Why you’re probably doing labour forecasting wrong

With the adoption of systems like digital point of sale and traffic counters, companies have more data than ever to keep track of exactly who is buying what and when. But the reality is that most retailers and restaurants are not making effective use of this data—especially when it comes to its usefulness when forecasting demand to plan the right level of staffing.

Not using all the data you have at your disposal to meet demand with labour means that you could be doing forecasting wrong. This can lead to understaffing at critical times, disgruntled employees, unsatisfied customers, and tens of thousands of dollars in unrealised gains. Furthermore, as the options for how a customer can interact with a business continue to grow (in person, on third-party apps, by calling in, etc.), forecasting demand and matching labour effectively is becoming more complicated.

The right tools for demand forecasting

Getting forecasting right can be easy with the right tools and approaches in place.  Traditionally, forecasts are based on expected and historical sales revenue, seasonality, or other high-level factors. The problem is that revenue-based forecasting, or forecasting as a flat percentage of total projected sales will yield inaccurate forecasts that will hinder your ability to meet demand with labour.

In order to do an accurate forecast you have to include the actual drivers of demand. This brings us back to all that data that’s sitting unused by so many businesses. Technology makes it possible to take into consideration large amounts of data, already at your disposal, to forecast demand accurately, always. From item-level sales and foot traffic to real-time external drivers of demand like local events and weather, you can consistently approximate what demand will be at any given time with an astonishing accuracy of over 95% if you use the right tools.

Examples abound. For instance, sporting events can drive alcohol sales for a restaurant, so it would be necessary to be able to identify this as one of the drivers of demand and staff bartenders accordingly when an event is happening (as opposed to, for example, hostesses and servers). Also, because foot traffic forecasted at an hourly level is a better insight into labour than sales, retailers can use that info to maximise conversion opportunities and ensure that there are enough people at the front of the house. At a QSR for example, productivity for a sandwich compared to a pizza might be very different, so driving staffing decisions based on total sales dilutes all that useful granularity. Machine learning makes it possible to capture all of this nuance and act on it.

The fourth generation of labour optimisation software

Running predictive drivers such as foot traffic, events and even weather, through a productivity model -which identifies how much work each employee can accomplish within a given shift- gives a much more specific and accurate labour forecast. The model can indicate the number of drinks that can be served per 30-minute increment—and factor in fluctuations based on sporting events or promotions.

This might sound overly complex. And it would be, if we were still only using third-generation tools for labour optimisation. Specifically, these older tools are software that mimics manual practice – making it easier but not necessarily more effective. These tools also might add mobile capabilities, but the decisions continue to be manual input based on estimates from managers. These tools typically find it tricky to account for variability as one moves from daily to hourly data. According to Estimating the impact of understaffing on sales and profitability in retail stores study, the coefficient of variation—that is, how variable traffic is—increases from 10% to 31% when one makes that shift to smaller time increments. This means you have to make that leap if you want to produce optimal labour schedules consistently.

In order to schedule labour at small increments, you also have to understand demand at small increments. Machine learning makes this possible. This is one of the key distinctions between the third and fourth generation labour management software. The fourth generation of labour optimisation software, driven by Legion, actually makes it possible to predict, manage, and take action.

Meet Legion: The demand-ready labour platform

At Legion, we build forecasts using machine learning models that learn patterns from all data, and then predict future values based on these patterns. Legion forecasting models look beyond the last few weeks of data to learn annual, weekly, and hourly patterns.  And in addition to your own internal historicals, Legion enables you to include the impact of holidays, nearby events like sports games or concerts, or local weather that can all affect your business. Taking all this data into account is nearly impossible for a person to do on their own but not for machine learning.

We integrate directly with major traffic counters, POS, and other business-critical software to create custom, accurate labour forecasts and schedules. Here’s how we put it all together:

1. Data aggregation: Legion aggregates data from your traffic, POS, and internal reporting systems.
2. Data augmentation: Legion augments your data with externals, like local events and weather.
3. Feature evaluation: Legion’s ML technology auto-selects and weights the parameters that have the most impact on your forecast.
4. Model selection: When it comes to demand forecasting, you can’t take a one-size-fits-all model approach. We use a group of more than 50 models and select the best-performing ML algorithm from among them.
5. Model training: Machine learning ensure the system is continuously learning and applying feedback. So Legion automatically re-trains the model each week as new data becomes available.

With Legion, everything is automated. From model selection to feature correlation to weekly retraining/improvement and on-demand forecast generation. When you put it all together, it results in demand forecasts—and schedules to match demand with labour—that are accurate, granular, and easy to use, without compromising on employee happiness. By taking this approach, businesses can save around $35,000 per location (average roster size of 30 employees). This enables our customers to rethink labour deployment to precise drivers and serve customers better — no manual configuration and forecast adjustments required.

Learn how Legion’s ML based forecasting can help you optimise your labour spend.

Book Demo