Creating Conversational Experiences (I): Planning
This is the first of a series of 3articles where we’ll cover every step anybody needs to go through in order to successfully create a Conversational Experience, from planning to deployment and everything in between.
In this first article, we’ll go deeper into the planning phase by following a logical sequence of decisions we need to make before registering an account on any eye-catching AI-based platform.
Do I need a chatbot?
The first thing we need to ask ourselves is: ‘Do I really need a chatbot?’. Being probably the buzzword of this lustrum, everybody is talking about — and with — chatbots. You’ll be widely considered anchored to the past if you don’t have at least one chatbot with its respective out-loud press-release. But if you want to inform your visitors about your office hours— and nothing else — a line of text might in many ways be more helpful and efficient than a chatbot. 😉
Engagement and efficiency are the magic words you are looking for when deciding whether you need a chatbot or not. We understand engagement as having something that makes users want to come back, share personal details and spread the word, and efficiency is often related to process automation and cost-effective customer communication — from giving information to providing support.
Of course, as we’ll see below in the Objective section, your curiosity and the unarguable trend chatbots are starring might be reason enough to go and create one.
1. Objective
The first thing you want to do when planning the creation of a Conversational Experience is the goal that your chatbot will pursuit. If you don’t have a clear objective when it comes to building a Conversational Experience, maybe you should think twice.
These objectives come from both sides of the table: user and creator perspective. Here are some examples of goals from both points of view:
Creator’s goals
- Give support: the #1 goal for companies, probably because it’s directly related to cost. They all dream of a future with machines taking care of the dirty work — aka repetitive queries from users — where they now spend thousands of hours per year. Depending on the level of support you plan to provide and the combination of bot/human resources available, you can create from a bot that answers FAQs — some are actually specialized in this, like Microsoft’s QnA — to one that handles complex queries and that requires chatbots and humans to be perfectly aligned, with an accurate takeover.
- Announce or promote an upcoming product or service: presenting a product ‘like talking to a friend’ seems to be an efficient way of doing it, taking advantage from the conversation
- Build an audience: related to the previous 2 points, Facebook Messenger bots are usually created to build an audience you can later push with news or updates on your product or service.
- Get leads (emails, phone numbers, Facebook IDs, …): one of the most common goals: conversation increases engagement and therefore powers-up data collection, as we can read this great article by Jiaqi Pan.
- Sell (product advice/information, qualification): by solving and asking questions from users you are able to tell whether there’s a real sales opportunity or just some visitor prying.
- Innovate/be cool 😎 Let’s be honest here: most of the chatbots built by companies now are whether experiments with the technology trying to prove it adds value to a certain process or a way for them to flex out their techy muscles in front of the crowd. And that’s perfectly fine!
Also, your goals can be linked to specific KPIs — which is something we always recommend — so you can better measure their impact and make more accurate decisions. For instance, you may want to increase your NPS by a 6%, improve your visitor to lead conversion rate to an 8,7% or be able to automatically — cost-free — handle the 20% of your Contact Center tasks.
Moreover, objectives are also related to the temporality of the chatbot: is it going to be a 1-week promotion bot for a movie release between the fans of your page or a permanent tier-1 support-bot for your customers?
User’s goals
Are very related to creator’s, and here are some examples:
- Get information about a product or service.
- Get fast help/support, on demand.
- Entertainment, have fun.
- Experience something new.
All you have to do is match these two goals within the chatbot, which will define its features and scope.
2. Functionality
What will the chatbot be able to do? What it won’t? What kind of value will add to your audience? Depending on the goal you’ve set — from both perspectives — the bot will have one or another type of functionality and features.
In this section, we’ll cover three transversal functionalities that will help you determine which technology you need to implement, namely: natural language input, information source, and availability and human handover.
Natural Language Input
— “Will I be able to speak to it?”
It defines the different types of input users can enter to communicate with the bot and obtain information — or get things done.
In words of Golden Krishna, ‘the best interface is no interface’, and that’s essentially the conversation. But there are still different types of them depending on if we use text, voice, images or even movements to communicate.
There is a super-interesting concept introduced by Bill Buxton called “place-ona”: a persona whose location shapes/limits the type of interaction that makes sense for him/her. The most common example is the “driving placeona” that is hand and eyes busy, but ears and voice free. In this particular scenario, a voice control device seems to be the most logical alternative. Does this apply to a nightclub? The answer is probably not!
