Expert Interview with
The Conversational Marketing Masterclass
Srini is a chatbot designer and programmer. He has a Ph.D. in artificial intelligence and Natural Language Processing from the University of Edinburgh. He has published many articles on the subject of AI and created a handbook on Chatbots and Conversational UI.
It started 20 years ago – the first chatbot built was in the year 2000 – I was a undergraduate.
The project I was working on was to build a Natural Language interface that connected to a library database system.
My main motivation probably came from SCI FI Movies. Other motivation is that it is very challenging field. Natural Language has a inherent ambituity problem. As human we dont pay attention to these problems. I like to see how I can help machines communicate more effectively
There are lots of use cases.
Basically, chatbots automate conversations. Where we find conversations we find people. As we are social humans. We can put chatbots anywhere there is conversation.
From a business point of view:
Chatbots are used all over the customer journey. In product discovery, marketing, sales, and services.
I believe 80% of conversations can be automated – they are repetitive and simple. A lot of people have the same problems. And in many cases, you don’t need a human. Therefore you can use conversational intelligence to automate 80% of them. It’s Pareto’s principle. 80% traffic is generated by 20% of problems.
I believe the most impact is created in support. Customers need a web chat with an agent. However, there are a lot of customers and very few agents (long queue). As a customer, it’s not justified that you have to wait so long in the queue.
Chatbots lack the skills for critical analysis and emotional intelligence. So what you want to do is plot a graph
X axis = frequency
Y axis = complexity
Best use cases for a rule-based chatbot.
High frequency and low complexity.
Low hanging fruit:
Low complexity and low frequency:
Live chat agents (human take over):
High complexity and low frequency
High complexity and high frequency. The most challenging.
We can divide the conversation into a human and bot: the 1st part of the conversation the bot can handle – qualify and pre-screen the customer and then hand it over to a human
In that case, there is synergy between the human team and (AI) bot.
Collective intelligence – the model we should focus on.
Humans and AI can collaboratively work with each other. How we can provide extra support to the human team. I would like AI to be a superpower that you give to humans so that humans become super agents.
Humans do what they love and they want to solve the interesting stuff.
In the last years. a lot of their big players opening up their platforms for chatbot possibilities. Think about: Skype, Facebook Messenger, Telegram.
More and more organizations were automating their business through chatbots. When that happened there were so many players. A lot of small companies started up to build a solution for this. People were starting to explore this whole space. There were many questions and I just wanted to give people a birds-eye view on how to deal with this new technology. There were a lot of questions about how chatbots would change websites and apps. First, it was a way to provide clarity for me, and later I wanted to share this with others.
I have several questions that give you a high-level overview of what it takes to build a chatbot.
If you are a business and you want to explore conversational AI these are questions that will help you
Structure it in the same way as the business model canvas. A business model canvas tries to showcase the entire business model of a company in one canvas. All the essentials in one canvas!
During the interview, we discuss the different components.
Explanation of the components:
These components Srini also discusses in this article.
Chatbot design canvas:
The main question you want to ask yourself: What kind of tasks are you looking to automate?
It’s not difficult to build either of two bots. However, it’s definitely difficult to provide a great customer experience. You need to keep your feedback loop and see your analytics. Finding any sort of issues and try to repair them as soon as you can.
When it comes to the choice between AI bot and rule-based bot there is no strictly separate solution. You can have a hybrid solution and this is what’s working across the industry. The customer comes in and explains his issue to an AI bot. Therefore you use an intent classifier. This way you classify what intent is of the customers. Once you’ve predicted the intent you put them on a journey that is rule-based. You’ll be serving a journey that gives buttons and interactions that lets them navigate through the journey.
As a company, you want to share the control. First, you identify the problem (the customer is in control) then you want the bot to be in control (unless the bot has it wrong). The last thing that you want is the customer to be guessing.
The trends are looking very exponential for chatbots in the coming year. Exponential means both usage and effective use cases.
Single turn or multi-turn conversation
Things are moving very fast. The moment we have bots that can do simple FAQ-type conversations (question vs answer) within one turn or few turns.
Conversations about the database.
This is from a more functional perspective. Being able to get data from the database.
Natural language interfaces that connect to database systems are not mainstream yet.
Being able to identify the next thing to say. At the moment it’s very scripted. The bot makes a plan to achieve a goal state and knows the most efficient route to get there during the entire conversation. It’s similar to google maps that describe how to get you to your destination.
Expert (video) interviews