What is a chatbot? An easily ignored cheerful pop-up running automated scripts, or an integral part of the automation of menial tasks within companies? How much work can they actually do for your business and how are hard are they to integrate?
For many companies looking to cut support costs, these and more specific questions about the amount of work they can potentially allocate to chatbots is high on the agenda: Gartner cites a 160 percent annual increase in client interest.
Going back to basics, a chatbot, crudely, is a piece of software that can process and simulate human language. To do this they use natural language processing to answer audio or written queries. While chatbots are getting more sophisticated they often typically fall into one of two types, the first of which is the task-oriented bot that excels at structured single purpose objectives such as serving as a website’s FAQs function.
In this instance a chatbot will have access to the site’s database of pre-answered questions, and can select the most appropriate answer when queried.
The other type of chatbot may be the one people associate most with the technology, a machine that can simulate human responses to questions even if it has not been prepared for a particular line of query.
These types of conversational chatbots have been used to create digital assistants and they use a mix of natural language processing and machine learning programming. These data-driven bots can start to predicate a user’s preferences over time and begin to suggest more efficient travel routes or similar data derived insights.
What is a Chatbot in Enterprise Settings?
Duncan Anderson CEO of Virtual Assistant development firm HumaniseAI and former CTO for IBM Watson Europe, told Computer Business Review that what he has seen over the last few years is a a “sort of de-emphasis of the frequently asked questions side of things, and a plumbing of these Chatbot systems into enterprise systems so that the sourcing of answers to questions doesn’t come always from a static corpus of text.”
There was a big shift over the last few years from chatbot systems which accessed static data towards those that can handle more dynamic information, such as financial data associated with bank accounts that can change every second, or a website that displays news or sports results that gets frequently updated. This was a noticeable shift from earlier simplistic implementations of chatbot technology as answer retrieval bots, towards a more integrated enterprise applications approach.
What about integration with your back-end?
Adrian Thompson Founder of The Bot Forge told us that: “There can be challenges integrating into enterprise software stacks.
“First there are so many different types of integration requirements. There is always development effort for each specific integration with an existing system; ERP, CRM, CRM or whatever. Although APIs tend to be pretty good these days and easy to authenticate against. The integration technique tends to be the same; matching an intent with NLU and then calling the appropriate system via a webhook call so gather the required response. Things get complex when you have a number of system integrations which need to be accessed in one chatbot interaction and specific information held in the context of the conversation…”
Many chatbots are not trying to pass a Turing test, but resemble more assistants that suggest and notify employees about tasks or alerts. Anderson notes that: “Your interaction should be very short and sharp. It’s not a conversation.”
Chatbots free up employees to work on other task and in many case small business would not be able to respond to a fraction of their customers without the aid of a chatbot that handles and filters out easy to answer questions or tasks that can be automated. Bank of America, which built its “Erica” chatbot in-house, for example now sees it handle interactions with over 10 million users: there are over 400,000 unique ways that clients can ask Erica financial questions, with more being added.
Experts note that to replicate this success, you need to be crystal clear about the scope project, and clearly understand what functions you want the bot to automate or serve. You should also be prepared for the disruptive way in which customers interact with it once it goes live, as unlike machines we humans can be slightly chaotic.
As Anderson notes: “When you start giving it to testing, people start trying to see if they can break it and they start asking daft things.
“Some of those daft things you might conclude are not actually so daft after all and there things that real people might actually try to ask the system. So you need to be talking about that and you can often find that the scope expands quite late in the project as you try to address how real people actually behave.”
Get it right, and a chatbot with the right scope and function can be a key way to automate tasks or assist your employees with their objectives.