About Site Analytics Co. About Site Analytics Co. About Site Analytics Co.
News articles either by us or about our services
 
About Site Analytics Co.
For Retailers
For Developers
For Municipalities
Our Clients
News/Press
Careers
Client Login
Contact Us
Copyright 2004 Site Analytics Co.

 

Site Selection Models For Retailers:
Not Just For Site Selection Anymore


By: Adam Epstein
Reprinted with permission by the publisher from July, 1999

 
Being a real estate director for a national chain retailer can be fairly disheartening these days. Oftentimes, driving around a city for days only yields that each corner of every “Main & Main” is already taken.

As competition intensifies, the best spots dwindle away — and the cost of picking a “failure” site grows — innovative retailers are forced to find new ways of ensuring each of their sites will be winners.

Every retailer can point to a collection of both “slam-dunk” and “disappointing dog” stores in their portfolio. Why do some succeed while others don’t meet expectations? There may be thousands of possible factors which influence sales, but for each concept there are 20-50 unique critical success factors which really drive results. New technologies are enabling savvy retailers to dig through mountains of data to truly uncover what are the reasons why sites succeed. 

Armed with that enhanced understanding of their retail concept, chains are building statistical models which help them improve their hit rates significantly. The models, when used as a supplement to existing site selection techniques —and paired with the expertise of the real estate director — can help a chain grow more profitably:

  • The models can sharply limit the number of “mistake” stores a chain builds, thereby saving potentially millions of dollars in closing costs and headaches.
  • Models can boost the average sales at all new sites which they open.
  • They can be used by several departments within the firm to make their jobs easier too.

For these reasons, retailers have begun using these statistical models as tools.

“I’m a strong believer in models of this sort and have watched their sophistication improve over the last ten years, so that you really can’t live without them,” says Dan Clark, CFO of California Pizza Kitchen. “Not only did we build them once, but we updated them later based on further learnings.”

Yet there remains an air of mystery about models. They may seem like complicated “black boxes” which spit out conclusions through indecipherable methods. And it can be difficult to find retailers willing to share details about their practices. One real estate director at a prominent national chain declined to be interviewed for this article, pleading “I don’t want to give you any information about how successful our site model is because I don’t want to give any incentive for our competitors to go out and have one built too.”

Site selection models essentially add a touch of science to the subtle art of real estate; they become one additional tool in the chain’s arsenal of weapons. They aren’t correct in their sales predictions 100 percent of the time, but they can approach 90 percent accuracy levels. So they tell powerful, compelling stories about several aspects of a chain’s performance.

A complete model-building effort consists of two main components: Uncovering the critical success factors which lead to high sales or failure, and building a model based on these factors.

 1) Uncovering the critical success factors

Most firms think they know their criteria for success (e.g., “75,000 residents within 3 miles, HHI $40K+, and Median age = 32”). But often those goals have been derived only anecdotally rather than through rigorous proofing. They may become outdated as the chain grows. And it can be tricky to make clear decisions based on them. What if a site meets only five of the seven criteria? Is that good enough? 

A more helpful approach is to start with a blank slate and then perform a thorough analysis of the existing stores to find commonalties within the slam dunk stores and within the disappointing dogs. A chain can be surprised to learn what they thought was important turns out to be only vaguely relevant, or that previously overlooked factors can be crucial.

Allyn Taylor, currently Real Estate Manager at Extended Stay America, Inc., helped create Brinker International, Inc.’s models when he was in their real estate department.

“In many ways [building the model] was a good objective process to go through,” he says, “because it helped us look at things in another light, perhaps find things we hadn’t seen before. It was just a real good, analytical, check against our decision process.”

When examining the existing stores, a retailer can’t simply lump all of their sites into one analysis. The factors which drive sales are different at a super-regional shopping center than at an office park or a residential neighborhood. So they must segment existing units into appropriate groups and build separate models for each.

Pier 1 Imports has created multiple models they find useful. Rick Blackwelder, Pier 1’s vice president of real estate and development, says that analyzing basic demographics gave inconclusive answers, so a key was to “start looking at customer profiling and market segmentation. We broke our 700 stores into three basic categories. We found out the makeup in the trade area around our existing stores and really came up with a great model that we use, which has been very very helpful to us.”

To maximize efficiency, the retailer should compile information about every conceivable factor which could influence sales. PC-based GIS systems enable them to call up several thousand demographic variables for existing sites at the touch of a button. Retailers then look for patterns between those data and the sales figures. But they can’t just stop there — retailers need to test variables of all four types:

  • Demographic (e.g., number of residents, income, housing values, lifesyle clustering)
  • Business climate (e.g., competitive information, daytime pop, and types of businesses in the area)
  • Site specific (e.g., square footage, ease of access, location within the shopping center)
  • Operational (they may pick a terrific shopping center, but if the store is poorly managed then all the good site work is wasted).

