Inaccurate predictive models inevitably lead to bad real estate investments. Every year, retailers close stores that would’ve stayed profitable, prioritize stores that won’t pan out, miss viable opportunities, and cannibalize excessive sales from their other stores—all because their site models weren’t accurate.
The whole point of site selection software is to inform and validate your decisions. But if your software can’t produce models that accurately forecast sales and represent the choices before you, it can easily do more harm than good.
A poor predictive model makes your software about as reliable as intuition—maybe even less so. You’re not really rooting your choices in sound data.
Here’s how inaccurate (or “less accurate”) models cause problems now and down the road, plus what it takes to get predictive models right.
Inaccurate models lead to bad decisions and missed opportunities
Your predictive models can go wrong in three main ways. Each of them can lead you to prioritize locations you should’ve ruled out or miss prime spots you should’ve snatched up.
Most site selection software doesn’t incorporate enough relevant variables or use algorithms that fit your business, resulting in sales forecasts that are too high or too low.
For example, if your model doesn’t consider or properly weigh site characteristics that affect how easily people can see or access your store, there’s no way your sales forecasts can accurately reflect the store’s ability to capitalize on nearby demand.
Similarly, some locations may have specialized equipment, configurations, employees, or inventory that affects their services and revenue—your model and forecast need to include that. If it doesn’t, you’ll be basing your forecast on a model that doesn’t actually represent the location you’re considering.
When your sales forecast is too low, you could pass up a good location or wind up with a store that doesn’t have enough capacity or product to keep up with demand. The low forecast will lead you to treat the location as a smaller opportunity than it really is. And that could take years to correct.
When your sales forecast is too high, you could trigger a “go” decision when you should’ve made a “no-go” decision. Or, you might over build, leaving you with higher rent, too much merchandise, overstated revenue, and/or lower margins. Not to mention the opportunity cost of prioritizing these locations ahead of others that may have performed much better.
Say you’re trying to add stores in underserved markets or fill gaps between current stores. If your software can’t accurately represent infill spacing and uses a cookie-cutter approach, your cannibalization estimates will always be too aggressive or too conservative.
When your sales cannibalization estimates are too aggressive, you get:
- Too many new stores
- Unachievable net new sales
- Overstated development plans
- Overstated accretive revenue growth
When your cannibalization estimates are too conservative, you miss opportunities by passing up potential infill stores, leaving excellent locations available to your competitors.
Every store and every trade area is unique—so if you use a fixed radius (distance or drive time) for them all, you won’t see your real gaps in serviceable area. But unfortunately, that’s the only method available with many site selection software vendors.
Custom drawing your serviceable areas helps, letting you use insights and experience to estimate the actual opportunity each location presents. But if you want reliable infill spacing, you’ll need mobile movement data to see where each store’s customers come from and how they move throughout the trade area.
Your brick-and-mortar locations have a direct, tangible impact on nearby ecommerce sales. But a lot of software solutions struggle to represent this relationship. This can lead you to close stores that should’ve stayed open (because they were driving significant ecommerce sales) or overlook sites that would’ve generated enough overall sales to justify a store.
This isn’t rocket science. We use your historical real estate transactions and localized ecommerce sales data to understand the relationship between your stores and online sales. In the past, what has the impact been on nearby ecommerce sales when you open or close locations? Using these patterns, we predict the effect your decisions will have on ecommerce.
Most vendors use a single algorithm or modeling approach for all their customers, regardless of your industry and the characteristics of your stores and markets. But a fast food chain won’t have the same priorities as a grocery store or fitness center—so why would they weigh all their variables the same?
Tango Predictive Analytics selects a combination of up to 100 different algorithms, stacking them on top of one another to ensure your variables are properly weighted. Using machine learning, we run through thousands of combinations to find the model that fits your business best, leaving you with the most reliable predictions.
Want to prioritize your next real estate moves with confidence?
Request a demo of Tango Predictive Analytics.