The Electronic Retailing Association

Article: Artificial Intelligence: 5 important things you need to know about chatbots by Pablo Alvarez

Pablo Alvarez, IT Director, JML, UK

Pablo Alvarez: explorer, traveller, passionate about life and technology. For 16 years Pablo has been leading digital transformation journeys and strategic technical roadmaps in hospitality and retail businesses across Spain, Mexico and UK.

Background information

A chatbot is basically a computer programme that is able to hold a conversation with humans. The idea is not new; in the 1950’s, Alan Turing published an article which presented the idea that a computer programme could impersonate a human in a real-time conversation with another human, where the latter was unable to distinguish whether they were talking to a computer or a real human.

Early chatbots were developed following Turing’s principles like Eliza in 1966 or Parry in 1972. They were basically used to simulate written conversations, but it was fairly easy to identify the limitations of the conversation that could be held.

The first commercial chatbots appeared on websites and Alaska Airlines (2008), Ikea (2010) or Expedia (2011) were some of the early-adopters. Their chatbots were used as basic customer service agents that could advise customers on a series of pre-established subjects. In the majority of cases, the limitations of this technology were still obvious and customer service staff had to contact the customer to help with the query.

Up to this point, chatbots were merely used for written conversations but this situation changed when Apple announced the arrival of a new voice-assistant as part of the new smartphone's operating system in 2011. The increment of the smartphone's popularity, coupled with these new capabilities and fast internet-mobile connections, exponentially increased the interest of both technology companies and consumers.

Thanks to the wider development of machine learning, natural language and sentiment analysis, chatbots are today able to complete complex conversations, connecting information from multiple sources and behaving in a more-human way than ever before.

The impact of new technologies on emulating human communication

At the same time, the rise of AI (Artificial Intelligence) research, and specifically, two particular fields called machine learning and natural language processing, began to erode the barriers.

Machine learning is a fundamental concept of AI. Its main objective is giving the programmes the ability to learn without explicitly being programmed, meaning that the computer can learn from its own mistakes and experience, improving its own capabilities and responses in the future.

Natural language processing focuses on giving programmes the ability to understand human language. In any communication, it is extremely important that both parts understand the whole content of each sentence, don't just identify a few words, as each word or combination of words might change the whole meaning.

The developments in both fields are extremely important to face and overcome the next frontier to emulate human communication: sentiment analysis. The objective is to systematically identify affective and emotional states and adjust the responses based on them.

All these factors are key to understanding the one critical factor that will determine whether chatbots are a game-changer or not: that critical factor is trust.

The Hype Cycle shows the phases of development of new technology

The phases of development

As shown in The Hype Cycle Graphic (right), between the 'technology trigger' and the 'peak of inflated expectations' (early days of AI and chatbots) both expectations and trust grow. A speaker knows that he is not speaking with another human and it did not create any problem. Chatbots are used as tools.

However, there is a point within the evolution of technology and time where the trust drops. This is known as the 'trough of disillusionment'. This gap is created when a person interacts with something that behaves like an human, but a few (evidently not human) defects are perceived, causing an immediate rejection.

The following phase, the 'slope of enlightenment', is achieved through the evolution, improvements and application of the different aspects of AI described previously. All of them combined, produce the improvement and singularity that ends in the last phase, the 'plateau of productivity' which is achieved precisely at the moment where chatbots interact with humans exactly as Alan Turing described in 1950.

5 things you need to know about chatbots

#1 How to get a chatbot

A chatbot can be developed by an internal team, developed in-house using external resources or acquired from external providers.

If developing the chatbot in-house, it is important to consider that very specialised skills in AI and Natural Language areas are needed. An estimated investment of £40K would be required just for the development. Today it is unlikely for a company to take this route unless the final objective is to sell this service or to include it as a pack of services for customers, mainly because of the time and resources required to build everything from scratch.

In the second scenario, developing a chatbot in-house using external resources, the most complicated areas of development (AI and Natural Language) are left to external platforms. Amazon, Facebook, Google and Microsoft provide and offer these ecosystems, hosted on their own clouds where the developer can use pre-built connectors, cognitive services and templates. This is the most popular option as it offers flexibility to adapt the product to business requirements at a reasonable cost as part of an “all-inclusive package”.

The last scenario is acquiring everything as a service from an external provider (design, development, integrations, AI…). The whole platform, business rules, connectivity and data sit outside the business. The advantages are the speed of implementation, flexibility of resources and low upfront costs. The disadvantages are that any additional requirement needs to be paid, there is no deep in-house knowledge, and a third party company becomes involved with personal data.

Each business needs to evaluate carefully each option to choose the one most suitable for its requirements.

#2 What aspects need to be considered?

· Business goals: sales, retention, customer service, customer experience.

· Channels: the different platforms where the chatbot needs to interact with customers. Facebook Messenger, WhatsApp, Kik, Telegram, Slack, Twitter, Skype, Amazon Alexa, etc.

· Volume: the number of conversations (interactions) held with customers within a specific period.

These three elements will define the scope and the cost of the project as well as the on-going costs.

#3 In what stage of development are chatbots now?

For a long period, chatbots performed conversations at a basic level. They were able to maintain a conversation and provide answers for specific (and more importantly) predefined concepts. Frequently, human assistance was required as the chatbot could not process or understand the content of the request.

Even after the huge progress achieved in the last 5 years, chatbots are now in a stage between the end of the 'Trough of disillusionment' and the beginning of the 'Slope of enlightenment'.

However, thanks to the wider development of machine learning, natural language and sentiment analysis, chatbots are today able to complete complex conversations, connecting information from multiple sources and behaving in a more-human way than ever before.

Despite this, there is a long development path ahead and many improvements will still be required before customers and chatbots can fully interact in any scenario without the need for human intervention.

#4 How can your business benefit from using chatbots today?

Considering the actual stage of development, chatbots can be useful in scenarios where a business can define a clear set of queries, their answers and the links between them. For example, a business that commercialises a specific product and has a high volume of interactions with customers of which a large percentage are based on a few specific questions (Frequently Asked Questions), could consider to focus their customer service team on adding extra value to those customers who raise complicated cases or returns, and create a chatbot platform to deal with all the plain and simple queries that already have a clear answer, considering their 24/7 availability, anytime, anywhere. Not forgetting the invaluable customer information that they will capture and which can provide additional analysis.

#5 How are chatbots expected to develop in the next 3, 5 and 10 years?

3 years: Based on the existing development path, chatbots can be expected to manage complex queries like placing an order / request based on the best offer available to the customer or to deal with a complaint. This technology is already available today, but it is not mature enough to be deployed on a bigger scale.

5 years: At this point chatbots will be integrated with more internal and external resources. Human assistance will still be required in some specific scenarios, but chatbots will manage the majority of non-face-to-face interactions with customers (through phone and text) providing almost any information and assistance through multiple channels.

10 years: By then, the word chatbot will be an obsolete concept. Almost everyone will be using a "personal assistant" interacting just by voice on whatever device is carried at the moment (smartphones might be obsolete too). These assistants will be connected to an immense grid of companies and resources helping in almost every imaginable way.

In the medium-long term, chatbots will definitely change the way we interact with the rest of the world, as employees, end-consumers and human beings.

Additional resources: