What is Pay Inference?
Pay Inference is Aon's proprietary Artificial Intelligence engine for Compensation.
It has been developed by our in house Research and Development team, to provide powerful insights to our clients on a range of compensation topics.
Pay Inference is a future state compensation offering, leveraging recent advances in Artificial Intelligence. It is created using deep learning neural networks trained on our entire compensation database.
Pay Inference is available through the Location Analytics tool on the Radford Network. If you would like to discuss a custom requirement for support from Pay Inference, please reach out to your Aon representative for further support.
How is Pay Inference trained?
An outline of the creation process is given below:
- Our neural networks contain millions of tuneable parameters - these are at first randomly initiated.
- We choose a neural network architecture designed to capture the nuances of pay drivers, and how they interact.
- The training process iterates over all data within our compensation database (tens of millions of data points), each time processing the inputs into pay predictions. After each iteration, the model predictions are compared to the true pay values, and minute adjustments are made to the model parameters.
- After enough iterations, the end result is a calibrated model of pay which not only understands the impact of job, level, location, and industry, but also captures nuanced interactions between these variables to produce the most accurate and complete model of pay possible today.
How does Pay Inference make predictions?
An outline of the runtime prediction process is given below:
- The trained model returns pay statistics for a given set of job, level, location, and industry inputs.
- Because this model understands the drivers of pay and how they interact, it performs well not only when replicating existing benchmarks (medians) but is also able to extend seamlessly to new combinations of inputs.
- This makes AI Pay inference ideal for estimating statistically reliable pay predictions for jobs and levels which are new to countries or territories, or in places where there is insufficient data to produce a meaningful traditional benchmark.
What technology powers AI Pay Inference?
Pay Inference has been developed by Aon’s Research and Development team and is powered by cutting-edge deep learning neural network technology (PyTorch),.
Will Pay Inference / AI replace traditional compensation benchmarks?
Compensation benchmarks remain the gold standard for understanding the true pay levels in established markets and other tools in the Radford Network (e.g. Quick Benchmarks, Market Queries) should remain your primary source of benchmark numbers.
Pay Inference augments this and helps us to predict and understand pay in new territories and in locations where a smaller range of jobs have data available.