A chatbot, or bot for short, is an automated, Conversational User Interface (CUI) to any kind of service on the Internet. As opposed to a Graphical User Interface (GUI) such as a website, a mobile app, or a kiosk terminal, a chatbot exposes information or leads the user through a transaction via text-based chat, typically conducting a conversation with the user using natural language, buttons to advance the dialog, or a combination thereof.

Chatbots are typically embedded in messaging apps such as Messenger, Slack, or WeChat, but could also be offered on a website via web chat, within a mobile app, or over the carrier channels SMS and USSD.

Chatbots, bots, and "Interactive Text Response" (ITR), a nod to the well-known customer care technology of Interactive Voice Response (IVR), are used synonymously in this document.

The answers in this FAQ are geared towards chatbots for the purpose of serving and supporting customers of a business.

Both Interactive Voice Response and Interactive Text Response share the following value propositions:

• Cost reduction in the contact center through automation (“self-service”)
• Improved Customer Experience
  • 24/7/365 access to information
  • Faster access to data, no "on hold" for basic information

While IVR is meant to deliver on the above through the voice channel, ITR accomplishes the same on messaging/texting channels. In addition, ITR applications allow certain use cases that are not feasible or often not commercially viable to deliver over voice, such as collecting alphanumeric data as needed in name/address change applications, or password reset applications. On text, "what you type is what you get."

Virtual Assistants for customer service and support have primarily become known as implementations on Websites, often with the use of avatars or “talking heads.” These systems are meant to guide users around the content of the Website. They are typically implemented to help with frequently asked questions, by pointing users to a page on the Website that has the answer, or letting the user conduct transactions there, e.g. via Web forms. Virtual assistants often work by providing a list of FAQs the user can choose from, that are similar to the one the user typed in.

Chatbots are meant to be more conversational in nature, not only giving answers to questions and sending users to other resources, but also transacting business through a conversation, e.g. looking up the status of an order, making reservations, purchases, or giving out account updates.

Download rates of mobile native apps have gone down in recent years, and many companies realize that apps for customer service don’t see the adoption they had hoped for. Despite the ease of installation and use, we witness the growing phenomenon of app fatigue.

Similar to the mobile Web, chatbots don’t require the download of an app from the business first; other than mobile Web, though, the experience through a bot is more light-weight, and the conversational paradigm opens up new and more personal(ized) ways for consumer engagement.

For the reasons stated above, adding a conversational experience within a mobile app might not yield the business outcomes desired. Using previously unused or rarely used channels, such as SMS or Facebook Messenger, are more promising to yield the engagement rates that make a business case.

While messaging apps and pure texting channels share their message-based nature, SMS conversations can only be purely textual in nature and additionally have a character length limitation. Web chat is often implemented the same way.

Messaging apps like Facebook Messenger or kik additionally allow for rich media to be shared within a conversation thread, either in a dedicated way through the use of embedded images or carousels of text/image/buttons, or through “web views” that allow the display of web content within a message bubble. This allows for richer interactions, without overwhelming the user with a full-blown Web experience.

Note though that SMS allows the use of disposable apps, which are context-sensitive mobile Web apps that are invoked through a short URL embedded into a message, as in this example:

A handover from a chatbot to a live agent is needed for the following situations:

1. When the customer asks to be connected to a human
2. When the application logic of the self-service script requires human intervention (e.g. for compliance reasons, or to complete a transaction that requires a person)
3. When the script cannot determine the intent or meaning of the customer’s message (after at least 1 or 2 attempts)

Natural Language Processing (NLP) is a discipline in the field of Artificial Intelligence. It covers all aspects of handling "natural languages" – such as English or Mandarin – with a computer, vs. artificial languages such as programming languages (Java, C, …)

Natural Language Understanding (NLU) is a sub discipline of NLP, with the focus on programming a computer to interpret natural language so it can act upon it (e.g. to then translate it to another language, or execute commands, or simply converse with the user). The opposite is Natural Language Generation (NLG), which deals with programming a computer to generate language, e.g. when giving answers to a user, or creating texts automatically based on structured information (think software that writes news articles based on a database of information).

NLU and NLG (or some form of response generation) are needed to build fully conversational chatbot applications, as the user expectation sometimes is as high as assuming they can converse with the computer the same way they can converse with another human being. Some therefore argue that "chatbot" is a misleading term for many of today's implementations that rely on buttons to advance the dialog.

