How Artificial Intelligence Improves Workforce Management
May 16, 2023
by Farshad Kheiri, PhD
Legion helps businesses maximize labor efficiency and employee engagement simultaneously with workforce management software that’s intelligent, automated, and employee-centric. To maximize labor efficiency, we identify the precise amount of labor that’s required to meet demand.
Why Use Artificial Intelligence (AI) for Forecasting?
Traditionally, in a retail store, managers use the previous week’s demand to predict what will happen in the following weeks. This process is called “random walk” (which is different from the famous “random walk” as a random process in probability and statistics).
More advanced store managers use statistical techniques, such as moving averages, e.g., next Monday’s traffic will be the average of the last six weeks of traffic on Mondays.
Other store managers use a more complex process and even more advanced techniques like ARIMA (Auto Regressive Integrated Moving Average) or SARIMA (Seasonal ARIMA), which uses a linear combination of historical data to forecast the future.
Why is AI Better?
In recent years, machine learning techniques have been used extensively for many problems requiring automation.
In demand forecasting, machine learning (ML) can be applied to enable managers to
- capture more complex patterns over extended periods
- use more input data from different sources
- update as the data changes
As a result, ML techniques generate more accurate forecasting.
What data is needed?
Even though the Legion AI platform leverages many different kinds of data points, all that is needed from customers are demand drivers they want to forecast, such as traffic, sales, and transactions. Customers need to provide this data at the granularity they want to be forecasted – for example, 15-minute, hourly, daily, weekly, or monthly intervals.
In addition to the historical data for the specific driver that needs to be forecasted, the customer can provide the special events that they think will impact demand, such as promotions, sales, etc. Legion can include those events in its AI engine to improve the accuracy of the forecasts.
How forecasts are generated and why accuracy is higher
Sales, traffic, and transactions are driven by external inputs, such as the economy, weather, an event happening close by, etc. ML can consume these data points along with the demand at hand and generate more accurate forecasts because it can see how these events impact the historical data. Legion WFM uses hundreds of machine learning and statistical models and selects the optimal forecasting ensemble model.
But what if a new location opens or management starts to gather new data in an existing location? There are different techniques to generate synthetic data; each has benefits and drawbacks depending on the application. Data scientists have extensively studied data augmentation and synthetic data generation and have presented their findings in various literature. However, Legion developed its own patented algorithm that outperforms conventional techniques.
Data integrity plays an essential role in generating demand forecasts. In recent years, many workforce management vendors have focused on ML but not on the data integrity and quality feeding these models (“garbage in, garbage out”).
To ensure ML works as expected, it’s critical to monitor the input data to ensure accuracy and that it hasn’t been tampered with randomly or systematically through an integration failure or clerical mistake.
Data integrity can be examined and tested using many different algorithms, which should be a major component in any ML platform.
One way to monitor data integrity is to check the input data distribution over time and track any changes or abnormalities.
Legion helps businesses maximize labor efficiency and employee engagement simultaneously with workforce management software that’s intelligent, automated, and employee-centric. To optimize labor efficiency, we identify the precise amount of labor required to meet demand. A company’s demand is defined as the number of customers, sales, transactions, etc.
To determine the necessary labor, we need to identify the tasks that hourly workers will perform. In almost all hourly jobs, tasks are created based on demand. Hourly workers perform the tasks to serve that demand. Therefore, we must have a clear idea of the demand. Demand forecasting is the first step for any workforce management platform that has a one-to-one relationship with demand. An accurate demand forecast ensures the optimal published schedule.
How Accuracy of Forecasts Are Measured and How Can They be Improved?
Legion measures the accuracy of the forecasting models using weighted forecast accuracy, which is a modified version of Mean Absolute Percentage Error (MAPE) and Seasonal Mean Absolute Percentage Error (SMAPE). However, to monitor the models for our AI engine, we always measure the accuracy versus the most advanced statistical technique to ensure that the AI engine surpasses any other technique. By running model training every week, the most recent data is incorporated so the models’ robustness is guaranteed. Also, by having a monitoring system in place, the accuracy of the models is being reviewed constantly, and as a result, the accuracy is stable.
In short, by continuous training (weekly training or less if a monitoring alert gets triggered), model accuracy improves or stays consistent depending on the drivers’ predictability. By the end of the day, the accuracy is constrained by the randomness and predictability of the data; however Legion’s AI engine has been designed to ensure we always stay above or equal to the expected predictability of the data.
What if you Disagree with the Forecast?
Finally, although AI automates many repetitive and exhausting tasks, such as how the weather impacts traffic or sales and how a nearby sporting event can change demand, it should still allow for human intervention. Human intelligence can be captured and added to improve the AI. Techniques like online learning can help train the AI engine as it gets user feedback.
Even when the AI engine has captured a lot of external data about the business, it may have missed some unforeseen issues, such as an accident that has caused a road closure or a health inspection that closed the store. The AI engine should capture these events and learn how they impact the forecast to improve future estimates. Any AI engine working closely with a human operator must be able to gather feedback. When human intelligence enables artificial intelligence, forecasts are even more accurate. Legion’s AI engine empowers users to give feedback and rectify the results as it sees fit. This feedback is weighted against what actually happens and weighted accordingly in the future.
These are some of the key aspects of Legion’s AI capabilities, which enable intelligent automation to improve business agility, accuracy, and labor efficiency – automatically. Demand forecasting is required to build an optimal schedule based on business needs and employee skills and preferences.
What Does AI Improve ROI?
We have shown how every percent increase in demand forecasting accuracy translates to a 0.5 percent decrease in labor costs and a factor of 0.25 percent increase in sales depending on the conversion rate of the traffic that translates into sales.