Size of the Prize
There are multiple Questions to ask your Chatbot Supplier. The first is there are multiple platforms to build upon. Two of the biggest are Facebook Messenger and Slack. But there are many choices.
- It’s the most engaging channel with average 80% opening rates and 4 to 10 times higher click-through rates than email. Your customers are highly likely to be already using Messenger every day and your business could be part of that conversation.
- Facebook Messenger has 1.3 Billion Active users per month
- Two billion messages exchanged between people and business per month
- It has 11,000 active Facebook Messenger Bots. (The Verge)
- Voice and audio calling feature are being used by 300 million people actively. (Forbes)
- Facebook Bots now accepts significant vendors like Visa, MasterCard, and American Express.
- 53% People say they are more likely to buy from a company to whom they can send messages daily.
- Slack had gone up to achieving 6 million daily active users in 2017. (Statista)
- It owns approximately 2 million paid accounts.
- According to BotList 484, Slack Bots are currently available. (BotList)
1. What Kind of Bot do you want to build? Part One
More questions to ask your chatbot supplier will be around the initial foundation of your bot. Flow chatbots function via tree-based decision trees. The chatbot’s developer defines paths that the user is driven down. Not too dissimilar to building a customer journey map of your website.
Do you remember those books that allowed you to choose the ending and or decide which page you turned to next? Or even early adventure games like Leisure Suit Larry or Monkey Island, there was only ever a few set options to solve the puzzles.
In a Flow chatbot, it’s similar. The user can decide, but cannot change the flow from the set path(s) you will make available. Flow-based chatbots generally will serve up set buttons or critical search words and will limit or not allow the user to type anything. Just like the books, these are hyperfocused bots, and it’s unlikely for unpredictable events to take place.
Artificial intelligence bots
Artificial intelligence drives the experience. Artificially intelligent robots will provide a lot more freedom to the person. It’s more like a conversation as it would be with a human. Users can type and use text, get answers and follow up questions. What us humans call a discussion. Flow-based chatbots do not enable this. Then even within AI, there is more than one type.
One-way AI – “Level 1”: A bot that remembers
In unilateral one-way AI chatbots, the user to chatbot interaction uses artificial intelligence on only one side of the engagement. The goal of the AI is to understand what the user is saying – that’s all and nothing else. The AI is there to interpret the user’s input. Once it has associated an intent to that data, all it does it search for the answer from a database of replies preloaded by you.
Two-way AI – “Level 2”: A bot that learns
Two-way bots use artificial intelligence to drive the information back to a receptive user. Its no longer matching an intent and obtaining an answer from a pool of pre-written responses; two-way bots can build and return solutions on the fly – in a sense. Most two-way AI chatbots work by connecting to a database of available information that they can use to learn. These extra-smart AI chatbots can compute user input, understand the intent, then construct the most accurate answer before delivering it to the user. These chatbots can learn over time (trial and error, better database, better information to work with, etc.).
Hybrid – “Level 3”: A bot that understands you
Hybrid bots are the most well-known type of chatbots you will end up using or comparing against. Google Assistant is an example of this type. As the name suggests, hybrid bots take the best of both to deliver a better experience. These bots will still drive the user down a particular pre-defined path but will also allow free typing and interactions. Use of pre-defined triage can be enabled with apps that fully enable free text and communications.
Most chatbots use some flow interaction at some point, to improve UX and goal completion. While the AI is intelligent, we still can’t entirely leave it to its own devices. Imagine what it could say!
2. What kind of Bot do you want to build Part Two
a. E-commerce focused bots
E-commerce chatbots facilitate transactions. These bots are used to make buying items online more engaging and more personal. A bit like an assistant in a real-world store but less wooden 🙂 [jokes!]. If you are searching for a flight, rather than showing you a list of flights in a static form, the commerce bots will allow you to look at options and trips dynamically. It will “listen” [read] to your suggestions and change the flights it delivers to you based on your replies.
b. Notification bots
It does what says under the heading. These bots keep you updated and notified through your day and to inform you of any alarms or reminders you may have set.
c. Brand and Business bots
Brand based bots represent already established brands of which there are many examples. Brands use them for anything from brand awareness to allowing full transactional actions like Skyscanner.
d. FAQ and customer service bots
We have noticed the most enquiries coming through about this kind of bots. The operational costs of running a chatbot are considerably cheaper than outsourcing support agents either locally or in countries where salaries are a lot lower. Chatbots are available 24/7 and are more affordable to train and manage than humans. Bonus, they can interact with millions of customers at any given time. No human could do the same task. And would require massive capital costs and operational costs if you did want millions of human agents.
e. Fun bots
Some bots are just for fun and don’t have any real purpose but to provide entertainment. It may fit in with your brand and website.