In order to properly answer or comment on a statement or a question by the user, the intent of a message, all entities mentioned (such as things, people, situations), and optionally the sentiment of the message need to be extracted and analyzed. Various techniques can be applied to achieve NLU, such as

• a neural net trained on hundreds of thousands of similar messages that are manually tagged to teach the computer their meaning
• a strictly rules-based approach that uses linguistic knowledge such as part-of-speech and other lexical and syntactical information about words and phrases of a language
• a hybrid solution

Keyword or simply word spotting is a simplified version of the rules-based approach where the computer simply checks if key words that carry meaning can be found in a message, and if so, classifies the entire message accordingly. While this approach can create first results very quickly, it can also produce misclassifications easily. As an example, consider the linguistic nuance that makes up the difference between "Can I read an electronic book on my flight?" Vs "Can I book a flight?", which each contain key words, but require completely different answers. Only by knowing that "flight" is a direct object of the verb "book" can you tell this message apart from the first message, where "book" is used as a noun and is thus in a different relationship with "flight."

Artificial Intelligence is a multi-faceted term. Wikipedia writes: "Colloquially, the term 'artificial intelligence' is applied when a machine mimics ‘cognitive' functions that humans associate with other human minds, such as ‘learning' and 'problem solving’.” Following this definition, chatbots do fall under the AI umbrella, as they mimic human conversations using natural language.

While a lot of software technology recently seems to be getting the “AI” label, the academic and industry community is trying to distinguish this narrower or “applied AI” from “artificial general intelligence”, aka “strong AI”, which means a machine that can perform any intellectual (vs. mechanical) task a human can perform, including reasoning, problem solving, even creativity.

See also http://blogs.aspect.com/what-is-artificial-intelligence/ for some more depth on this topic.

In theory, you can distinguish two types of “dead end” events:

1. The chatbot cannot map the user message to any message/question the bot knows (when using a machine learning approach), or the chatbot cannot find any of the keywords/linguistic structures it was programmed to find (when using an approach based on word-spotting, see also What is keyword spotting vs. natural language processing vs. natural language understanding?)
2. The chatbot successfully recognizes the message/question, but doesn’t have any answer programmed for it

In practice, however, only the first event usually happens, since if the programming effort has been made to understand the question, the answer or response behavior is easy to add (it usually takes way more effort to interpret a message than to respond to it). IF it happens that the answer isn’t known or simply cannot be given without human intervention (e.g. if the company’s business rules or compliance reasons require a human check before conducting a transaction), or the chat medium is not well-suited for the answer (e.g. the answer takes much more space than fits into message bubbles, or an image/table simply “says more than a 1000 words”), then you have 2 choices:

1. Hand over to another self-service medium (e.g. Website) that is already well-established and can handle certain inquiries better
2. Hand over to a human (see also How do you handover a self-service chat to a customer service agent?)

Option 2 requires to have a form of “contact center” in place, whether that’s an established function in your organization or just a group of employees that can help out when needed. Modern cloud- and Web-based contact center solutions require minimal installation and learning efforts.

If you do not yet offer service or support on a messaging channel, consider starting with SMS, which is still the one channel reaching the largest audience – anyone can text, even users of old feature phones. The advantage of using a hitherto unused channel is that you won’t be interfering with any other customer care programs in your organization. You will have less procurement and approval hassle internally, and will be quicker at “getting your feet wet” with this new technology.

However, make sure that SMS is actually an attractive and affordable channel in the region you want to deploy this in. Different countries have different market saturation and pricing models for SMS.

If you already offer service & support on a messaging channel, e.g. web chat, start there.

Begin by automating the most mundane tasks your tier 1 agents handle today: authentication and pre-qualification. For authentication, you need backend integration, so if you want to get something out there quickly, you could skip that and just implement pre-qualification, i.e. determining why the customer is messaging you, so that you line up the right resource to help them.

Consider advertising to your customer base that they can now get to your agents quicker and with less hassle, simply by texting “call me”:

This will let you accomplish 4 things:

1. Reduce cost by bypassing your IVR, thus avoiding operational IVR cost (vendor technology, telephony)
2. Reduce cost by turning an inbound call to your toll-free number to a cheaper outbound call (check how telephony cost differs for inbound/outbound in your region)
3. Improve the customer experience, as your customer now no longer has to wait on hold to speak to someone
4. Improve the agent experience, as your agents now no longer have to repeatedly ask the same basic questions over and over again throughout the day

From there, you can later add “real” self-service; e.g. try to automate the top 3 conversations your customers are having with agents today, and only hand the ones over to humans that are different, require human problem solving intelligence, or required human emotional intelligence (e.g. to calm an angry customer down.)

