If you are looking to deploy a chatbot for support and have started to investigate the market for chatbot providers, you've probably been left with your head spinning trying to understand how they are different. Chatbot companies have different ways for building bots (how much is it out of the box ready, vs designed by you?) and different places they can deploy (website, facebook, slack, etc). This post is going to look quickly at the AI, or, in some cases lack of AI, that these bots have. In general you can divide them into one of 3 camps.
1. Scripted chatbots with regular expression matching.
This is where the vast majority of support chatbots are today. You use some kind of script builder to build query-response bots. So, when someone says "How do I change my billing address?" your bot responds "You can change your billing address here" and inserts a link. The question has to be a regular expression match, meaning it must be reasonably close to the question the person typed to the bot.
The benefit of scripted chatbots is that they are simple to understand, and usually quick to build and deploy. The downside of scripted chatbots is that they are brittle - when you type in something that isn't a very good match, they break. They can break on relatively easy questions by missing synonyms, for example. The other issue is, if you have a ton of possible questions, it can take a really long time to build out the flow. Finally, if you have a lot of scripted questions and answers, maintaining them can be difficult. Bad maintenance will equal bad answers and unhappy customers, so be careful.
In general, scripted chatbots are great for quick small experiments, but you should probably move on to someone more complicated if you want to deploy a bot at scale.
2. Vector matching Q&A bots.
The second most common type of chatbot for support is what I label a "vector matching" Q&A bot. These bots use NLP technologies like word vectors to better match questions so that the bot isn't so brittle. Word vectors can match questions that don't have the same exact words. They can figure out that "how do I update my credit card information" and "where do I change my billing information" are the same thing. These bots can match various questions, with various paraphrases of those questions, to an answer.
The benefits of vector matching bots are that they solve the brittleness of scripted chatbots, so perform much better for end users. The downside is that you still have to keep tabs on the question and answer matches, and keep information up to date. Most of the forward looking companies building bots are using this technology.
3. Ensemble workflow bots
The last category is what I call "ensemble workflow bots." These bots do two things. First, they use a whole bunch of different NLP and Machine Learning technologies together, so that they can have the best chance at getting the answer for a customer. For example, they may combine a machine comprehension technology (which actually reads in your Support documentation and understands it) with word vector technology, with a default to keyword search. This ensemble of methods improves performance dramatically.
Secondly, they are more than a standalone point solution - they are integrated into a workflow. So, for example, they might open tickets for questions that have no easy resolution, or may catalog knowledge gaps for questions that have no support documentation or answers yet. They have methods to interact with support reps to receive ongoing training, so the bot can get better.
The beautiful thing about ensemble workflow bots is that they don't really take much more time to setup than other chatbots. Because they are embedded in workflows, they can learn quickly from those workflows rather than being part of a big data import and onboarding project.
Support chatbots are an important part of any CX strategy. They can be an anchor part of your CX automation toolkit, which makes your customer facing organization stronger by automating away monotonous tasks and focusing on higher level, more strategic, more meaningful engagement with your customers.
If you are thinking about a support chatbot, Talla falls into that third category. We would love to talk to you and show you what we can do for your organization.