects of HR that chatbots can be applied:

  1. Information service
  2. Recruitment
    1. To shortlist candidates (Bhaduri, 2016)
    2. Q&A facility for candidates (Sullivan, 2016). However, this may not be relevant for the communications agency environment, as much of the employment information on the company is typically considered confidential, due to competitive poaching, and would not be published on public forums. The highly customised nature of compensation and benefits in the industry also rule out the use of chatbots in this aspect as chatbots can only provide standardized responses
  3. Onboarding (Boulton, 2016) – use of a chatbot to contact new employees with onboarding information and links to the required HR forms, in place of a HR generalist
  4. Self-service registration / employee information update (Ames, 2016)
    • Benefits enrollment
    • Editing personal or family information

Considerations for implementing chatbots in HR:

  1. Deploy in current tools of communications that employees already use – Integrate into existing platforms of communication, rather than force employees to adopt another tool (Ames, 2016)
  2. Security of information – due to the nature of the function, exchanges with employees will involve confidential and/or personal data. Is the environment where the bot is hosted secure? Can it be integrated into an existing HR software pre-approved (security-wise) by the company? (Ames, 2016; Boulton, 2016)
  3. Data collection and privacy – what kind of data will the chatbot collect, and how does it impact privacy of employees, particularly when they may ask sensitive questions, or questions that may influence others’ perception of them? Will the questions asked be able to be directly traced to individual employees? (Maass, 2014a)

Some examples of organisations developing HR Chatbots

  • U.S. Army: Sgt. Star, commissioned by the US Army from NextIT, to help with recruitment efforts. Status: Launched since 2007. (Maass, 2014bNextit.com, 2016)
  • Intel: Ivy, developed by NextIt, is a virtual HR agent that uses a combination of natural language processing, artificial intelligence and optimized search hosted on their Intranet to help answer Intel employees’ questions about their pay, stock, benefits, or other HR programs (Ask Ivy, 2013, Boese, 2013)
  • Talla: Employee onboarding. Status: to be launched Oct 2016 (Talla, 2016; Knight, 2016)
  • ADP: HR task automation, such as sending a job to a prospective hire to alerting employees to use their accrued vacation time. Status: not available commercially yet (Boulton, 2016)
  • Aspect: Mila replaces the employee hotline for sick employees at overstock.com and allows change of work schedules. Status: available internally since July 2016 (Greenfield, 2016; Hobson, 2016Aspect.com, 2016)
  • RAMCO: HR chatbot to respond to queries and helping in tasks like checking leave status and calendar before applying for leave. (Sangani, 2016)
  • WIPRO: using chatbots to judge the mood of the employee. (Bhaduri, 2016); currently being trialled
  • FirstJob, Inc: Mya, the chatbot, vets and interviews job candidates as well as administering tests, providing application status updates and tips to candidates. Status: in private beta (Deutscher, 2016Trymya.io, 2016Dishman, 2016)
Advertisements

Now, how do I design a bot?

After my small success yesterday, I started thinking about what I need to do next to further my understanding of building a chatbot.

Of course, what I managed to do yesterday just skims the surface and there are a lot more chatbot functionalities that I need to learn to be able to build a protoype. However, what struck me, when I was writing the bot, was that I was learning how to execute, but I hadn’t learnt how to design one.

What is the logic that drives the design of a bot? Even if a bot had excellent functionality, it would still be useless (or not used) if the user experience is bad.

So, this will be an additional branch of reading that I will be looking to explore. Some initial resources that I’ve found:

  1. Twine
  2. Why does your chatbot suck?
  3. The chatbot design playbook
  4. When bots go bad: Common UX mistakes in chatbot design
  5. Principles of bot design
  6. Designing a Chatbot | UX Design Process | Case Study
  7. Chatbots made easy

I built a bit of bot!

I want to order screenshot

After a frustrating afternoon, I finally made a personal breakthrough.

I built a bit of bot!

I tried building a chatbot from scratch on my own using DIY blog posts as guides, but it was too hard because I don’t have programming knowledge, and those instructions start with setting up a server, when I was raring to have a go at building the bot itself.

So after hours of trying today, I decided to take a different approach and take a shot at using a chatbot toolkit. I chose to use wit.ai as it was recommended by a friend, and also the fact that Facebook bought it.

Following instructions from the wit.ai starter guide, I managed to write a simple yes-no response. On top of that, I also “trained” the bot to respond to variations of yes/no, such as “yeah”, “yup”, “nope”; similar to how a person might respond over WhatsApp or FB Messenger. This is the most basic of basics, but looking at where I started just a few weeks ago, with my fears and concerns, this is really a leap forward.

So here it is, a screencast of my efforts today: