Simulating risk scenarios across space and time.
Many businesses are struggling with the dilemma of reopening their sites and trying to return to normal levels of operation, without compromising the safety - and privacy - of staff and customers.
Parallel was commissioned by Sightline Innovation to help them develop a vision for Safe at Work - a product that uses building telemetry data to provide digital twins of workplaces to calculate COVID-19 exposure risk.
Hoping that things will be okay is clearly not a viable approach. If measures are not taken to reduce the risk of infection, businesses that do reopen are likely to end up in a vicious circle of reduced operations and reputational damage amongst risk averse customers.
But any approach that is too cautious does not benefit businesses either. If all staff were sent home when one gets sick, it would reduce their risk, but effectively shut down the firm.
A more nuanced approach is needed that equips businesses with insights about the specific interventions that will be most effective in reducing risk for them.
Given the global scale of the COVID-19 threat, it was not surprising that there were already a number of solutions on the market, competing for the attention of operations leaders looking to re-open facilities with increased confidence.
We carried out extensive competitor research and analysis across a broad spectrum of levels of sophistication. This led to the identification of four product categories, which in turn helped us to evaluate where Safe at Work should play and how it could win.
Many of these competing solutions appeared to have been built rapidly, often re-purposing an existing offering, for example in tracking for logistics or professional sports. These providers tended to have long-standing products in areas such as 3D environments and location analytics – putting Safe at Work at a disadvantage.
However, given Sightline Innovation's credentials in the domain of Data Trusts, there were two key competitive factors that provided an opportunity for differentiation: enhanced privacy and probabilistic risk modelling.
These factors became key components for an enhanced value proposition for Safe at Work and provided a set of principles for developing a vision for the product.
Product vision – Physical risk
In order to provide a more granular level of risk insight, it would be important to provide companies with some level of ability to map the specifics of their facilities.
Different areas within a site will have varying risk profiles and this information will be material in understanding how best to mitigate risk, beyond two individuals simply being in proximity to each other. For example, two people could be either side of a partition wall, but tracking signals could suggest that they were in direct contact.
Given that staff were yet to return to work, there would be no existing location analytics data available.
One approach to resolving this could be to use the triangulation from installed beacons, to encode the physical space into zones. A survey team could assess the levels of different types of risk in each zone - either remotely or by walking the facility and inputting data on the move.
This activity would provide a valuable baseline that can be tracked over time and compared with ongoing telemetry from individuals once they are back at work.
Product vision – Behaviour risk
In an ideal situation, staff would enter a facility in an orderly and socially distanced way, making their way to a workstation with minimal contact with team members.
Normal patterns of movement / behaviour are much less controlled and consequently more likely to increase risk.
By using bluetooth beacons or UWB devices attached on or near objects, an accurate picture of employee behaviour patterns and proximity events can be observed and analysed.
This approach could help prioritise cleaning efforts and frequency and highlight to operations teams the areas within facilities that are likely risk hotspots.
For an entire facility - or a smaller zone within it - anonymised aggregations of proximity events could be observed - from brief passing contacts that represent little risk, to riskier sustained contact and the use throughout the day of shared objects, such as tools.
With staff movements modelled using actual data or synthetic simulations, it becomes possible to model how behaviour affects the level of risk within a space.
Taking inspiration from Nathan Yau’s A Day in the Life of Americans alternative visualisation approaches could be explored that provide a level of abstraction that anonymises the movement patterns of individuals - instead aggregating them by teams, zones, floors etc.
Product vision – Contact risk
The original vision of the Safe at Work product was devised around a graph database - and it was assumed that a graph visualisation would be the best way to indicate if contact had occurred between people.
However, we felt that there were a number of issues with this approach. Not only can these visualisations quickly become very dense and unwieldy, but given the time sensitive nature of the problem - isolating people who may be infected - it was important for operations teams to view lists of people they need to isolate, test or inform, rather than picking over a complex visualisation.
Instead - for reasons of both cognitive load and privacy - we proposed a simpler, timeline based view to help provide context and rationale behind any actions to be taken and to give an at-a-glance view of the scale of the issue.
This contact visualisation could be toggled between unprocessed events / people and those that have already been dealt with / resolved.
Lists of proximity based contacts could be organised according to a company’s policies, making use of the underlying data that specifies the duration and frequency of contacts (without overloading users with too much information upfront).
Privacy was identified as a key differentiator for the Safe at Work product, but there is clearly a tension between this and needing to know exactly who has been in contact with someone who tested positive.
Our approach to this was two fold - firstly any contacts or proximity events, older than 28 days were anonymised. Secondly, only those contacts that were measured as sustained would be shown within the system. This was also central to the ability of firms to balance risk with operational capacity.