by: Posh’s Business Development and Data Science Intern
What is a chatbot and how does it work?
Most people have used a chatbot before and probably think of it as “that bubble that pops up when I’m purchasing a flight or having trouble processing a return.” What really is a chatbot, though? The fancier answer: conversational artificial intelligence streamlines human interaction with computers. For simplicity’s sake, though, you can really think of chatbots as another Customer Service Rep. Like humans, they understand human input through written or spoken language. Like reps, they’re designed to hold complex conversations about specific topics.
There are many different types of chatbots in the world today, from e-commerce chatbots to customer service chatbots and virtual assistant chatbots (e.g. Siri or Alexa). Whether chatbots use speech or text to communicate, they usually rely on AI technology like machine learning and natural language processing to respond to users.
Interested in buying a chatbot for your company? First off: congrats! A chatbot can really transform a company’s communications both internally and externally. Given that there are so many types of chatbots out there and so many reasons to use them, though, you should first take a step back and ask yourself the following questions before making this key purchase.
What will your chatbot be used for?
Chatbot usage can be difficult to pin down depending on the size and structure of your company. So let’s break it down. You should first ask yourself whether you’ll be using a chatbot for your customers or employees. An internal, employee-facing bot can automate employees’ questions about various HR and IT matters. If you plan to use your bot externally (for customers), consider the product or service that your company provides. Are you in financial services, hospitality, retail, government, etc? Different industries will necessitate different use cases. For example, financial institutions often use chatbots to enable their members to check balances and make transfers, retail companies use their bots to initiate purchases and returns for their customers, and government agencies use chatbots to schedule appointments. A lot of companies further use chatbots for lead generation, marketing, and sales while still others leverage them for educational purposes (e.g. financial literacy). When purchasing a chatbot, prioritize bots that have proven success in your industry and already support the use cases you’re seeking.
How much do you want to spend on a Chatbot?
Now let’s talk about money. How much you want to spend on a chatbot depends on your business goals (see above). If you have a sophisticated tech team, you can create your own chatbot in-house. If you go this route, you’ll have to purchase some additional tools to create your chatbot, hire additional engineers, or sacrifice other projects to create the chatbot.
Alternatively, you can use a 3rd-party company (vendor) to create and implement a chatbot on your behalf. Often these companies have broader product suites so, in addition to a chatbot, you can potentially bundle other products into your purchase such as onboarding software, live chat, or innovative phone technology. If you use a vendor, you’ll have to pay an upfront implementation fee and then an ongoing monthly license. Be wary of chatbots that don’t have an implementation fee or aren’t super pricey; this most likely signals that these bots won’t integrate with your broader systems and aren’t built upon advanced machine learning or natural language processing.
While an “in-house” route may seem more appealing, it is important to consider the technical aspect of this decision. In order to create the best bot for your company, you’ll need to train the bot on your data for it to perform to your needs. This means that by the time you hire engineers, build your bot, and deploy it, it could take close to 6+ months. On the flip side, while having an agency made chatbot may seem more expensive, the time and energy saved by paying a vendor most likely makes it a better option.
Finally, keep in mind when you want your chatbot to show up to customers. Do you want it available 24/7, just during working hours or after hours as well? Sometimes bot availability might affect pricing.
Does the chatbot have industry experience and domain knowledge?
“Garbage In, Garbage Out” is a key phrase in machine learning; it basically means that an AI company is only as good as the data it’s built upon. Since chatbots are trained on such data, it’s crucial to clean this data thoroughly so that the bot can accurately, efficiently, and naturally answer user’s questions. You should really ensure that the chatbot you’re purchasing is trained in your industry’s data. There are a lot of chatbots out there that work across industries; while this industry agnosticism is good for the breadth of their businesses, it’s not great for the depth of their products. If you work in financial services, you want your chatbot to be a finance expert, understanding all the terms and acronyms of the business not to mention the recent regulations and protocols. On the flip side, if you work in hospitality, you want your chatbot to understand the intricacies of that vastly different industry.
Can the chatbot work with existing software?
You’ll also need to consider whether your bot will be “in front of the pin” or “behind the pin”. In front of the pin means that the bot will simply sit on your website, answering basic FAQs about brand hours and ATM locations, for example. “Behind the pin,” as you can guess, means that the bot is integrated with your broader systems and technologies so it can execute actual workflows. An example of a “behind the pin” bot would be a Banking Bot that integrates with a financial institution’s core system and digital banking system, gaining access to members’ Personally Identifiable Information (PII) so that it can greet members by their name, verify their balance, make transfers or put travel freezes on their card, for example.
If you want a more integrated and thus effective chatbot, you should ensure that the chatbot you’re creating in-house or purchasing externally can easily integrate with your other systems. More specifically, you should seek chatbots that have open API-suites, which means that they can integrate with any API-friendly system.
Also, depending on the complexity of your existing software, you should ask prospective clients how long their chatbot will take to integrate with your systems. If you want an “in front of the pin” chatbot, integration may only take a few weeks, but a “behind the pin” chatbot integration can take up to over 12 weeks.
Finally, regarding the technical infrastructure of your prospective chatbot, you should consider if it is omnichannel (i.e. works across web, mobile, SMS, and phone technology) and if it works with cloud, on-premise, and hybrid integration environments. Equally important, you should ensure the chatbot complies with privacy and security standards.
How is the chatbot engineered?
The scalability of a chatbot depends heavily on the way in which it is engineered. If purchasing a bot from an outside source, be sure to ask how the bot is engineered. A chatbot that is engineered with advanced machine learning and natural language processing (NLP) will generate the best user experience for your customers. Natural language processing finds patterns and relationships between computers and human language and gives the machine the ability to understand given input. NLP enables a better understanding of user intent which provides a correct assessment of the information given to the bot. Whereas chatbots engineered on just basic triaging and flowcharts are robotic and frustrating for end-users, bots built upon machine learning and NLP are far more human-like, able to understand linguistic nuances and execute complex tasks.
Can the chatbot be personalized to your company?
Depending on the service you want your chatbot to perform, personalization can be key. In order to personalize your chatbot, look for vendors that have Content Management Systems. Some vendors crowd-source frequently asked questions or desired workflows from industries and enable you to pick and choose which questions and workflows you want to create. You can then brand your chatbot in your colors and with your logo as well as adjust the content to include your company’s name, hyperlinks, and style of speaking. You can further personalize your bot by having it integrate with your customer data so that it greets users by their name and knows their backstory (i.e. their date of birth or recent spending history) before beginning to converse with customers. If you’re instead building your bot in-house, you can create your own content by consulting your sales, marketing, and customer service teams to unearth customer requests and trends to ensure the most personable and comfortable experience possible for users.
Can the chatbot seamlessly handoff to a live agent and display analytics?
While considering which chatbot to buy, you should ask about success metrics, prioritizing chatbots with a high (over 90%) containment rate (the percentage of customer questions that the bot can handle completely alone without involving a human rep). Given that all bots will require at least some human assistance, you should then determine if a chatbot works in tandem with live support. In order to create the best customer experience possible, your business must find a chatbot that allows both parties to support the customer without frustration. As soon as a customer asks to speak with a live agent, the bot needs to be able to connect them to not just any agent, but the correct agent. The bot should also retain enough information from the conversation to direct a customer to an agent in the correct department at the company, contextualizing the conversation up until that point so that the customer doesn’t have to repeat herself.
In order to enact seamless handoff, the chatbot can either integrate with a live chat vendor (e.g. Glia, LivePerson, Oracle) or instead be able to create a ticket. Creating a ticket entails notifying an employee in real-time that a specific customer has a question that the automated bot can’t answer and needs to be called or emailed about this ASAP. This smooth exchange can dramatically enhance the customer’s experience with the bot.
An additional feature you should seek in a chatbot is Analytics; you want a chatbot that comes with an analytics dashboard or email digest so that you can track its performance and your return on investment (ROI).
At this point, you may have already made up your mind about if you want a chatbot and which version you prefer. If you’re still struggling to decide about which chatbot is best for your company, though, consider speaking with your customers and/or employees who will actually be using this bot to get a better idea for what they genuinely want and need. If you already have a chatbot-like system being implemented, ask if this bot is satisfying the needs of the users or if a more innovative bot should be considered. Regardless of what chatbot usage stage your company is at, though, hopefully, this gave you a great framework to think about bots and conversational AI!