Using AI to analyse users ‒ promise or threat?
Why analytics can offer much more than just compliance monitoring.
Few literary characters have had more exposure to the fluidity of ‘truth’ than Winston Smith in 1984. Central to the novel is that the collection of data about the citizen empowers the state to the disadvantage of the individual. At the time of writing the logistics of such an operation was not sustainable. Operatives would need to be nearly one on one to citizens to detect anomalous behaviour. Orwell would have written a very different novel today if he’d known that AI could simultaneously monitor every aspect of citizens’ digital lives and judge what was, or was not, within acceptable parameters.
This is increasingly the focus for many organisations faced with cyber threats. Zero-day attacks and user-triggered malware mean that conventional anti-virus solutions are not enough. Instead, complex telemetry and monitoring is used to judge whether user behaviour is normal, potentially threatening or actually compromised. Understanding and responding to user behaviour goes beyond simple geographical location and ‘geo-fencing’. It extends into data taxonomy, business context and role-based activities.
AI and big data ‒ two digital disruptors
This is an area where AI and big data combine very effectively, providing real time analysis of a workforce’s digital activity and automating defensive and preventative actions against attack or data loss. But is there an option to take the same data, combine it with business data, and use it for a much more positive purpose?
The use of AI in medical profiling by organisations like Google and IBM has been well documented, but similar application to workforce productivity has been less well explored. Let’s first consider the possibility of linking business and role-based data. If we can use AI to cross reference specific metrics and look for patterns, can we use this for business advantage? Specifically, can we consider costs, behaviours, engagement and outputs (or outcomes) to generate some notion of business value – and potentially of best practice?
From a cost perspective we can allocate specific values to a particular role, based on average salary, resource usage (premises, software, services) and overheads. This gives us a cost-profile for the role.
If we are delivering a digital workspace to the user we can use the in-built telemetry to analyse behaviour. This can be as simple as patterns of usage (times and locations) or more detailed analysis of interactions with application and data. This is an example of the use of the same data used to determine anomalous or potentially threatening behaviour.
Engagement and outcomes
Employee engagement is a major concern for most organisations, and measuring the extent to which an employee feels included, communicated with and valued requires regular, interactive communication and assessment.
We need finally to consider how to measure output or outcomes. Clearly this varies by role but, ultimately, we need to identify some metric by which the effectiveness and productivity of the role can be judged. Ideally this should include some added-value element rather than being simplistic or transactional.
In the world of 1984, analytics would be used to identify malcontents and despatch them to Room 101, but the opportunity here is to do something much more positive. If we can work back from identifying the most productive users, can we use AI to identify those behaviours (both direct and indirect) that are linked to effectiveness?
Such analysis can identify the positive benefits of training, new ways of working or investment in new systems. It can also identify negative patterns of behaviour or user bias that prevent effectiveness.
Dangers and opportunities
This area has to be approached with caution, as there is always a temptation to use them for punitive purposes. But if the data can be suitably abstracted and anonymised, then we have the opportunity to provide very relevant and specific advice to all those in a particular role about how they can improve – and can accurately monitor that improvement. Finding an ‘exemplar’ and sharing the positive impacts of their way of working is by far the best way of driving long-term behavioural change.
Winston Smith was tormented by the equation 2+2=5, used to show that the state controlled ‘truth’, but in fact the ability to add business data to employee behaviour and performance data does just that – it creates a data set that can be readily analysed by AI, delivering a business value which is much more than the sum of its parts.
This article was first published by techUK.