The Business Leader’s Guide to Launching AI Chatbot Programmes

From Ada to Leo to Temi, Artificial Intelligent (AI) Chatbots are increasingly finding their way into the wish list of business leaders. I am still amazed at the number of organizations that are just starting to dip their toes into AI chatbots, given the high costs of maintaining customer service centres for Africa’s large population.

A Chatbot is a software that can simulate a conversation (or a chat) with a human user in natural language through messaging applications (WhatsApp, Messenger), websites, mobile apps or through the telephone. Chatbots are important for Africa at this time because many factors have aligned perfectly to make Chatbot programmes strategically important.
Such factors are
• the Covid-19 pandemic forcing most organizations to rethink how business should be run virtually,
• the emergence of messaging platforms like Whatsapp as the primary communication channel amongst Africans
• The high cost of providing personalized services to our large population in Africa
• The success of the SMS/USSD banking bots.
• The increasing number of customers willing to transact and buy over messaging platforms like WhatsApp.

Most businesses in Africa can benefit from an AI Chatbot programme however the ‘perceived’ complexities can scare many managers. Some managers that considered launching AI chatbot programmes in the past may have been turned off by all the technical terms like Natural Language Processing, APIs etc that technology professionals like to throw about.

However, a Chatbot programme is much more than the underlying technology. If anything, AI Chatbot programmes are 80% process and 20% technology. It is far more important to get the process right than to pick the hottest AI technology tools. In this 3- part article I present an easy to follow step by step guide for every business leader looking to start or manage an AI Chatbot programme in their enterprise.

1. Start with why

As with any product, a chatbot should solve a fundamental strategic problem for an enterprise. It is unfortunately very easy for organizations to fall into the me-too trap and launch AI Chatbot programmes because competitors are doing it. The results from such me-too chatbots are usually muted.

It is important to have solid ‘Why’ behind any AI Chatbot programme and as such, a solid business case must be made before starting to talk about technology.
As part of defining the business case, managers must define the benefits for the two sides of the conversation. The benefits to the customer and the benefits to the organization.

Benefits to the business are usually based on 3 key factors
• Increasing Revenues (e.g New Sales Channel)
• Reducing Costs (e.g reduction in customer service costs)
• Minimizing Errors (e.g automation)

Benefits to the customer have usually based on the core behaviours that drive customers in digital interactions (see Digital Transformation PlayBook book by David L. Rogers). These are
• Access – providing a faster and always-on experience for customers
• Engage – interacting with the customer one-on-one
• Customize – providing personalized services to an ever-increasing divergent customer base

• Connect – giving the customer a voice and making their opinions matter
.

2. Know Your Customer

The second stage of managing a chatbot project is understanding the habits and preferences of the end-users. Understanding your customers helps informed decisions like what interfaces the chatbot can interact on and how the conversation flow should be designed.

For example, if most of the end-users prefer WhatsApp over telegram then an organisation can better focus its efforts on developing a chatbot for the WhatsApp channel rather than develop for all platforms.

3. Determine the Primary Role of the Chatbot

“We want the Chatbot to be able to do everything”.

This is a common statement we hear from clients when discussing chatbot projects. This of course is not possible!!

A Chatbot must at least start off doing one thing and doing it very well. For success in any chatbot project, organisations must determine early on what the primary role of the Chatbot should be. Making this decision early on helps guide the team in later stages and helps keep any AI Chatbot project on track.

In the book, ‘A guide to AI chatbot Project management’, the author (Igor Luzhanskiy) presents the six different purposes of chatbots.

These are:

• Sales – for helping customers make purchases
• Lead generator – for keeping users engaged and converting leads
• Interface – for providing access to external services.

• Informant – acts as a directory service for providing information
• Helper – customer service bots that help users solve their queries and issues
• Psychologist – for providing advice and counselling

4) Define the Intents

Once the primary role of the chatbot is known, the next step is to define the list of possible intents the Chatbot should process. Intents can be seen as the jobs to be done by the user when interacting with the Chatbot. The most important question to ask at this stage is ‘What is our customer trying to achieve whenever they use our products/services?’

For example, possible intents for a Supermarket AI chatbot could be :

• Buy groceries
• Check price of milk
• View opening times

The list of intents defined at this stage should be as comprehensive as possible (aim for 70-90% confidence level) and as such, managers must gather data from all customer-facing teams on needs and wants of customers. Other sources of data like Website traffic, emails, FAQs can also be reviewed

Finally, it is important to note that the list of intents should be in line with the primary role of the chatbot. For example, a psychologist AI Chatbot has no business selling airtime credit to end-users.

Read Also: Flutterwave commits N5m, partners Lagos State to help businesses rebuild after looting

5) Define Entities

Entities can be defined as the pieces of information required from the user to achieve an Intent. For example, the entities required for an Open Bank Account intent could include:
• First name
• Last name
• Date of birth
• Email Address
• BVN
• Phone number

For each intent defined in step 4, the required entities need to be clearly defined.

