We’ve all seen the alarmist robot headlines and hyped supplier claims about how the fourth industrial revolution will change the nature of employment for humans and the way our communities are organised. But most of the AI that we currently see has been in passive technology in retail-driven areas where there is a wealth of available data and where pattern recognition and predictive models have been deployed, often around your next purchase or preference on Netflix, Amazon and the like.
These systems deploy cutting-edge models and there is no doubt that they, along with improvements in hand-held devices and home-based technologies, have changed the way we live our lives, embracing health and wellbeing, shopping, dining, finance and entertainment. Social housing can’t ignore this trend in consumer-driven technology, but how can AI and machine learning be deployed effectively in the sector?
This article looks at the conversational agents or ‘chatbots’ which have proliferated over the past two or three years. Despite chatbots having been developed by commercial businesses, it is widely accepted that few businesses are currently getting any real value from their attempts to enhance conversational interfaces with AI, or at least what they describe as AI. This lack of current success is attributed to a combination of poor deployment (esp. a lack of thought about the customer journey), the quality of the conversation supported, and user expectations.
Why use chatbots?
So why are businesses pursuing the deployment of chatbots? The two main reasons are the belief that automation will save money and the desire to provide a more satisfying experience for the target users (whose expectations have been shaped by instant transactions on Amazon).
Where are chatbots currently deployed? Customer support is the most common starting point as a use case. In the housing sector, this might be internal customers (e.g. contact centre agents dealing with requests raised by their colleagues) or external (e.g. tenants). The efficiency gains and cost savings associated with shifting enquiries from face-to-face and telephone enquiries into other channels have been successfully demonstrated by many of Alysium’s clients, although the behavioural shifts that come with this achievement are generally hard-won and require a clear strategy, detailed planning, and sustained effort.
To date, most of the automation has been around allowing human contact centre agents to work more effectively and to ensure efficient escalation where necessary, while simultaneously allowing customers to contact the agents using the multiple digital channels that they are already using in their personal lives. Clearly, the deployment of a chatbot which can adequately respond to a meaningful number of enquiries which would otherwise require agent intervention will enable the contact centre staff to focus their efforts on other enquiries and getting things right first time.
When it comes to the quality of interactions with a chatbot (which is closely tied with user expectations), there have been significant advances in natural language processing and the ability to integrate with other information sources (such as information already available from a website).
Chatbot conversations
However, no-one should expect to be able to have a ‘normal’ conversation with a machine. At present, people tend to guess what a chatbot might understand and often talk to it as if it were a child. Chatbots represent progress on a single-pass enquiry which you might undertake with an internet search engine (they are designed to ask questions of the customer in order to refine a response), but more complex queries requiring more utterances require significantly more complex models and more ‘true’ AI.
The key challenge continues to revolve around just how hard it is to recognise the different ways that people say the same thing. Future advances may allow a bot to develop context in a conversation and recall and reference issues already discussed but typical chatbot interaction currently involves around three utterances leading to completion of the task or provision of information. For voice interactions, a single turn remains the norm. At present, it would be more accurate to re-state the rationale above as “a desire to provide a more satisfying experience for the target users with simple enquiries.”
To deliver real efficiencies, the chatbot has both to recognise the customer’s intent and then either deliver an adequate summarisation of information or complete a task. If a chatbot when asked “What do I do about my noisy neighbour?” simply provides the email address for the ASB team then, although it may save the customer a little time by not having to call the contact centre, the subsequent email enquiry will still go through the existing channels.
Automation and application integration
To gain the benefit of customer-facing automation, an intelligent chatbot must be truly integrated with systems including the housing management, CRM and repairs systems and these systems (and the organisation’s work practices, including authorisation for work) must be able to act on automated requests. Achieving this business process re-engineering is part of the digital transformation journey and represents an opportunity to strengthen processes and improve customer satisfaction.
A chatbot which undertakes tasks for tenants, akin to the functionality delivered through the portals and apps that a number of housing providers have bought, is achievable and developers such as FuzzLab have been drawing on their repairs and maintenance expertise with housing providers and local authorities to develop and deploy chatbots which allow residents to diagnose repairs and make direct bookings.
Common questions and tasks
Rather than investing in the development of new models and neural networks for deep learning to give the bot the ability to devise responses to new questions, with the development investment and delivery risk that this entails, many organisations are choosing to focus on common enquiries and provide their chatbot with a bank of answers based on templates used in the pre-AI CRM. This can offer good value, with limited development costs and lower risk. A simple chatbot capable of answering a bank of common questions and even completing tasks such as booking appointments is now an achievable and relatively inexpensive project, and a number of housing providers are engaging in trials of resident-facing chatbots.
Most bots are initially designed for text and messaging only. The choice of development platform is an important factor to consider; for example, the larger platforms make it relatively straightforward to add functionality such as a voice interface through speaker technology such as Amazon Echo, Alexa and Google Home, computer services such as Cortana or mobile phone technologies such as Siri.
The speech recognition software used by these systems is fairly impressive but each has its limits and harnessing these with emotion and sentiment recognition requires a more complex integration project. The investment case for commercial off-the-shelf interfaces does not rely on the development of incremental AI beyond the technology already embedded in these assistants, but will for now be based on the cost savings and customer service improvements associated with the channel shift to chatbot interaction versus the cost of device deployment and maintenance.
Andrew Webb is the CEO of Alysium Consulting.
