Photo: Getty Images
Is AI the key to finding the right location, location, location?
The conventional wisdom is that being in the right location is critical to success in retail. As one Japanese convenience store pursues an expansion, it may be getting some non-human help to decide where its stores should go.
Convenience store chain Lawson is considering using artificial intelligence (AI) to determine where to place its new store locations, according to the Japan Times. The chain plans to use AI to collect marketing data, such as household distribution patterns and traffic volume, to determine a given store’s chances of success in an area. Generally, the chain makes such decisions based on information gathering and analysis of an area carried out by staff. The chain is the third largest convenience store chain in Japan with 13,000 locations; 7-Eleven is first with more than 20,000 stores.
Using AI to crunch the numbers on demographics or habits could lead to some interesting and unforeseen strategic maneuvers, depending on what data points the chain folds into the calculation. But passing on locations that the AI doesn’t recognize as being profitable may have pitfalls, potentially ruling out otherwise good locations with extenuating factors that a human would recognize but an AI would overlook.
And there is also the question of the cost attached to such predictive AI technology. If an AI solution were to only pick locations that a human would select anyway, it might not be worth the investment.
Meanwhile in North America, convenience stores have been expanding, not through finding new places to build, but by acquiring existing locations.
Early this year, 7-Eleven acquired the 1,030 convenience stores owned by gas station chain Sunoco, bringing its total store count up to around 9,700 in the U.S. and Canada.
EG Group, out of the U.K., made a move into the U.S. convenience store market through the purchase of Kroger’s entire multi-banner convenience store operation. The company purchased the 762 stores for $2.15 billion, beating out a high-profile bid from U.S. convenience store Casey’s General Stores.
Discussion Questions
DISCUSSION QUESTIONS: Is using AI to sift through data like household distribution patterns and traffic volume becoming necessary for determining the best store locations? What risks do you see in using AI to replace the traditional location evaluations done by humans?
Poll
BrainTrust
Joanna Rutter
Marketing, Dor
Ricardo Belmar
Retail Transformation Thought Leader, Advisor, & Strategist
AI is being applied to a myriad of data sets to inform decisions and using it for location selection makes good sense. While the hope is that AI systems will be able to identify trends or patterns that traditional methods may not, there’s no guarantee AI will identify the best locations. AI is only as good as the quality of the data and the algorithms which are designed by humans. While there is great promise in what AI based analysis may reveal, human oversight should ultimately guide the final decisions.
It is smart and it will be the future. AI has limitations that need to be accounted for today; however, the scope of data AI can process and correlations it’s able to make are already far beyond what humans are capable of in any reasonable time. I think the best bet for retailers in the short term is for AI to lead the charge and humans to do a sanity check before pulling the trigger on new locations.
I think retailers can’t ever be too off the mark when driven by a desire to serve their customers better, especially when they’re taking a calculated but brave risk with new tech. Whatever it takes to optimize your organization’s operations and streamline customer service, I say! Plus, there’s a reason they’re called convenience stores — the more data-informed you can be about delivering on that convenience factor, including where you place stores, the better.
Using AI technology is fine, but we’re still a long way from it being wholly perfected and therefore it will not give 100 percent accurate information. There are still too many questions that only humans will be best suited to answer. We see this everywhere today where businesses are pushing AI technology because it’s cheap. But remember, you get what you pay for. Until AI can fully understand things like a person’s feelings and “honest opinions” humans must remain involved in the decision making process.
This is a classic AI problem, where the inputs are so varied that it is very difficult for a human to create a weighted analysis across so many dimensions of commerce, demographic and geographic data. Thirteen-thousand store data points is more than enough to create significant insight into large dimensions of data, as long as the data is structured well enough and normalized appropriately. Current techniques are more than sufficient.
Of course, one shouldn’t blindly go forth with AI without an analysis of its top recommendations. In fact, there can be remarkable learnings where AI makes recommendations that defy current understanding, providing insight that helps guide the future of the business. AI is clearly the path forward here.
At this stage of AI’s evolution, it’s too premature to solely become dependent on AI technologies to replace the traditional location evaluations. Certainly AI will enhance, optimize and improve the store location sourcing process, however there are other intuitive thought processes and instincts that should be considered instead of relying on data.
It’s another case of combining the art and sciences in retail. As AI technologies mature and become more operationalized within a retail organization, then there are clear advantages to be gained from sifting through the data. Another factor to consider is the corporate culture considerations, as retail organizations embrace, understand and leverage AI capabilities.
Data helps makes decisions. Having AI comb through more data than the mortal human can comprehend can be an advantage, as long as you know what you’re specifically looking for. For convenience stores, this is easy. Traffic patterns, demographics of consumers in the traffic, average spend, etc. are all good data points. For other types of retailers, this information can be good as well. Comparing to other locations, malls, etc. can be helpful. Putting science behind decisions is good business, but don’t let it completely take away from creative thinking and even a little “gut feeling.” New ideas and breakthroughs don’t happen when you follow what everyone else is doing.
