Challenges:

People cost forecasting is traditionally a time-consuming procedure involving numerous Excel templates. Payroll, benefits, travel costs, PTO accruals, and other accounts are all factored into the people cost forecasting process. Furthermore, there are several moving pieces, such as employee attrition, temporary employment, new hires, and so on. To make projections, all of these must be merged in an excel template.

Input data:

We must first prepare our Input data file before we can utilize Machine learning to analyze multiple data points and estimate the expense. Historical data for spending accounts, drivers, and future projections should be included in input data files. Let’s take a look at expense accounts first.

Expense accounts:

First, expense data from ERPs must be reviewed, which includes the history of various accounts such as salary, wages, taxes, PTO, and benefits. The granularity of data must be determined in relation to the output required. You must include division level data in your input file if you wish to predict at the division level.

Drivers:

The majority of the people’s costs comes from the existing headcount. However, new recruitment and attrition account for the variance from month to month and quarter to quarter. As a result, you should include the following data elements from your HR systems.

• Last headcount
• Incremental headcount
• PTO accruals
• Health insurance costs
• Pending job requisitions, etc.

These factors will act as drivers for the expense prediction

External drivers:

We all know that the economy, and thus the employment market, has an impact on hiring. We can use GDP as one of the driver. Also we can consider the pending job postings for similar jobs in a Job site.  This can tell us how hot the job market is and how long it takes to hire the right person.

Prediction:

We can use several machine learning methods to analyze and anticipate the expense once the input data is ready. You can hire a data science team or utilize a tool like Tadaa.ai, which analyzes data and generates forecasts automatically. Machine learning models like XGBOOST, Light GBM works well with time series tabular data .

Results:

Because of the underlying assumptions, human-generated projections always lean to the baseline. When machine learning is applied to this data collection, however, it blends baseline, seasonality, and trends data from the history of expense accounts with the influence of numerous internal and external causes. The forecast numbers that are generated from an ML model are more accurate than those created by humans. Depending on the needs of the organization, more drivers can be added to fine tune the accuracy.

If you want to know how this can be implemented feel free to contact us.