Chatbots are becoming a common part of the online experience, most likely unfolding in the customer support domain, where websites often use chatbots to help with a purchase or address specific issues. Increasingly, organizations are also implementing HR chatbots for online recruitment, helping prospects find jobs that are good matches and assisting with the application process.
But while these two use cases have proven to be early chatbot hot spots, one area that seems primed for chatbot use has been slower to develop: internal HR service delivery.
The idea of chatbots answering the gamut of routine questions HR professionals field on a daily basis represents a holy grail of sorts for HR leaders and chatbot makers alike, but a host of limitations have prevented the market from taking off.
For starters, today's chatbots lack the capabilities required to support internal processes.
"It's like having an employee, a really crummy employee," John Sumser, principal analyst at HRExaminer, said of current HR chatbots. "The kind of employee that only does what you tell it to do, only stops when you tell it to stop, and has no ability to have intuition."
In fact, Sumser tells of an interaction he had in which a chatbot couldn't understand what he was asking, and it wouldn't hand him off to a human being until it actually understood that it wasn't capable of understanding.
"Is that a good experience to give your employees when they're on the job trying to do other things?" he asked rhetorically.
Data the key to realizing HR chatbot benefits
One of the main technological holdups is the inability of chatbots to get at the data needed to answer HR-related questions satisfactorily. Rob May, CEO of Talla, which makes artificial intelligence-powered customer support chatbots, said one big problem is the limited ability of today's natural language interfaces to databases (NLIDB), which are unable to get at dynamic information such as data residing in an external cloud-based system like Salesforce.
"If that technology gets better, it takes natural language instructions and turns them into API queries into a database," said May.
Talla is working on improving AI-enabled chatbots, but May isn't in a big hurry to jump back into the internal HR chatbot market. The company was an early entrant into the arena, but organizations weren't ready. May said engagement levels were typically high immediately after deployment, but would quickly drop off. It wasn't long before a company with 1,000 employees was only answering a couple of dozen questions a week with an HR chatbot, not enough to warrant the investment. As a result, Talla abandoned the effort and re-focused on its customer support sweet spot.
That said, May expects internal HR chatbot use cases to start becoming more common in 2020, and for demand to start rising soon after.
"It's a market that's going to develop a little bit slower," he said. "By 2021, I think there will be a big market there."
Having a good NLIDB solution ready will make Talla's re-entry into the market more compelling. But only a strong commitment and consistent usage will help HR chatbots truly take off. May said that chatbots are doomed when they're isolated. His thinking is this: If employees only see a chatbot in a couple of settings, they don't get habituated to using it. They literally forget that it's there. However, if a company starts using chatbots as often as possible, employees become acclimated to their use, and they start looking for opportunities to use them.
"If everything you do inside of a company involved going to a bot, you'd start going to the bots," said May. "It's like how employees first forgot about intranets."
But there's a Catch-22 in May's perspective, because he also believes that companies currently must look upon HR chatbots as experimental. He suggests starting with a small pilot project, supporting a small group and then learning from that before expanding chatbot use.
Implementing an HR chatbot
In choosing a starting point, May recommends identifying places where chatbots can have the biggest impact. This often means targeting parts of the business that will benefit most from cutting costs. He said the prospect of saving millions on customer support efforts is what drives relatively small companies to use Talla's technology.
Before a pilot is launched, it's critical to ensure that the chatbot works well, and that it's not asked to do too much. Too often, May said, companies roll out a chatbot before it's ready, and the disappointing experience it provides undermines the effort. The chatbot must have access to the needed data so that it can deliver meaningful results, and that data should be as clean and structured as possible.
Lisa Rowan, research vice president, IDC
Data structure is important because a chatbot is only effective if it can answer questions the way users ask them. May said that users will test chatbots by throwing all kinds of complicated queries at them to see what they can or can't do. The more structured the data that's powering the chatbot, the more likely it will bring back useful results.
Lisa Rowan, a research vice president at IDC, said this connection between chatbots and the data behind them is the key to their success. Bottom line: If your HR department doesn't have a handle on its data, a chatbot isn't going to help.
"If you've got HR answering the phone and unable to find the answer, they're not going to be able to put a chatbot in place to find the answer," said Rowan. "A chatbot can't cover up for an underlying failure to find the information."
May also emphasized that companies need to promote their chatbot pilots, providing constant reinforcement that the chatbot is there, ready to use. If employees know HR chatbots are there, and they have good experiences using them, the payoff will come.
"It will lead to really big improvements in productivity," said May.
As of now, however, Sumser said the two types of chatbots he sees in the marketplace still have critical flaws. The first type, which is driven by script automation, is dependent on decision trees that work great so long as there are clearly binary responses. While these chatbots are good at supporting processes that require a very rational juggernaut, the problem, Sumser said, is that "there aren't very many processes that are purely rational flowcharts, particularly in HR."
Meanwhile, chatbots powered by artificial intelligence often come up against the limits of that intelligence. For instance, Sumser noted that Apple's customer support chatbot experienced a bug that confused the use of the pronoun "I" with Apple's exhaustive lineup of "I" products. Throw in the rampant use of colloquial language in HR settings and things can get messy in a hurry.
For that reason, Sumser cautions companies thinking about deploying HR chatbots not to expect that much. Once the technology and data issues are sorted out, he expects the market to take off, but he sees today's HR chatbot efforts as research and development for the industry rather than a successful stake in the ground.
"This will be the way business is done in the future, but it's not ready yet," said Sumser. "Adopting now means you're going to help sort out the problems rather than benefit from market maturity."