We just wrote a benchmark report on location-based analytics and, to be honest, the findings shocked us. Almost half of surveyed retailers don’t use any GIS system to help them site stores or distribution systems, let alone infusing them with AI.
I think part of the problem is that we’re flooding the market with terms, technologies and concepts all at once. I don’t think retailers can keep up. So let’s go back to the basics:
Baby steps. I think we have to start back at baby steps.
AI is becoming a fundamental piece of any marketing automation and prediction model. So it only makes sense that it will be used also to determine the right location for expansion. However, I’m not sure this is the most suitable task for AI. Using big data methodologies by humans may yield better results since some of the data streams would be subjective and inconsistent. Furthermore, planning path to opening new locations is a long-term planning project, and not real-time operational analysis. The speed of AI doesn’t necessarily have much value in this type of project.
My daughter did this task for a major retailer for eight years. While lots of demographic data was used in her analysis for new store locations, her selection successes came from the on-site and in-area visits she did. She found that certain centers worked better for her retailer due to the other retail makeup in the area. But this algorithm worked differently in different parts of the country. After looking through all of the data analysis, it took her intuition and lessons learned to make a final location recommendation. As good as AI tools might be, there is no substitute for human intelligence to review and adjust the AI results.
Site evaluation in the c-store industry was historically a mixture of “gut” and analysis. Someone from the company would spot what looked like a good corner, drive around and look at the number of rooftops, traffic patterns, etc. The analysis portion was fairly light. Over time data analysis replaced gut feel in the process. The move to using AI to do the analysis is a logical next step.
However, for Lawson the one missing ingredient is the store site’s known and possible future competitive set. I have worked with a site evaluation firm in the c-store industry in the U.S. They include an evaluation of the current competitors and research to determine the possible future ones. Also included are factors such as changes in destination drivers (current and known changes) surrounding the site.
Bottom line: the use of AI to handle a portion of the process makes sense, but until an AI can measure and evaluate the other factors relying solely on an AI for these decisions is a mistake.
As an aside, the Lawson name originated with a c-store company in Ohio. They worked with a company in Japan and sold them the rights to use the name in Japan.
Matthew answered this nicely — it’s an empirical question of whether AI makes the same decisions cheaper or better decisions that turn out to be more profitable. But keep in mind, just because they have lots of data to work with (from 13,000-or-so stores) does not mean most of the data is useful. That’s the fallacy of AI and Big Data solutions — believing that all that data is worthwhile.
It isn’t the end-state to use AI for the entire decision but it can certainly automate the large volume of manual processes to get to a smaller decision set. Yes, EXCEL DOES STILL EXIST in large retailers making these types of decisions. Using AI to automate the core data-set analysis, with even more contextual data sources being utilized, will only increase accuracy and velocity. Let the data analysts work on strategic work for retailers and not production-level data wrangling! This will only help retailers be more successful.
Everyone needs competitive advantages. Any metric that helps with picking the right location is a worthy investment for sure. But creating a brand that increases preference is equally important. Maybe even more so. Powerful branding turns a convenience-driven business into a destination.
I think it comes down to whether this is the right kind of problem for AI to solve. On the surface, it seems like the answer is yes. There’s a lot of data, and there may be depths to that data that a human may never pick up on.
On the other hand, location analysis may really just come down to traffic flow and demographics, and if that’s the case, it’s overkill.
Ultimately, though, I believe you have to have someone show up and actually look at the location — something an AI can’t do. I saw a baby/toy store go through exactly this a couple of years ago. On paper the location was perfect. Traffic, demographics, recent renovations to the parking lot and the strip mall facade, no competition anywhere in sight. The reality? There was a paint store next door, and contractors were coming and going, crowding out the baby store customers and creating noise, truck exhaust and even some hassling of young mothers trying to go to the store.
“Perfect” analysis is still subject to the harsh realities of real life that no amount of data training may ever pick up.
AI in most cases is simply data crunching to produce insightful deduction. Can more data points and better computation generate outputs worth consideration? Absolutely. But the importance of visually presenting results should not be overlooked. Interpreting the outcomes is the critical success factor.
I’m a big fan of leveraging AI, true machine learning, to ingest vast amounts of data that would require too many humans to do the same job with less accurate results. However, the sources and types of data to be ingested are key, of course. There are several external sources, much of which is unstructured and requires cognitive computing capabilities to derive usable insights. Also internal data, often 80 percent+ of which is not even seen by corporate systems, can drive location selection based upon best case scenarios.
While AI is great, it’s a tough sell to say it can replace any human analysis on selecting a new store location. Given comments from Paula elsewhere in this discussion, I find it hard to believe this is a trend yet for retailers if other existing technologies are so poorly used. I think we often fall into a trap assuming that if retailers are made aware of new technology, it will be adopted because the use cases are “so strong, how could they not!”
