As consumers, we’ve come to expect a lot from our machines: they are basically everywhere, and they are basically smart. Conversational UI is the way we can interact with appliances where a mouse, a keyboard, or even a touchpad is too clunky.

When it comes to recruiting and talent acquisition, messaging is quicker, easier, and more personal for applicants. Automating parts of the process through chatbots can also save recruiting departments a great deal of time and money.

But if you’ve ever messaged a chatbot, you know the feeling: there are many seemingly simple things it cannot do, and it’s unclear what it can do. Faced with a chatbot, many end users both underestimate and overestimate its capability - paradoxically, at the same time.

What does it mean for a chatbot - a conversational UI with a smart engine behind it - to be usable? There are two design and psychology concepts that are especially useful here, introduced by Don Norman in the classic interaction design book “Design of Everyday Things” (1986): the gulf of execution, and the gulf of evaluation.

Have you ever opened a chatbot and hesitated, thinking, “what should I type?” That feeling reflects the gulf of execution. You have some idea of what you want to achieve, but you are not sure which action to take. In the 2017 Designing Interactive Systems (DIS) journal, Klopfenstein et al describe this need for guidance:  “because of the free-form nature of [chat], it is easy for bot users to get lost and [not know] what commands“ can be used.

One way to provide guidance and support to the user through quick, clickable replies, as in the screenshot below from our careers chatbot

At this point, this is a very common approach to make chatbots more usable. But guidance can be a general guiding principle. Klopfenstein et al  also suggest as a chatbot design principle:

“Each single message ... should contain the full context of the conversation and should embody what a single UI screen represents for mobile apps. ... Each message has an atomic meaning and stands on its own.”

In addition to providing job search functionality, a major part of our offering is automated FAQs. A well-designed bot flow typically has a few dozen responses, but one of our FAQs bots might have a couple hundred distinct answers. It is the job of the underlying system to determine, based on the question, which answer to give. But what makes an answer good is how well it can bridge the gulf of evaluation.

Suppose the happy path for your chatbot includes the following exchange:

• Prospective job applicant: Can I upload my CV here?
  
• Non-atomic answer: Yes, you can do that after you select an available position and answer a few screening questions.

In practice, though, there are many ways to ask that question, and this answer doesn’t always make sense:

• Prospective job applicant: How can I submit a resume?
  
• Non-atomic answer: Yes, you can do that after you select an available position and answer a few screening questions.
  

The alternative is to use stand-alone, atomic answers when the goal is to provide the user with information. In this example, a better option would be “To apply, select an available position, answer a few screening questions, and upload your CV.” You can try out a demo recruitment FAQ bot here!

The Actions on Google design guidelines offer an overview of techniques for graceful error recovery. The guidelines focus on speech (rather than typed text) input. The aim is to provide the right amount of guidance. For example, repeating the available options for answers, but not ad nauseam. In this use case, though, the bot is asking the questions, and the user is giving the answers. This technology supports making purchases, reservations, and so on. There are relatively few types of questions the bot asks, but it is necessary to ask them in different ways, and to include additional, flexible information.

In the talent acquisition space, however, a major way to help save time is to automate answering frequently asked questions. Because the aim of this kind of bot is to improve processes of existing teams of people who are responsible for responding to questions, it is important to provide control, transparency, and maintainability to those teams. In this case, designing a chatbot that guides the user means clarity and consistency.

Over the last two years we have been able to work with great teams and companies to test, iterate, improve and validate our chatbot technology for many recruiting and HR use cases. Read more about our research activities!

Norman, D.. (1986). The design of everyday things.

Klopfenstein, L. C., Delpriori, S., Malatini, S., & Bogliolo, A. (2017, June). The rise of bots: A survey of conversational interfaces, patterns, and paradigms. In Proceedings of the 2017 Conference on Designing Interactive Systems (pp. 555-565). ACM.