Getting to know on which scenario people will use your bot is critical in order to decide which type of input should we go for. 90% will stick to text, but Alexa’s timer is still the most used feature on Bezos’ gadget, and the kitchen the preferred location for it!
Information source and availability
— “Will it provide real-time information?”
Knowing where information is and how is it going to flow, is a critical part when planning a chatbot. In other words, you first define what is your chatbot going to know about and how much is going to know about these topics. Is it going to cover simple FAQs, first in-app steps or deep technical issues? Whatever you decide, you need to set user expectations from the very beginning, being transparent about chatbot’s capabilities to avoid frustration and abandonment — the number 1 enemy of our logical friends now.
Then, you define whether the information will be there as a “preset” of the chatbot triggered by keywords or buttons from the user, or if it’s going to consult that information somewhere else depending on user input — what we name “callback”.
In the first case, we have static information: you need to manually add all that data and, whenever you want to make a change, go back to edit and change the original content. It’s tricky but fast and easy to have up and running.
On the other hand, we’ve got dynamic information. You are going to retrieve info from external sources, like a CRM, a database or a website, and you need to integrate your bot with these sources, usually through their API. This way, you’re able to provide dynamic input that doesn’t depend on presets and that’s virtually unlimited. You can probably add a higher value to your users with this type of information, but the scope and therefore complexity will also be more prominent.
Human handover
— “Will there be any human around?”
Knowing if your bot is going to team-play along with some humans it’s important as it’ll determine which technology, tools, and channels you should use. Some examples where human takeover makes sense:
- Complex Support. Many companies want chatbots to take care of their FAQs while having humans to cover the most difficult tasks. In this scenario, the human takeover plays a critical part of the process, so you need an appropriate tool including an interface for human intervention.
- Lead Scoring. One of the great abilities of chatbots is acting as receptionists: they can explain the basic stuff to your visitors, ask them certain questions and, whenever they identify an opportunity worth it enough to be handled by a human, perform an automatic handover.
So go and analyze if your chatbot is going to cover some of the above points and find out which tool and channel fit you best! Some examples of tools that allow you to handover conversations are Facebook Messenger, Operator from Intercom, Chatfuel, or Landbot.io.
3. Technological Approach
Probably the hottest topic when it comes to building a chatbot has to do with whether we want it to be rail-based or able to process Natural Language — NLP.
For some people there’s no such dichotomy since usually chatbots are a combination of both — and they are right — but I believe this distinction has more to do with the type of technology you need to back one or another. IMHO, the key question is “will the chatbot be able to infer stuff by itself or will just deliver outputs as the result of a predefined combination of inputs from the user?”.
Rail-based interactions are those where the user needs to choose between a certain amount of preset options. So, we’re basically asking them “Do you want information about 1 or 2?” and they’ll answer “1” or “2”, being delivered to one or another piece of information depending on their selection. Most of the times, if they can enter any input — and not just click on one option — and they type “3”, the system won’t be able to identify it and will ask to try again.
Or, it could have some basic keywords to cover the most common situations (when someone insults the chatbot, when the number is not 1 or 2, etc). So we’re watching a combination of keywords and options, which makes this approach a hybrid between rails and Natural Language Processing — obviously closer to a train than to a T-300, but more than good for 80% of the use cases with chatbots.
Natural Language Processing — the application of computational techniques to the analysis and synthesis of natural language and speech — is helpful if you’re building enterprise-level complex chatbots that aim to last for years. But the reality is that, nowadays, the technology is not there yet. This means that, even with the hottest technology stack available, you won’t make anybody believe they’re talking to an actual human. I’m sure that in some years time technology will be there, though, it’s just a matter of time.
You can go for any of them as long as you pay close attention to the most important thing when it comes to building a chatbot: expectations. If you don’t try to convince users there’s a human speaking when there’s not and don’t over promise with what the chatbot is able to do, you’ll be good to go. Oh, and having a different answer for every insult you may receive as a chatbot is a clever move too: accept you’ll be fooled!
In this line, the biggest NLP platforms are Google’s API.ai, Facebook’s Wit.ai, IBM’s Watson and Microsoft’s LUIS. Depending on what your top priorities are you can go for one or another.
First time bowling and feeling good with rails? Then you might want to stick to powerful yet accessible platforms like Chatfuel (one of the biggest names in the industry, code-free Messenger and Telegram chatbots builder), Motion.ai (flowchart based, code can be added for further complexity and fully trainable NLP) or Landbot.io, (based on web, fully-customizable Conversational Interfaces with no coding and a drag-and-drop builder).
