Chatbot conference organized by Chatbots Life is among the best conferences for all the chatbot enthusiasts, bot entrepreneurs, and bot investors. It gave a perspective on where the market stands today and where it is heading in the future. As a co-founder of a chatbot company, I found the sessions to be useful to make the right business and technology decisions. Following is an attempt to summarise my learnings from the session.
“Though chatbots market is cluttered with over 1000 players, only a few companies are adding real customer value”
Over the past 3-4 years, a wave of AI-based companies has emerged, of which a good portion are into providing chatbots solutions. However, currently, there’s a massive gap in what these companies promise and what they finally deliver. The expectation gap has resulted in dissatisfaction among the customers and resulted in a bad reputation for the chatbot industry. Companies that have no real AI expertise are claiming to have AI in their product, whereas what they have is a simple rule-based responding engine. Also, companies are trying to provide end-to-end solutions without a specific focus area. ‘I am an AI company, and I can solve any problem’ – kind of approach is turning out to be a difficult problem to solve.
There is a small portion of companies that are providing real value to the customers. More than the technology edge, companies are striving towards the right business model and thus seeing success. Their solutions are backed by the right technology to meet customer needs.
Eventually, the market will get corrected with companies settling down to their niche and evolve as an ecosystem.
“Chatbots are best fit for external customers than internal customers.”
Enterprise applications of chatbots can be bucketed into four categories
- Contact Center
- Human Capital Management
- Enterprise resource planning/ Supply Chain Management (ERP/ SCM)
- Customer Relationship Management (CRM)
While Contact Center Use Cases are external facing, the remaining use cases are for internal users, who could be either sales or operations teams or any employee. The value-add that chatbots bring to external customers is higher than that for the internal customers. The reason is very simple. For internal customers, chatbots are more of an additional layer of comfort which they can live without. In contrast, with the increasing number of customers and the fundamental need to satisfy customers, contact centers find great value in using chat and voice bots. To summarise, there is a great demand for contact center chatbots than for HR or sales bots.
“State of the art chatbots are at Level 3 – capable of handling variations of conversation flow”
Rulai (listed among the top 50 promising AI firms) has come up with a mode to assess chatbots, which they have termed as VAAM (Virtual Assistants Ability Model). There was a need for someone to come up with a standard, and I sincerely appreciate the effort taken by Rulai in coming up with VAAM. Yi Zhang, CTO of Rulai, made the presentation.
As per VAAM, there are five categories into which each bot can be classified.
- Human-agent assistant
- Partial automation
- Conditional Automation
- Fully automated in designed domain
- Human-like automation
The categories can be as – Level 1 (Human-agent assistant) classified as simple, and level 5 (Human-like automation) classified as sophisticated and smart. Rulai CTO, Yi Zhang says that most of the chatbots in the industry are in 1st or 2nd class, and the state of the art chatbots are in level 3 – conditional automation.
Level 1 virtual agents (VAs) assist human agents but do not take control.
Level 2 VAs can take control of specific tasks, but the human agent should be ready to take over when the conversation deviates from the design.
Level 3 bots are capable of augmenting the work of agents and have the ability to handle complex variations of queries. However, they are dependent on human agents to handle complex tasks.
Level 4 VAs operate most of the time independently and can find answers for some unseen questions based on reasoning.
Level 5 is human-like automation, where VAs are recruited and trained. Humans, only manage VAs.
I believe the high-level categorization of VAs based on ability is excellent and helps in communicating the right message to the client.
“Start small and take along the customer through the journey. And don’t over-promise.”
The realization that many chatbot developers have today is to add real value to the customers, which gives long-term benefits, instead of riding the hype wave. It is the responsibility of the vendors to educate the customers about the ability of chatbots and provide the right measures to ensure there is a value add. This approach will separate the winners and losers in this game.
Here I have summarized the approach that was discussed at the conference. I took the liberty of adding my interpretations and understandings.