This question about core functionality is probably more for yourself rather than questions to ask your chatbot supplier.
Language / Dialect questions to ask your chatbot supplier
Ultimately bots are built on the principle of language. AI, Machine learning, Natural language processing are words synonymous with the rise of chatbots. Chatbots are essentially a mix of keyword/phrase matching, NLP, and machine learning. Below are some important language and dialect points you need to consider.
3. Brand chatbots need to take into account domain and related vocabulary to understand what a consumer is saying along with inferring the intent. In case of Insurance for example words like car insurance or buildings insurance etc. How does your chatbot cope with this? Limitations that may arise need to be thought of?
4. How will your chatbot handle dialects? Will a person from Newcastle pose a question in the same way as someone from Liverpool? Sometimes yes and sometimes not! A consumer can ask a question or make a statement in different ways. Dialect, typos, inability to spell or even not having ones glasses on could affect the input. Put yourself in your client’s shoes and try simulating as many conversation scenarios as possible. Design a test with different users before you go live. Don’t use friends and family to test. Try usertesting.com or similar.
5. Every machine learning product needs training data. How does your chatbot have access to this training data? A customer service bot, for example, could have access to your customer service team’s knowledge base and chat logs. How will the platform improve over time once it gains access to more diverse conversations?
6. This one is a doozy especially with as ambiguous a language as English. Pronoun disambiguation! An example “repeat my last order”. Here your chatbot will need to match “my” to a user account.
7. How does your chatbot handle out of the context questions? Does it feel such conversations gracefully or not? Or does it just have a meltdown?
8. How does your chatbot handle overall conversation/dialogue? The ulterior motive is to make the user forget they are talking to a bot. How close is the discussion to real-world natural language interaction?
Coding questions to ask your chatbot supplier or developer.
9. What is your chatbots portfolio? What case studies do you have?
10. What have you learnt from building other chatbots?
11. Can you repeat to me what my chatbot will do? Now document it.
12. Who is responsible for writing the content? The flows? The canned responses?
13. What if my chatbot stops working? Who will I go to get it fixed and up and running again?
14 Where and how is my chatbot hosted?
15. How and where will people interact with my bot? In other words, where will people find and speak with your chatbot?
- How difficult will it be to have the bot on our website?
- Can the bot be used with our mobile app?
- Facebook page?
- Do we need to add channel functionality at the beginning or can we add another channel at a later date?
Start with the channels that are already where your punters are and where they are the most engaged.
If you have all your fans on Facebook and only a fraction on Twitter you can probably pass on a Twitter bot for now. With many companies, it might be the help and support page on your website.
16. What will the bots data sources be?
No chatbot will work without access to data. This data could be from your website forms, knowledge base, faqs, existing documents, needs analysis, reservation systems, patient records, shipping information, product inventories, mobile app and many other sources.
What you want the chatbot to do will define and decide what sources of information to make available to the chatbot. How data sources will be integrated will be one of the most critical questions to ask your chatbot supplier. Say you’re an internet service provider that needs a chatbot to help their customers manage their account or to resolve a problem with their internet connection. To efficiently perform all the tasks associated with these types of transactions, the bot will need access to:
- Information about the customer’s subscription
- Previous ticket history
- Account information
- The isp’s billing functions and systems
- Knowledge base of prior problems
How the bot will access data sources will need to be mapped out. Is your current knowledge base adequate, will it be able to cover all scenarios? Will you need open APIs, how will integration work? Will PIN or password data need to be accessed?
17. What, when, how will chats be transferred to a human?
Bots like humans are fallible. There is always one user who will throw something at your bot that it can’t handle. A problem even it cannot solve or provide a sensible answer to.
Examples of when an escalation to a human agent could be required,
- if a customer requests a phone call
- when language is used, that indicates frustration
- when managers monitoring the chat take over / override
- as a customer repeats a question multiple times
- when a question is asked that the bot doesn’t know how to answer
18. Will the customer have to repeat their story if there is a handover to a human agent? At any point will the customer be told they are talking with a bot? If it was not a manager override, will the agent taking over be given a summary of the conversation with the bot?
19. Are you using any other services to create my chatbot?
20. What if my chatbot gets a gazillion people all messaging at the same time? Probably a question for you more prominent brands only 😉
21. More important questions to ask your chatbot supplier will be around the IP rights for your bot? Whose bot is it?
22. How can I drive engagement and marketing for my bot? Any built-in features?
I hope you find the list useful. Here at Growth Hakka we can of course help you on your journey to build a bot, whether its a text based bot or even an Alexa voice bot. Read more about our Messenger chatbot & Alexa Voice chatbot solution.