4. Audience
Just like if you were designing a website, depending on your audience your bot will have a totally different look and feel when it comes to things like tone of voice, interface design or message length. Moreover, is not only the people but where and how is those people going to interact with your creation that will determine its full design.
So first you need to define your audience and their context to then adapt your chatbot to it. Use sociodemographic variables for audience definition and pose virtual scenarios to analyze context.
One of the most important aspects of your chatbot that will vary a lot depending on your audience and its context is the tone of voice. Harriet Cummings, from Distilled, deeply analyzes brand’s tone of voice in this great article where she clarifies that ‘it’s not what you say, but how you say it. This encompasses not only the words you choose, but their order, rhythm, and pace.’ You wouldn’t expect a mortgage chatbot or a robo-advisor to be super-joker or a sports results bot to talk to you like a 90-year-old ancient kingdom’s spokesman would do. Adjust the tone of voice to your audience, functionality, and context.
And yes, this is closely related to channels. For instance, if you find out that your audience’s avg age is 24, Messenger might be the right channel to deploy. But, SMS could make more sense if we’re talking people in their 50s. Same with the country: LATAM is a lot into WhatsApp while Korea loves local Kakao Talk. So this audience analysis will help you on our next step: channels.
Regarding context, talking about personal finance or health might be easier on a more anonymous, web-based environment. Also, if your chatbot is about real-time traffic information, allowing voice input — remember the concept “placeona” — would make much sense!
As obvious as it gets, keep in mind logic things like keeping messages short if you expect people to talk to your chatbot on mobile or avoiding media sending if the bandwidth of your target region is far from optimal.
5. Channel
As we introduced in the previous point, the channel will determine critical things like how much you can play with the interface, what kind of audience you’ll find and which interaction timing should you expect.
You’ll also need to bring forward things like if it will be accessible to everyone or where is the interaction going to happen (laptop, tablet, smartphone)?
We have different categories though that can help us classify these channels according to their nature, and match our purpose and needs.
Instant Messaging
Messenger, WhatsApp, WeChat, Telegram… Instant Messaging is eating the world — already surpassed Social Media in terms of active users, I just won’t share Business Insider’s chart again — and those are great channels if you want to reach billions of people in a mobile environment.
It varies a lot from channel to channel. Messenger and Telegram, for instance, are wide open when it comes to the chatbot developing community, while giants like WhatsApp are still very restricted — although this is going to change.
Oh, and yes — SMS and iMessage are instant messaging channels too so take advantage of how widely extended they are!
Web channels
Web channels — from bots that live in webchats to full-screen web-based conversational interfaces like Product Hunt’s Upcoming Pages or Landbot.io — are very accessible to anyone without the need of an app and offer a wider range of possibilities in terms of interface customization.
While you need to stick to a certain amount of rules when developing a bot for Kik or QQ, web channels open a world of possibilities in terms of Conversational Elements personalization allowing to shorten many development processes — why develop a full Google Calendar integration on Messenger when you can iFrame it on a chatbot’s message from the web?
Of course, you might find some issues regarding user verification: without the right IDs they are just visitors and you’ll be unable to know if the same person is coming from both mobile and desktop without the right process.
Voice, movement based
From Amazon’s Alexa, to Microsoft’s Cortana or Apple’s Siri, voice-based bots — especially smart speakers — are gaining more and more prominence open a new world of possibilities in terms of interaction. The counterpart is that they need a specific device to work so their potential volume of users is not even comparable to other channels. Some people might call these chatbots voicebots.
Xbox Kinect or PlayStation Move are also some examples of channels/technology not based neither on text or on voice but, in this case, movement. With VR — Oculus Rift, HTC Vive, PlayStation VR — we’re discovering new possibilities too in terms of hybrid channels. And well, ordering a pizza with some kung-fu moves might not be the most common thing — but give it some time!
Every good thing comes to an end — although temporary — and this is it! Thanks for reading and feel free to check our next article Build and Design: Creating the perfect chatbot flow.
Don’t miss any article of this series:
Don’t forget to share some love ❤️ — and some claps 👏 — if you liked the article and let me know if there’s something you miss or feel like planning the creation of a Conversational Experience should be done any other way, would love to hear it!
