A published algorithm for a radar-based hail indicator was found to correlate with hail damage, so it was applied to real-time 3D weather radar data to flag locations with potential hail. These locations were compared against geocoded customer policy data to identify possible damaged areas. Auto-generated email alerts went to the research team to then communicate the unfolding event with the wider company.

After extensive floods in 2011, insurance companies needed to start offering flood insurance, but could only do so if the risk was calculated individually per property. A team in Insurance Australia Group gathered hundreds of datasets from around the country and standardised calculations of flood risk across numerous methodologies. The end result was a modular system selecting the most accurate flood risk assessment at each address to enable policy pricing.

Many areas with flood risk do not have any formal flood studies to quantify the risk, and performing such hydraulic studies is very costly. A model was created to rapidly calculate a rough estimate of flood flow and height at each point along all watercourses in a catchment with minimal data. The estimate was based on assumed 100-yr and PMF rainfall events upstream plus sampled elevation data for cross sections, with smoothing of flood heights along the watercourse for sensible outcomes.

Natural peril event models usually utilise a severity parameter such as wind speed or hail size. Flood intensity is more difficult as it is highly location-specific and more usually specified by ARI. A synthetic flood event catalog was generated in each catchment by working backward from ARI to interpolate an aggregate loss amount. Realistic multi-catchment events were then generated using historic cross-catchment correlations to select from the catchment event catalogs.

To assist in internal operations, a dashboard of natural peril warnings was built that gathered real-time warnings from many agencies and hazards along with internally built alert sources. The warnings were categorised and displayed with corresponding colours to indicate likely property impacts.

Most weather maps are either static images with zero customisability or interactive 2D map tiles with fixed colour scales and zoom levels. None of these maps accurately illustrate the physical shape of Earth. Globe Weather Maps was built to show weather data on a 3D Earth representation with smooth zooming from whole Earth to city level. Because the data was rendered live, the map styling and colour scales could also easily be customised by the user.

OpenAgent needed to understand the comparative listing rates of their referrals to real estate agents across different cohorts of users. An algorithm was developed based on an observed profile of listing conversions by referral age. The profile could be applied to any cohort of referrals to count how many had converted to listings and thus project the ultimate listing rate of the cohort.

Based on historical agent sales performance and referral success, several factors were developed to select the best set of agents to present to users in order to maximise conversion to listings. The new approach was trialled with the sales team, requiring updates to employee tools and backend databases and APIs.

Automated home valuations can be based on numerous factors but often fail to consider subjective aspects especially condition and quality. A new method was developed to start with multi-variate regression on readily available data and add input from a user on comparative quality of the subject address versus other properties in the same area.

Mosaic is an interactive web map showing various urban planning spatial data layers across three states. Its aim is to make basic planning data easily viewable within seconds rather than checking numerous web sources and complex PDF maps. It was built using Mapbox maps and a custom interface that smoothly transitions across state boundaries. Backend data processing includes a weekly download and reformatting of government-published spatial data.

NSW land use definitions, of which there are nearly 200, are standardised in planning instruments but are defined in a complex system of groups and hierarchies. Permissibilty in each land use zone is specified with some selection of these terms which may be group or parent terms. To simplify assessment, a calculator was created to embody the definition logic, allowing for testing of individual terms and zones or bulk processing of all terms and zones for analysis.

For transport project business cases, a benefits calculation includes an analysis of future land use changes due to the project. By combining strategic planning knowledge, housing/employment market understanding, and urban design, scenarios were produced of granular spatial land use change, which were then quantified and reported with commentary. Specific future transport projects included Sydney Metro lines and NSW Fast Rail.

Land use forecasts are provided by NSW Govt for thousands of small areas across a 40-year time range. Outside of Greater Sydney these forecasts are based on a very simplistic extrapolation so do not accurately reflect local conditions or growth plans. A recalibration was done for 2021 and future years based on data between Census years and higher quality local population and employment forecasts.

An algorithm for forecasting visual quality of sunrise and sunset was created based on input cloud forecast data and light rays' geometry and optical properties. The algorithm was used to feed a publicly available app, Sun Chaser, with accompanying backend services created to allow users to login and unlock additional app features.