In 2019, chatbots (or AI Assistants) have now passed through the “peak of inflated expectations” and are starting to become truly productive for business. Given that there are now many effective options then we need a framework to help decide on the most appropriate solutions.

Types of chatbots

“Chatbots” come in two main types – those that are an adjunct to human customer service and those that are fully automated. This is characterised as a choice between “Live Chat” and what are generally know as Chatbots (or AI Assistants). We’ll use chatbots to mean either and Live Chat to mean only Live Chat and hopefully that will make sense in context.

The correct choice of one or the other is a crucial first step in implementing a chatbot project. but it is not quite a simple as just choosing between the two.

Live Chat Projects

Live Chat projects can vary from the simple – just connecting agents to a website visitor – to the complex. At the complex end, the Live Chat system might incorporate AI support, as for an AI Assistant, and integrate into a client’s CRM, email marketing system AND their customer support case management system.

Without detailed planning and testing, that kind of complex application of Live Chat can backfire on both customers and employees who find themselves on the receiving end of a garbled thread of interactions. That’s not a project to take lightly.

For that reason, in real life, most of the Live Chat implementations are simply connections to agents with a backup of email support. Live Chat offers human intelligence to prospects, but the caveat, here, is that this strategy only works when live chat agents are available. Unless a company opts to staff their support channels 24/7, there’s typically a period of downtime for every site.

Another disadvantage with Live Chat is the analytics insights derived from the very unstructured conversations are limited. Although it could be done by continuous natural language processing in association with the website meta-data this is not usually the case.

On the other hand, AI Assistants are available 24X7 – no downtime – and because of the structured conversation clearer analytics and goal conversation rates are available.

Chatbot Projects

The vital first step with an AI Assistant chatbot project is to understand clearly the main use case you wish to solve.

Lisa Bouari, CEO of Outthought, classifies 4 main classes of applications:

1. Frequently Asked Questions

2. Triage Assistants

3. Lead Generation

4. E-commerce

Being clear about the objectives is important, as each of the categories above requires a different set of core features and integrations in order to be successful. For example, you will typically ask an FAQ bot more questions than it asks you, but a Triage bot will ask you more questions than you ask it. Therefore an effective FAQ bot can be built on relatively simple keyword recognition and searching, whereas a Triage bot needs access to natural language processing and a smart decision tree.

Obviously, a Lead Generation bot needs to be integrated into an email marketing system and an E-commerce bot into a payment system and logistics system.

In 2019, it is still very important not to be too ambitious with a chatbot project. While they might become mainstream in 2020+ there are still challenges today. We are in year 2–3 of a 15–20 year period in the growth of chatbot technology.

Don’t overestimate your Chabot’s common sense

The biggest challenge today is that chatbots are relatively easy to prototype, but hard to get right. Developing a conversation flow is quite an art and depends on the underlying technology being able to support a “human-like” sequence. This is exacerbated by the fact that people’s expectations are ahead of the readily available technology, because of what they see in the movies and also because of their experiences with the likes of Siri. Siri is no bolt-on $30,000 project!!

Even something as simple as handling follow-on questions is handled poorly by most chatbots, as the context is not retained. If you try Siri or Alexa you can explore the extent to which they are managing to retain context during a flow of questions – and even with those huge $$$hundreds-of-millions investments, it is still limited.

For humans common sense is simple but for chatbots it is still the unsolved problem.

That means today, in 2019, the most successful chatbot projects will have clear objectives, be focused on clear payoffs for the users, and not try to be too many things to too many people.

Done properly, as a well-designed project, bots do yield good payoffs. For example, it is reported that chatbots have delivered a 15% to 35% lift in total order value for orders on eCommerce sites when the chatbot is built correctly.

For example, Tennis Australia released a bot that guided users through the different types of tickets available for the 2019 Australian Open, then linked them directly through to Ticketek’s payment page to complete the purchase. In its first week, the bot delivered 170 per cent more conversions than linking users directly to Ticketek, at a lower cost conversion rate. More than 600 conversations took place in that week within the bot, generating 141 conversions and 25 times the return on investment.

And for, say, lead generation applications, a well-designed chatbot can achieve high rates of lead capture. Chatbots do a great job in capturing small and medium-size business leads, and this is a very viable application in 2019.

One of the advantages of chatbots versus Live Chat for lead capture is that they eliminate that dreaded wait time when you are waiting for a Live Chat agent to come online. The drop-off rate while in the queue is high, and those are missed opportunities.

Understanding levels of Chatbot projects

To summarise the state of chatbot projects in 2019, consider these three categories of projects:

1. Type 1: A clear and relatively simple goal with a relatively simple form of structured data as the base material, and the potential of a structured flow into and out of a simple decision tree, and probably stand-alone. This suits FAQ and Lead Generation projects. This is a typical “Manychat” type of application. Typically, this would be a $3,000 project.

2. Type 2: A clear goal using a broader range of structured data than Type 1 as the base material, with more complex branching and responses. This requires that a history of enquiries is available to inform the conversation development process, and that all branches of the decision process have useful data to support that stage of the enquiry or a contextually-informed set of options to keep the “conversation” flowing. Type 2 will also have more complex integrations into an existing ecosystem. This suits Triage E-commerce Assistants.This requires a bespoke project using a good commercial agent platform, or could a narrower solution focused, for example, on Social Customer Support using the chatbot system integrated into a larger platform like Sprinklr. Typically, this would be a $30,000 project.

3. Type 3: These projects also require a clear goal and a clear payoff for consumers, but will get there through a more complex analysis of data and logic – this requires a serious strategic approach and a collaborative project, something like the bots released by some of the major banks. They use structured and unstructured data, and the application of Natural Language Processing, plus contextual retention, and a high degree of integration with core enterprise systems.Typically these would be build using a bespoke chatbot platform rather than a chatbot application i.e. a robust AI / Machine Learning-based chatbot framework which is powered by various industry-leading engines e.g. IBM Watson. Typically, this would be a $300,000 project.

Underlying all these three levels of projects – modest, ambitious, and complex – is analytics. You should always aim to get the best interaction and goal analytics available for any solution you choose. The progressive enhancement of your solution will depend on these analytics, and your understanding of customer satisfaction will depend on the insights from the analytics. Without excellent analytics your chatbot investments, at any level, are trying to grasp smoke.

Conclusion

Properly designed chatbots are yielding real benefits right now for businesses which have approached their development the right way. In the future, according to Lisa Bouari, organisations will have several AI Assistants with different roles and capabilities that are all aware of each other and where to route users based on which assistant will serve them best. That’s an exciting opportunity for brands that can get it right.

Contact Us for social media audits, strategy, and for augmentation of your customer experiences enhanced by social data.