What are Smart Differentials and how are they calculated?

What are Smart Differentials?

Smart Differentials provide like-for-like compensation cost comparisons between locations.

Key features are:

1. Assessment of all jobs and levels selected in each location, irrespective of underlying data availability*.
2. A normalized calculation, which normalizes over all jobs and levels without being impacted by the headcount in each location.

This is achieved using Aon's proprietary Pay Inference engine, which leverages state-of-the-art AI neural networks designed to capture a global understanding of compensation markets.

Similar to benchmark differentials, Smart Differentials are presented as a percentage difference relative to an anchor (reference) location which a user selects. For example, France can be compared to New York for all HR jobs.

Unlike benchmark differentials, Smart Differentials are normalized to account for differences in job, level, and industry distributions between locations. Because they are based on powerful and generalizable neural networks, they are also less affected when data availability is low. This helps with filter selections with less common jobs or locations with low headcount, such as emerging markets.

* A small number of jobs/levels/locations may still be excluded as part of our model governance framework. This governs where the model is expected to perform well, and where the data available for the model to learn from is too limited.

How are Smart Differentials calculated?

The following methodology is applied:

• Find all valid combinations of  job and level within the selected filters.
• For each valid job combination and each location:
• Use AI Pay Inference to infer pay at the selected location and the reference location.
• Calculate the percentage difference between the selected and reference locations for that job combination. (Sub-differential in the table below)
• Finally, for each location average across the sub-differentials, to return a normalized Smart Differential.

*When we calculate a Smart Differential, every job, level, and industry in a filter selection is counted and balanced equally, regardless of the headcounts found within your selected locations. This is made possible using our AI models, which can accurately infer pay for all jobs, levels, industries, and locations, even when we have little or no data available for that selection in that location.  

Example: What is the Smart Differential for Data Scientists and Data Analysts (Level Category = Professional) in London, relative to San Francisco?

There are 12 valid combinations of job and level. Therefore, we must infer pay for 12 combinations for each location, and then calculate the average of 12 sub-differentials
 

Job	Level	Inferred Pay London	Inferred Pay San Francisco	Sub-Differential (SF vs London)
Data Scientist	P1	50k	110k	-54%
Data Scientist	P2	66k	135k	-51%
Data Scientist	P3	84k	150k	-44%
Data Scientist	P4	108k	180k	-40%
Data Scientist	P5	121k	200k	-39%
Data Scientist	P6	140k	234k	-40%
Data Analyst	P1	35k	82k	-57%
Data Analyst	P2	51k	110k	-54%
Data Analyst	P3	64k	125k	-49%
Data Analyst	P4	78k	140k	-44%
Data Analyst	P5	90k	153k	-41%
Data Analyst	P6	98k	170k	-42%
Smart Differential				-45%

This result:

• Is not affected by the proportion of headcount between Data Scientists and Data Analysts in London or San Francisco.
• Is not affected by whether there is any data for these jobs in these locations.

This means Smart Differentials are not just ideal for providing a normalized comparison between locations, but also for calculating comparisons for jobs and locations with very little data, such as new or emerging job markets.

How are different jobs weighted in Smart Differentials?

Because there may be no data for certain jobs in some of the locations you have selected, we do not apply any headcount weightings to the different jobs. 

As such, each job is weighted equally in the Smart Differential calculation.

 What adjustments are applied to Smart Differentials?

Smart Differentials normalize over all job, level and industry filter selections. To understand why we do this, consider the example below:
 
The charts below show the US distribution of headcount by Level for a basket of roles in New York and Maine:


Notice that Maine has no headcount for the higher-paid Data Scientist AND fewer Product Managers.  Whereas New York has representation for all roles.

Even if New York and Maine paid the exact same salaries per role, the overall average salary in New York would be higher than that of Maine. This is because New York has more headcount in the higher paid roles.

What data does Smart Differentials use to make predictions?

Smart Differentials are powered by Pay Inference, which has been trained with many millions of rows in incumbent compensation data – the same data used to provide benchmark reports to our clients. Pay Inference models are able to understand the complex array of factors that influence pay markets for different jobs in different markets.

In particular, when calculating a Smart Differential, the AI will have learned from:

1. Exact matches for the selected job(s) and level(s), in the selected location.
2. Data for the selected job(s) and level(s) in the country (where a micro region has been selected).
3. Data for similar jobs in the selected location.
4. Data for similar / ‘close’ levels in the selected location.
5. Data for the selected job(s) in other locations.
6. Data for the selected level(s) in other locations.

 
This means that while there may be limited or no data that is an exact match, the product is still able to provide differentials that reflect the expected cost differentials in that location.

What does prediction support mean?

To help users understand more about the Smart Differential data they are using, hovering the mouse over a selected location provides a summary of the range of the data available to the models when inferring compensation trends about that location.  This is called prediction support.


Example tooltip showing prediction support for Bangalore, for a Smart Differential with "Life Sciences" selected in the filters.

What if there is no data available for the job I’m looking for in a selected location?

If there is no data for the exact job(s) and level(s) in the selected location, then Smart Differentials will still attempt to provide a cost comparison for that location. 

This will be based on what the AI Pay Inference models have learned from supporting data (on related jobs, levels, and locations) in order to provide a differential that reflects the expected cost differentials in that location.

What if there is lots of data available for the job I’m looking for in a selected location?

Where there is good data coverage available for the job(s) and level(s) selected, Smart Differentials are still available. Due to the way the Artificial Intelligence (AI) works and has been developed, Smart Differentials work well in areas of good data coverage and so should provide a good differential.

Note - this number will not be identical to a Differential calculated only from the “exact match” data available, as it is normalized across all of the jobs and levels selected.

When would I use Smart Differentials instead of benchmark differentials?

Smart Differentials provide normalized pay differentials with greater global coverage than is possible from looking at the data for each location alone.

As such, Smart Differentials allow a greater focus on the difference in pay which is based on location alone and reduces the effect of by other conflating factors.

They also provide previously unavailable insight into emerging markets, where limited data is available, or for new or priority jobs, for which data may have limited global coverage.

Does Smart Differentials work for all jobs in all locations?

No.  Depending on the compensation metric you have selected, Smart Differentials may not be available for some level categories (Executive, Professional, Management, Support and Technical).

As part of our model governance framework, the algorithm measures the data support for each prediction, to determine whether the prediction should be provided.

If a prediction is not available for your request, you will see a message like this in the Smart Differential tooltip:

Example showing the tooltip message when attempting to view Equity Smart Differentials for the Executive level, in Costa Rica. 

If only part of a request is available, then Smart Differentials will still be provided and a warning message will indicate the levels that have been excluded.