Seeing is believing, and I suspect we could come up with many examples of the selection process that until someone actually went to the location to see it — it was thought to be perfect. Then the site visit proved otherwise! AI should be one tool amongst many retailers rely on to choose the perfect location — it’s too young to be solely relied on.
So the idea of a company sifting data on where to open stores is something new? I realize we’re supposed to be wowed by use of the term “AI”, but without expansion on that phrase it doesn’t mean much.
The risk, of course, is that the AI used isn’t so smart after all, i.e. the algorithm turns in nonsensical results, but is blindly followed, and that isn’t realized until too late.
This application does not sound like AI, it sounds more like complex statistical modeling. We need to be careful not to get swept up the excitement that everything which is complex and uses a computer equals AI. Remember such scenarios happened before we started using this term this year!
Next, AI has to be taught. If we teach it badly, the results are going to be bad. It still needs our human contribution and responsibility. If AI performs badly, we should not jump to blame AI. We should look closer to home, and see how well we trained the system.
Yes, if the absolute best store location is the goal, then it is necessary to crunch a lot of data, but it is not sufficient. The big data crunching (note: I’m not saying AI because in this case, it’s a bit of a stretch) is going to get you halfway, the human half is to put that in context with all the data that didn’t get crunched.
The BrainTrust has covered this topic well, the only thing I would add is that for retailers and c-stores that have a number of locations and worry about cannibalizing their brand AI could be a useful tool to map out the new locations and driving patterns of their current clientele and flag them allowing retailers and c-stores to avoid those locations if possible.
I come from the retail real estate industry and have represented both retailers and the landlord/developer and as many of my colleagues point out finding great retail spaces can be an art and a science. Most retailers, even after crunching the data want to have a staff member spend significant time in the area and at the location.
Other than the acquisition of existing locations — true and tried locations — AI might be the best tool available for determining locations.
In California, we saw examples of bad store locations from Japan’s “Famima” convenience stores and England’s Tesco “Fresh and Easy” grocery stores. The locations appeared to be determined by affordable real estate locations, and while they were built in the right communities, some of these locations were not in the traffic flow, the appropriate households vs industrial distribution, or income levels. I don’t know what the final criteria were for the aforementioned locations, but AI could help validate or question the potential of a location and mitigate failures.
Using AI data to crunch through drone-captured parking and traffic data (including supplier delivery truck data) or data from cell towers or maybe trash analysis or sales tax — to back into the number of categories/sq/ft needed to find ROI and negotiating strategies vs. say an enhanced vending machine approach — might be a good use of machine learning.
Like others have said, the power of AI was enabled by the ability to handle more data. That ability was thanks to new high performance computers (GPUs etc) and AI software and methods.
The system figures out the weights and relevance on what matters. Traffic volume is cool, but what if the type of car and brand was an indicator?
I have a use case for this. A good one on why AI is the only future for site selection. Today there is infinite data, which is good, but it is also bad because its infinite data. It would take an expert the rest of their life to sift through and glean insights from the data … such as the one I chat about here.
Two eye lash extension franchise units are successful. Doing well. Nothing wrong, so nothing pops up in anyone’s head, not from the team nor the analytics. In the meantime the AI is churning in the background and comes up with an insight, a problem, but poses a question to both units. “How many beds do you have?” The one on the southwest side goes, “I have 6.” The one on the northeast side says, “I have 8.”
The AI responds, “found it.”
The unknown problem detected was customers we’re migrating from the southwest area’s unit to the larger northwest unit on the trade area’s fringe. Nothing accounted for it. Not the drive time, competive landscape, ampliphiers, suppressors, nothing … so its an anomaly, its timely and it’s live. That is where the AI shines and trumps an expert. It looks for the oddities that go missing.
Why did it ask about the beds? What it saw was the anomaly and taking into account everything else, only the merchandising (for lash extensions the “beds” are one of the main products) was unknown. It looked at the POS system and the scheduling system and saw that bookings were packed weeks in advance for the southeast store, primary culprit was too few beds and for eye lash extensions, that is too long. The wedding is this weekend. That big opera is Saturday, so people farther out gave up and took the extra drive for faster turn around.
So it suggested increasing the bedding by 4 to meet demand but more importantly, it stated if they do not do this, they may be out of business within 9 months.
Why? Because the area so ripe for picking. Good old supply and demand. If a competitor was parked across the street from the southeast store and saw the volume, smiled, and opened a 10 bed unit before the existing store took notice, it would spell the end.
Now let’s say we didn’t have the AI, what would have happened a year from now? The southeast store gets a competitor, it scrambles, tries to regain and retain customers. Eventually the competitor opens a second store … we know how this story ends. But what did the retailer say: “Couldn’t be helped, we didn’t see it coming.” And that is where AI is the game changer in site selection.