Today, the term artificial intelligence (AI) is thrown around rather generously. As businesses around the world become more open to making waves and ditching legacy technologies in their quest to become data driven, an ever-increasing number of tech deployments are claiming to use AI or machine learning (ML), writes Soeren Bech, VP EMEA, Persistent Systems. One prominent example is in the use of chatbots to resolve customer enquiries, complaints or comments rather than human customer experience (CX) personnel.
Although chatbots have been in use for over two decades, reactions to this technology have been mixed, with many left dissatisfied or frustrated by the rudimentary and often inefficient service they provide. However, developments in recent years have elevated simple chatbot functionality to the more refined realm of Conversational AI (CAI) to deliver much more than automated responses – instead using sophisticated AI to not only access and harness the intelligence of a whole organisation and its systems, but also to provide a realistic human-like contextual and branching conversation, in order to enhance and maximise CX.
Why CAI?
The motivations for using CAI are multifaceted, but a key driver is usually cost efficiency. It’s an inescapable fact; an AI-powered technology solution is a more cost-effective solution than employing a team of 25-50 people who are restricted in terms of speed and working hours.But it’s not just a matter of money. If somebody has an urgent enquiry ‘out of hours’ or needs an answer immediately, CAI bots can deliver these quickly and at any time, as well as being able to directly access all business data in order to solve the query.
Technical requirements
While implementing conversational AI chatbots, one significant challenge is how to select the appropriate platform from all the available options. There are many platforms in the market, such as Kore.ai, AWS Lex, Google DialogFlow and AzureBot. Most of these support a standard set of basic functionalities to which they keep adding extra features and differentiators to make them stand out. When choosing a platform, it’s important to consider a number of technical aspects, such as the need for robust support for backend APIs; the number of supported channels through which the bot will have access to or interact with end-users and the ability to seamlessly connect to ‘human support’ when required. It’s also vital to assess the sophistication of the available NLP / ML algorithms and the number of languages which need to be supported throughout bot applications.
For any bot that needs to have “real-time” data to answer a customer enquiry, they will need to access APIs for various management systems including ERP, BPM and CRM software.
Follow The Stack on LinkedIn
Typically, this is achieved via a Proxy façade, which allows these two pieces of technology to interact and communicate. These systems sit behind maximum security walls and no organisation wants to expose them directly to outside contact. So, a thin layer of API gets written over them to supply just the required amount of data to the CAI bots. For example, when a customer places an order, the API for this object might be exposing 100 fields, but CAI bots will probably only need access to just 10 fields. A proxy layer sitting between core APIs and CAI bots can solve this discrepancy by hiding non-required fields as well as mapping and transforming required fields to make the process of connecting the AI with the bot easier and safer.
For CAI bots that are expected to handle free-flowing text conversations with customers, it is also important to design specific training modules to educate the bots so that the right amount of language is captured. This can include vocabularies including domain-specific words and abbreviations and regional dialects. This allows the bots to speak the language that end-users are comfortable with instead of forcing them to learn new terms – this in turn enables a more natural ‘human-like’ interaction.
Infrastructure is key
In order for CAI systems to work effectively, they need to boost internal operational effectiveness, as well as deliver specific use cases that directly impact and benefit customers. As such, it’s crucial to ensure bots are platform-oriented and underpinned with the sufficient infrastructure to support a wide range of uses.
However, where CAI excels is in situations demanding quick decisions based on actionable business data. Bots can generate responses and intelligently hold conversations around logical information, giving yes or no answers, finding information, suggesting options, and directing people to certain locations or documents.
A great example of Conversational AI in use is AGCO, the global agricultural machinery manufacturer, which has recently implemented CAI technology to deal with customer enquiries. CAI bots can not only answer questions about which component is required and the stock levels of those parts, but can also elevate the conversation to include a call to action, asking customers “do you now want to order this part?” before processing sales requests.
Delivering consistency
Conversely, many of the qualities we often attribute to human behaviour do in fact make CAI stand apart.
As many of us know, there are few things more frustrating as a customer than having to repeat ourselves to different customer service representatives as an enquiry gets passed around individuals or departments. Failure to recall your query, your customer details or your enquiry record can lead to a lot of time wasted on lengthy calls, and frustration at having to explain your story again and again.
Enter the CAI bot which, while not human, has the ability to remember and unlock conversations and records through its direct access to all of the data within the organisation. In these situations, giving your customer number to a bot can eliminate hours of lengthy conversations and give you a quick, uncomplicated answer to your query – and on top of that, it’s impossible to frustrate a chatbot.
Conversational AI chatbots and generating business decisions
Where CAI differs from traditional chatbot technology is its ability to perform analytics on conversations and queries like these, exploring which questions or comments are asked the most, what people need and what frustrates people. The benefit to the business is that strategic decisions can be planned and executed off the back of this data, putting measures in place to anticipate these enquiries and deal with them swiftly and effectively.
For AGCO, this could be an additional tangible benefit of using CAI. By performing analytics on the bot data, it will be able to assess which parts are requested most regularly and, as a result, make sure these are in stock. This simple yet quantifiable benefit will be felt by both customers, who feel satisfied with quick and competent service, and the wider business, which can use this data to forward plan and future proof its operations.
Additionally, using centralised AI is a great way of unifying the customer experience. As the bots are programmed with specific software and commands, their answers will be consistent, meaning that all customers will be treated the same way and get the same experience.
That said, customisation also plays a big factor in how CAI has evolved since the early days of chatbots. For example, today’s CAI solutions can distinguish between regional accents and dialects, pick up slang and abbreviations, and adapt to nuances within individual conversations. Additionally, sentiment analysis enables the Bot to detect for instance frustration or confusion and as a consequence re-direct a conversation.
The future for CAI
As with any technology, solutions in CAI are constantly evolving and new functionalities and possibilities are emerging, which all have the potential to transform businesses and the way they talk to and interact with customers. One of the most exciting developments currently being explored by those creating CAI solutions is the combination of AI with Augmented Reality (AR). By adding a layer of visualisation to customer conversations, the bots can ‘see’ the problem with a machine or a part, or access information on a laptop to give a tailored and accurate solution. Using this additional visual stimulus, the bot can then suggest a specific solution, point to a page in a manual, or elevate the query to a human expert.