By going this route, you are playing it safe: you are setting the right expectations with your customers (e.g. that the service cannot yet solve their questions, but can make the path to the agent faster and less burdensome.)

Alternatively, you can start right away with automating one or more of your most frequently asked questions, which will lead to savings quicker.

In general, chatbots can live on any channel that can carry written dialogs and that expose APIs so that automated applications can be deployed. These include:

• Carrier channels
  • SMS
  • USSD
  • MMS

• Non-carrier channels
  • Facebook Messenger
  • (WhatsApp does not yet officially provide an API, as of February 2017)
  • WeChat
  • Twitter (public timeline or direct message dialogs)
  • Kik
  • LINE
  • Viber
  • Slack
  • Skype for Business
  • …

• Embedded chat
  • On websites
  • Within mobile apps
  • On kiosk terminals (typically also Web-based)
  • …

With Aspect CXP 15.1, Aspect supports SMS, USSD, FB Messenger, Twitter, and embedded chat out of the box. Any other channel offering an API can be integrated through a Professional Services engagement.

Great question! We would love to share more with you. If you’re interested in learning more about what we think about this question, please contact us.

Please read What kind of skills are needed to build chatbots? first.

When working with Aspect, expect to be asked to provide access to business analysts that understand your business and can assist with scoping the solution and the entire discovery process. Furthermore, an IT resource will be needed to help with backend integration, opening of firewalls, etc.

Great question! We would love to share more with you. If you’re interested in learning more about what we think about this question, please contact us.

Great question! We would love to share more with you. If you’re interested in learning more about what we think about this question, please contact us.

Project timelines depend on the complexity of the bot, i.e. predominantly the number of questions it is supposed to answer and the complexity of the backend it needs to integrate with. Simple chatbot projects can take 8 weeks from project discovery to deployment; complex bots can take 12-16 weeks or longer.

That depends on both the platform you chose for bot development and deployment, and the “transport” channel (the messaging channel) acting as the carrier of the messages. Aspect CXP can scale to handle several thousand concurrent dialogs (with no limit to the total number of messages), while you initially probably only need to handle a few dozens initially. If you choose a cloud platform, concurrency shouldn’t matter to you.

SMS as transport channel has bandwidth restrictions set forth by the carriers the application will be deployed on. Furthermore, there are also limits as to how many messages you can send outbound at once.

This depends on the platform you chose for bot development and deployment. Aspect offers deployments in all 3 models.

Great question! We would love to share more with you. If you’re interested in learning more about what we think about this question, please contact us.

Security is a function of trust and effort. There is no 100% security when it comes to mediated communication, yet every effort should be made to achieve a level of security that aligns with the risk your business is willing to take in order to offer great customer service. This question cannot be answered in “absolute terms”, but needs to be explored together with your compliance and risk assessment group.

It is common practice not to share PHI (protected health information) or certain financial information such as credit card numbers in an email, but rather point to a securable Web resource (like a disposable app, i.e. a mobile Web app) that can implement payload and/or transport encryption. The same is advisable for the chat medium, and thus chatbots.

Data security standards such as PCI-DSS or HIPAA define rules to avoid security issues, and these practices should be applied for chatbots just like for mobile apps, your phone channel, or your Website. Aspect offers PCI- and HIPAA-compliance in the cloud, and can help make your premise environment compliant as well.

Great question! We would love to share more with you. If you’re interested in learning more about what we think about this question, please contact us.

The Messenger bot should have the same ID as the Facebook page of the business. You can find out the ID by clicking/tapping on the bot’s name at the top, then select “Open Facebook profile”. Have a look at the part in the URL coming after facebook.com.

Scan the Facebook page for previous customer communication. If the wall of the page is empty, or the page has very few followers despite a well-known brand claiming ownership, you have a right to be suspicious.

If you are not sure, or the company has deployed a bot on a dedicated ID, ask the company on a trusted channel first.

The effort to maintain a bot should be comparable to similar self-service efforts in your larger CX organization, whether that’s maintaining your IVR, your Web FAQ, or the content behind your mobile app and website.

