At a recent lunch with senior executives from housing providers, the conversation started with all the excitement around the potential of AI, including automated translation for tenants not confident in English, the identification of stress or emotion in conversations to prompt interventions, predictive maintenance, energy efficiency and the value of identifying and acting on what our data might tell us about the health and well-being of tenants. However, I certainly felt a consensus emerged from the discussion; that the economic case for investment in complex AI or machine-learning systems has yet to be convincingly made.
Given the proliferation of AI-related headlines, can this really be the case, and can the social housing sector really ignore these technologies? And while true AI remains expensive, will rapid changes in cost and availability shift the balance in the near future?
AI developments
Looking at the first area that caused a buzz, the potential of applying translation algorithms to speech-recognition technology is very real and techniques are improving, although the accuracy of speech recognition is hovering around 75 per cent due to the varieties of accents and pronunciations in the real world. Similarly, research to detect the impact on our voice patterns when we are stressed or experiencing emotion are now being implemented in commercial systems. While insurance companies are reportedly interested in the relationship between emotion-driven voice patterns and the incidence of fraud, in the housing sector the ability to prioritise calls or contact from vulnerable or distressed members of the community offers real benefits to housing providers and our partners in other social and blue-light services.
Factories and complex machinery are increasingly making use of sensors to allow predictive maintenance, with real-time flows of data allowing AI models to determine how and when intervention is most efficient. A number of housing groups, including Coastline (with the University of Exeter and their Smartline project) and York City Council (with Pinacl Solutions), have been running trials where sensors are placed in homes and data on a range of parameters, including temperature, moisture levels and carbon dioxide, captured and analysed to develop models for damp prevention, insulation improvements and boiler maintenance. The interim results, with the current costs of sensor purchase, installation and monitoring in addition to the cost of strengthening the infrastructure for increased data access, don’t yet appear to demonstrate a compelling economic case based solely on house and equipment maintenance. However, the sensor data are throwing up interesting insights into living conditions which will inform debates around the roles of housing organisations, social services, health and blue light services, and around data privacy and the ways in which data could be shared across partnerships.
Letting and estate agents are making increased use of imaging technology to provide prospective occupiers with virtual walk-throughs, with AI tools analysing preferences to suggest new viewings and undertaking a lot of the transactional side of the lettings process on mobile devices without the need to come into an office. These applications are directly applicable to choice-based lettings but perhaps more innovative would be risk or repair insights identified by AI algorithms from videos or images made during visits to homes for repairs or void turnarounds, although of course the considerations around the protection of personal data become extremely important in imaging applications. Companies such as Fujitsu and Hitachi have been working on a variety of industrial projects which harness imaging and AI to deliver impressive and sometimes unexpected outcomes.
Machine learning & predictive models
AI advocates make great claims for the benefits flowing from machine learning-based predictive models based on the processing of data ‘lakes’ – i.e. all the data held by the enterprise. In housing, these might include income collection by predicting the ability for tenants to pay rent and putting support in place, translating behavioural patterns which signal risk into earlier interventions and improving customer satisfaction through pro-active human contact following indications of stress during contact centre communications etc. However, businesses can achieve significant benefits ahead of any application of machine learning through the process of cleaning up their data, structuring it correctly and using applications like Power BI to interrogate and present the analysis to inform business planning. Making greater use of existing data, alongside more real-time monitoring of vulnerability and financial inclusion flags are the current focus of many housing providers. Once the value released by these activities has been demonstrated then greater use of predictive models could be considered, but in our experience the incremental benefit of AI cannot be assessed until this has been achieved.
The development of an enterprise-wide AI system can’t yet be achieved by procuring a set of tried-and-tested software applications that neatly integrate. Instead, organisations are identifying and implementing discrete projects where a use-case can be identified and where scale and benefits justify automation. The hoped-for larger benefits can often only be accessed when several discrete elements can be combined with connected technologies. Companies such Converse360, the CX and communications specialist which has been focusing on AI products, have been working with housing groups, software suppliers and providers of contact centre services to develop AI offerings in the housing sector in more holistic ways and have some interesting development trials underway. It seems likely that these will develop into add-on modules within existing commercial housing packages.
A coherent digital strategy
The fragmented landscape of AI possibilities and potential benefits continues to evolve rapidly and can be bewildering. While the claims of suppliers, the apparent success of peers and the advertised ‘standalone’ nature of an AI project can be tempting, we believe that organisations are better served by identifying and prioritising projects in the context of a coherent digital strategy. The strategy can (and should) allow pilot projects to be created at relatively low cost to provide proof-of-concept and to test RoI and customer take-up. However, just as with other IT development projects, housing groups need to consider all of the implications of the future deployment and support of AI projects, including how risks around mission criticality will be managed when moving from trial to production, staffing needs (perhaps including the recruitment and retention of data engineers and scientists), data privacy and security, and infrastructure needs (including the potential impact of large data flows on existing infrastructure usage, even if cloud-based).
Of course, the automation of customer interactions and the back-end business processes discussed in this article are enabled by passive AI and pattern recognition. Active AI – robots that interact with us and each other and learn by trial and error rather than operating under prescriptive rules – are coming. They will serve us coffee. They will transform manufacturing. And they will undoubtedly be in our homes. Just as with self-driving cars, the possibility of human error will be removed from tasks such as inspecting critical safety systems, like gas meters today (although presumably our homes will have migrated from hydrocarbons to other sources of energy by the time robots are ready for widescale deployment in our communities). But, despite the extensive R&D and jaw-dropping investment, home robots will not arrive for some time to come and can certainly wait for a future issue of Housing Technology.
Andrew Webb is the CEO of Alysium Consulting.