6) Select Channels

This is the perfect stage to decide what platforms the AI Chatbot will run on. If the steps in this guide have been followed up to this step, the team would already have enough data (like customer preferences, customer intents) to make the right decision on the best channels to deploy on.

The possible list of available channels are as below :

• Email
• Facebook Messenger
• WhatsApp
• Telegram
• Email
• Phone
• SMS
• Website

7) Select Conversational Styles

The conversational style of the Chatbot is the mode of interaction between the end-user and the Chatbot.

There are three conversational Styles to choose from

Structured

This conversational style lends itself more to the use of fixed menus and options. The chatbot presents the user with a defined set of menu options from which the user can choose from. Consider using this conversational style if the Chatbot will have a clear and defined set of Intents or if there is a need to use local languages to interact with users.

Unstructured

This conversational style is based on human natural language for the interaction between the Chatbot and the user. For example, users may be able to interact with an Airline Customer Service Chatbot using statements such as ‘When is the next flight to Abuja?’ and the Chatbot can similarly reply with statements like ‘The next flight to Abuja is on Arik Air at 2:30 p.m. Would you like me to book a seat on this flight for you?’.

Consider using this conversational style if the Chatbot will have a very large number of Intents that makes it difficult to display in a menu format or if the primary role of the AI Chatbot does not lend itself to using menu options (examples are Counselling Chatbots).

Mixed

Using a combination of structured and unstructured conversational styles. Consider using this conversational style if the Chatbot will have a mix of users with varying technical savviness.

8) Decide on ChatBot character

Chatbots should absolutely have characters. Chatbots can be funny, polite, professional, aggressive etc. The Chatbot character helps define the tone and nature of the conversation with users.
The Chatbot can also be given a name as the likes of Temi and Ada.
In defining the Chatbot character, the brand and the values of the organisation should be considered.

9) Design Conversation Flows

This is the stage where we put pen to paper to design. In this stage, a script is developed to model the conversation between the Chatbot and user. A script needs to be generated for each intent defined in Step 4, and the script should define the full conversation up till the user achieves the required goal.

The product manager, along with other key stakeholders need to be involved in this process.

For example, a sample conversational flow for an open bank account intent could look like below:-

Bot :- Hello, how can I help you today?

User : I would like to open an account.

Bot : okay great, can I take your first name, please?

User : Tolu

Bot : Thanks Tolu. Can you tell me your surname?

User : Adelowo

The selected character of the Chatbot must be used as a guideline when crafting the responses of the Chatbot. For example, a professional Chatbot could respond “Thanks Mr Tolu” while an informal Chatbot could respond “Thanks Tolulope”

Read Also: Emerging technologies are fueling digital revolution in the Oil and Gas industry – Ayinde

Select Development Tools

At this stage, the development team should select the most appropriate tool for creating the chatbot. The selection of the development tools could possibly happen earlier but it is better to use the learnings of the previous steps as a guide in selecting the most appropriate tools.

The key development tools required for an AI chatbot implementation are listed below:

Programming language
Database
Conversational Engine – this component helps manage and keep track of the dialogue between the Chatbot and the user.
Natural Language Processing platform : chatbots with unstructured conversational Styles require a natural language processor to interpret user input and predict the user’s intent.
Live Agent platform : It is important for a chatbot to be able to pass the conversation over to a human agent if Chatbot is having issues understanding the user. The Live Agent platform is the platform to be used to continue that conversation between a human agent and the user. Examples of Live Agent Platforms are Zendesk, Intercom, Zoho etc

Create Training set

Creating a training set is optional depending on if a Structured or Unstructured Conversational Style Chatbot will be implemented.

If going with an unstructured conversational Style Chatbot, then a Natural Language Processing platform will be required. Natural Language Processing (NLP) platforms analyze customer input and attempt to predict the user’s intent.

Like a toddler learning to communicate for the first time, the NLP must also be trained to communicate with the users. The process of training the NLP consists of feeding the NLP with past customer conversations possibly from a Contact Center, emails, contact forms. This process helps the NLP understand how the user speaks. The data to be used for this training is called the Training Set.

Define Test Cases

As with any technology project, testing must be done after implementation is complete. At this stage test cases need to be defined to guide the testing of the Chatbot after implementation. The Test cases should be defined for each intent.

Implement and Train

At this stage, the software development team will implement the Chatbot and train the NLP.

Test

During and after implementation, the chatbot should be tested using the test cases developed earlier.

Retrain

Learning is a lifelong vocation even for Chatbots. If implementing an unstructured conversational Style chatbot, the NLP will require regular retraining as it interacts more with users. In particular, the NLP will need to learn from failed attempts to predict the user’s intent so it can get better for next time. As more conversation history is generated, the Chatbot should be retrained with the updated conversation history on a regular basis.

As a final thought, enterprise Chatbot projects are not Deploy and Forget Projects. A Chatbot is like a child that requires constant nurturing and monitoring to become better. It is not unheard of to keep the project team behind for a period of 2 -3 years while they continually improve the Chatbot from feedback. It is therefore important to think medium-to-long-term when planning for starting an Enterprise Chatbot Programme.