Statistical analyses narrow down this list of 1,000+ variables to those which are truly the most important for each concept.

 2) Building the models

Once a retailer has fully evaluated what makes their concept tick, they weave those learnings into a predictive tool. There are four major types of statistical models; some solutions may integrate the results from more than one of these methods:

  • Analog models rate potential sites on a numeric scale. That rating is then compared to existing sites with similar scores — the sales at the new site should be analogous to those at the existing units.
  • Regression models are equations which assign weights to site data to forecast a sales figure directly.
  • Gravity models are spatial analyses which focus on the number, location, and drawing power of competitors in the region.
  • Neural network models have historically been used for pattern recognition applications such as fingerprint matching. But firms such as Neilthall Associates are now applying this technology in retail not only to inventory replenishment systems, but also to site selection.

The end result of any model-building process is a series of equations which the retailer can run easily — when coupled with data — to help predict sales.

Once the models are in place, the retailer may be astounded at how useful they are not just for real estate purposes, but for several different departments within the firm. Senior executives, therefore, often initiate the model-building process. Allyn Taylor says about his experience at Brinker: “Not only was (Norman Brinker) one of the model’s biggest supporters but he was probably one of the biggest instigators. He was constantly wanting to learn from past experience. One bad restaurant can suck the profits away from five to ten good stores, so it’s just critical to eliminate those mistakes.”

Benefits throughout the organization include:

  1. Site Selection. At their basic level, models help chains decide between two seemingly-equivalent sites.
  2. Market Selection. Armed with a sales forecasting model, retailers scan an entire city at once to find the likely hot pockets for their concept. Once they have determined how many units a city can support and what its predicted sales are, they compare this information across cities to find the best ones for their future expansion. “We’ve looked at some markets and intuitively said ‘this just doesn’t feel right, we probably shouldn’t be here’,” says Bob Goehle, vice president of real estate at Staples, Inc. “But then the model supports otherwise,” and the site proves successful. Pier 1’s Blackwelder agrees. “There are some markets that we thought of (entering) but now have backed away from based on this model as well,” he says.
  3. Marketing. By refining the customer profile, the chain can create detailed maps to find the buyers. “Knowing who our customer is and where they live is far and away the most important thing we’ve gained from this (model),” says Blackwelder. “It’s helped us in our advertising approach —we do monthly insert mailings and we use this (model) to help target where those mailings go.” Marketers also appreciate the higher “hit rate” the real estate departments achieve because they no longer have to devote so much time to “cleaning up” as many “mistake” sites.
  4. Human Resources and Operations. The sales models also “benchmark” existing units. By comparing actual revenues to those predicted by the model, retailers highlight which stores are underperforming expectations — possibly due to subpar management. CPK had one such unit, and as Dan Clark says, “When we strengthened the management team, the awareness and sales (moved up to where the model had expected).” Conversely, if a store is exceeding model predictions, they may have discovered a new Operations techniques which other stores in the chain can copy.
  5. Strategic Planning. Higher hit rates permit a chain to grow faster. Also, knowing how many units are feasible in each territory promotes efficiency in the budgeting and long-range forecasting process.
  6. Wall Street. Stock analysts love models and carefully laid-out development plans. When a CEO tells the investment community about the success of their model, it can set the groundwork for continued confidence in the growth plans for the company. And for private companies, having this type of analytic system in place can help secure loans and private investments.
  7. Mergers & Acquisitions. When a retailer first eyes a potential merger partner, they may think that the target has dozens of attractive sites which are performing well currently. But will those same sites continue to perform well when converted to a different concept? By plugging those sites into their own site selection model, a retailer can pick its partners more carefully.

Site selection models can be built in-house, or there are now consulting firms with different specializations to guide retailers through this process. The price ranges from a one-time cost of $20,000 to an annual fee well above $100,000. But as Blackwelder says, “the cost per store is really minimal. And consider that one bad choice or one right choice pays for the whole thing for a year.”

Because these tools can be so helpful throughout an organization, the savvy real estate director enlists the support for a model-building project from colleagues in other departments. They can spread the cost of the project as well as the benefits. As Bob Goehle at Staples says, “Our model is really shared by every department in the company, and our research department gets pulled in a lot of different directions” to make use of the models.

With the models in place, retailers will not just pick better sites, but they will also outsmart their rivals.
 

[Back to News/Press]