Start small with a measurable proof of concept with a very well-defined scope. Understand the real problem area of the client where bots can be of value instead of imposing unwanted technology. Use technology as per the business need. Have a shorter timeline and minimum budget allocation. Come up with the following documents:
- POC scope document
- POC success criteria
- POC timeline
After POC completion, take time to validate the use case. It is OK to drop at the POC stage if warranted, instead of pushing hard for the full project. If POC gives good results and confidence to the client, then go to the full project execution.
To start the full project execution, conduct a workshop with all the stakeholders, if possible, involve the end-users of the bot too. From the vendor side, there should be a bot architect, a CX expert, and a copywriter. Discuss the project scope, POC results, and the time for the project execution, training, and rollout. A live pilot with a small target audience can be included as a part of the final rollout. The result of the workshop should be the right inputs for the project implementation plan.
The full project rollout, in some cases, take from few months to over a year, depending on the use case. Sensitizing the customer about these timelines is very important to set the right expectations and right budgets.
We are still at the evolution stage in the Conversational AI space, and therefore the timelines are so long. Soon, there will be a day when technology will solve many of these hurdles and standardize the approaches with the right amount of data, and thus, reduce the project execution time. I am looking forward to a bright future for conversational AI.
“Conversation design is a very crucial step in bot building”
Hans Van Dam, CEO of Robot copy and co-founder of Conversational Academy, gave a fascinating talk about the importance of conversation design. I’ll write a separate article on conversation design; here, I’ll try to provide a high-level overview of the topics covered during the session
What is Conversation Design?
Conversational design is the first step in building a conversational agent. Just the way blueprints and wireframes help in developing web and mobile applications, it is essential to design conversations before initiating any technical implementation. Conversation design is the process to identify possible scenarios where a VA interacts with humans and models appropriate VA responses keeping in mind human emotion and persuasion.
Three mistakes in Conversation Design, as identified by Hans, are as follows:
- Start with technology
- Focus on knowledge management
- Start with the business process
Many of the bot developers start with one of the above steps and struggle to meet the desired outcome.
Conversation Design Process:
Conversation design has the following steps in the order:
- Use case identification
- Identifying bot needs and user needs
- Sample dialogues building
- Testing the dialogues in the early stage
- Expert rewrite
Identifying the right use case is the first step in the design process. Use-case is defined well if we capture the following details.
- Use-case name – Ex: IT help desk assistant
- Target user persona
- Identifying the scene when a user would interact with VA
Identify User Needs and Bot Needs Separately:
Following is one example.
The user, who works for the finance department wants to reset her outlook password. She is in the middle of a busy day and doesn’t want to waste much of her time in this process.
A bot can reset the password for 30 out of 40 available systems. The bot needs to know the system name, employee id. Before resetting the password, the bot needs to get a final confirmation by providing the one-time password.
Sample Dialogue Building:
Sample dialogues can be built with the help of a role play with two people. One person is the bot and the other person is the user. They sit back to back to prevent non-verbal clues. The person playing the role of the bot has the first draft of the script and the other doesn’t hold any script. The conversation is initiated and iterated till you get it right.
Testing the Dialogue in the Early Stage:
Replace the person who plays the role of the user with a new person who has a little clue on the conversations designed. Iterate once with this person and see if you get it right. Make minor changes at this stage if required.
During the sample dialogue-building process, draw flow charts to capture the conversation flow.
Once the dialogues and the flows are in place, use a professional copywriter to improve the conversations.
Detailed Conversation Design
In the detailed conversation design, identify the intents and entities.
The human-centric design process ensures cost and time saving during implementation, and also provides good CSAT post-implementation.
Importance of having a bot persona
Emotion is one of the key components of conversation. Making sure the right emotion gets communicated can be handled by the use of appropriate words in the bot messages. The next question is how do you define the right set of words? We can do so by first deciding the bot persona which helps in picking the right word. Some of the common personas used in bot building are – friendly, empathetic, professional, complete assistant, assistant cum analyst, analyst, etc. Adding demographic features like age, sex, name, nationality, etc will help further in response building.
To summarize, the chatbot conference was successful in bringing the right set of industry leaders in conversational AI space under one roof. This helped in sharing knowledge and helped the industry take a step forward into the future.