With today’s intelligent tools, the smaller the problem, the better the answer. To the extent that you can define, describe and contain the thing you want to apply “intelligence” to, the more likely it is a solution that is feasible.
This is particularly true with chatbots. They work best when tackling small, simple problems. There are two primary flavors of chatbot: scripted and unscripted. Here are a few ways to tell them apart.
Scripted bots ask questions and record responses from a set list.
Scripted bots walk users through a decision tree and are effective ways to collect data for precisely defined situations. College degree? Check. Driver’s license? Check. Late-model car? Check. Willing to relocate? Check. Seventy-five percent travel? Check. Stand on head for hours? No check. Sorry, you don’t qualify. Having a machine work through the drudgery of qualifying 500 applicants on these knockout questions is a godsend.
Unscripted bots take user questions and offer answers from a data set.
Unscripted bots attempt to get answers for users from an archive of company information. They are like a search engine designed to provide answers to employee questions. When applied to HR, they tend to be tools that guide benefits choices, answer questions about policies, help employees use their benefits, provide intake for complaints or navigate administrative processes.
The knowledge-acquisition section of the bot consumes company policies, procedures, FAQs and other documentation to create a pool of indexed, searchable data. The translation function interprets user queries (in text or sometimes voice). The matching operation pairs the two to produce an answer.
Unscripted bots can be trained to deliver the right result approximately 80 percent of the time. If the company data are complete (and that’s often a big, expensive assumption), an unscripted bot can understand the question and deliver the correct answer four times out of five.
When bots work well and when they don’t.
If the problem is simple or just a little complicated (such as determining who oversees sexual-harassment reports), answers are easy. So if what you want is a rigorous and unforgiving application of policy to specific transactions, chatbots are a logical solution. HR certainly has its share of procedural areas for which exacting precision is a necessity.
Chatbots (and all current intelligent tools) work best when the project involves a question that can be clarified with a structured set of subordinate questions. But things get more difficult when the topic is complicated and become impossible when it is complex or chaotic.
It’s worth elaborating on these categories.
- Simple circumstances are ones like filling out a standard form and transferring the data from one system to another, tasks that are easy to describe and easy to repeat. Chatbots, intelligent forecasting and robotic process automation all excel in this area.
- Complicated circumstances are ones like transferring an employee to a new location. Sometimes we can break them into smaller simple problems, but not always. Success usually requires coordination with different people, multiple teams and context-specific expertise. Things often go wrong and timing and coordination can cause problems. Some of these problems are also amenable to automation.
- Complex circumstances are ones like developing a cadre of managers. Once you learn how to transfer an employee, you can repeat the process with other employees and perfect it. One transfer is like another transfer, but that’s not the case with developing a layer of management. Although developing one cohort may provide experience, it does not guarantee success with the next. Expertise is valuable but not sufficient. The next group may require an entirely different approach from the previous one, which points to another feature of complex problems–their outcomes remain highly uncertain. Yet we all know that it is possible to develop a team of managers well even if the process is complex. Complex problems also involve rules that change based on circumstances, making it often hard to know when the change that matters will take effect–and rendering AI difficult to use here.
- Chaotic circumstances are ones that defy predictability. Many of the core processes of our organizations are chaotic. They are driven by the operation’s response to emergent phenomenon like key personnel shifts, reorganizations, market changes, regulatory surprises, missed financial targets, continuous imprvement or new disruptive competitors. Most of the organization’s processes are driven by difficult to predict factors, which are not favorable for automation projects.
There are two keys to successfully implementing AI projects: picking areas where the tools will work and having a larger vision of the fully automated organization. Among the latest tools out there, chatbots are a great example of the strengths and limitations of contemporary intelligent tools.
I’m already beginning my preparations for the HR Technology Conference next fall in Las Vegas. As a part of my research, I’m anxious to speak with practitioners who are implementing intelligent tools in their organizations.