• As your business evolves with product launches or changes of your business model, new questions need to be added to the arsenal of questions your bot answers
• As your business rules change with new legislation or changes to your business model, the answers and/or the dialog flow of the bot need to be adjusted
• Language is not static – every year, hundreds or thousands of new words enter a language, either for new things (“smartphone”) or concepts (“Brexit”). While new synonyms or changes in the nuances of the meaning of words should be maintained by the vendor of your bot and NLU platform, new words that have relevance for your business need to be added to the dictionary of the bot and might require new handling

As with your agents in the contact center, it comes down to training, training, training. Shortly after go-live, read every single chat transcript the bot has had with your customers and identify areas for improvement. This can and should also be a service of the vendor you chose for your bot.

Aspect offers services to not only build and deploy a bot, but also analyze and tune the system over time.

The following types of changes need to be made on an ongoing basis:

• Changing existing answers to questions
• If this is about tweaking an answer in terms of the words used, these are easy fixes and should be possible by your business analysts, without requiring change requests through your vendor. If it is about adding more words to an answer, make sure that the message still fits into your chosen messaging platform’s message restrictions, if it has any
• If this requires a change of logic, i.e. a change of follow-up questions, then app development expertise is needed. (See also What kind of skills are needed to build chatbots?)
• Adding new questions with their corresponding answers/handling
• When using a rules-based approach (see What is keyword spotting vs. natural language processing vs. natural language understanding?), these changes require coding new NLU rules and thus require specialists (see also What kind of skills are needed to build chatbots?). If the new question is similar to existing questions the bot handles and only differ in certain linguistic nuances, you easily run the risk of impacting interpretation accuracy of existing questions when adding a new one without due care. Adding a new question requires an analysis of all existing questions to avoid overlaps and handle linguistic nuances accordingly
• When using a machine learning approach, these changes still require manual effort to find enough samples per question to train the system with. Furthermore, care needs to be taken retesting what has been added, so that there won’t be misunderstandings if the new question is very similar to existing questions.
• Improving the understanding accuracy for an existing question
• If you find gaps in the understanding of a message that you as a human understand just fine but the bot either misclassifies or doesn’t understand at all, the NLU rules need to be adjusted. This requires an NLU specialist.
• With a machine learning approach, new data needs to be collected to re-train the model

This depends on the availability of the messaging platforms in the desired target countries, as well as the platform you chose for bot development and deployment. Aspect can offer chatbots in all regions of the world in a premise deployment model. When using a cloud or hosted model, availability depends on the cloud zones Aspect offers Aspect CXP in. For details, please contact us at www.aspect.com, or reach out to your sales representative.

That depends on both the platform you chose for bot development and deployment, and the messaging channel acting as the carrier of the messages.

Aspect CXP supports all languages of the world. To leverage Natural Language Understanding capabilities, Aspect NLU currently supports 14 languages. For all other languages, chatbots can be built using keyword spotting techniques.

The costs to design and develop a chatbot are comparable to the cost and efforts for a speech-based IVR self-service system. From the work of a business analyst to determine the scope and domain of the bot, to the designer creating the dialog flows and crafting each and every bot response, to the developer implementing the design and backend integration, to the deployment and QA efforts – prepare for app development costs that can lie anywhere between USD 20k and USD 150k+, highly depending on the number of questions it can answer, and the complexity of the backend and/or contact center integration. Add to that the cost for the platform chosen and potential per-message costs stemming from the selected carrier channel. SMS incurs transport costs, while most if not all messaging platforms (OTT, or “over-the-top” channels such as Messenger or Twitter), do not charge per message.

Any vendor selling you just a platform and maybe services for developing a bot should be checked for follow-up costs. Prepare for ongoing tuning needs that might incur costs for change requests. Make sure that you understand the tuning, change request, and maintenance costs before signing up a vendor, whether that’s in the form of an annual maintenance fee, or a bucket of hours that can be used post go-live.

Furthermore, ask your vendor if they charge platform fees for adding a new messaging channel (say you start with SMS but later want to add FB Messenger and Twitter), or a new language (besides the obvious services cost to implement that new language; see also How are multiple languages handled?)

Aspect discusses tuning cost up-front, and does not charge extra platform fees for adding a new messaging channel (as long as it’s one that is supported today) or language later, apart from the app dev cost these additions incur.