Theft is a huge problem in the restaurant business. There’s a laundry list of studies that show what a huge impact theft has to the bottom line. The National Restaurant Association estimates that 75% of inventory shortages are caused by theft. How is theft manifested? Employees give away free food, ring up false transactions, take money from registers or steal tips – all just a few examples of employee theft.
Not all theft can be caught. However, the point of sale (POS) is a huge advantage for the restauranteur. The POS maintains a repository of transaction data that can be used to determine what was stolen, when it was stolen, and how much it impacted the bottom line. In more technology-forward industries, data mining for shrinkage has been in use for decades. In fact, the statistical techniques employed to find suspicions transactions are not all that complicated.
In the restaurant industry these techniques are rather new, and are something that should be employed at every restaurant given their low cost of implementation. Many POS companies will call these solutions “exception based reporting” or “loss prevention”and charge $50 – $100 per month. Here’s how they work:
A line of code crawls over all the transactions that were recorded in the POS. The code starts to assemble relationships between transactions and builds a database. The results are lists of items that were sold, and items that sold with those items. For example, the relationship could be that for every 10 hamburgers sold, 1 coke sells. Using these types of relationships a confidence matrix is built, giving the probability of theft for transactions made on the POS. All of these activities can be drilled down to shift and employee.
In this way you can use data to determine suspicious activity, and what employees to monitor more closely. Just don’t pay $100 a month for the solution – it’s not that complicated.
The restaurant industry has long been a technology laggard. It may be that the business is inherently low margin (unless you’re McDonald’s and control your logistics and distribution) and restaurants don’t see value in technology; or that the industry is incredibly fragmented, with only 30% of the 1,000,000 US restaurants represented by chains, and few technology companies want to chase a thin market. Whatever the reason, on its face there seems to be plenty of opportunities to help restaurants with technology since so few use it today.
Big Data is an elusive term. Depending who you talk to, it can have different definitions. In our world Big Data involves taking lots and lots of data and using computers to find correlations that a human never could. As computers have gotten faster, databases bigger, and overall processing costs lower, we can now use Big Data to help industries that previously couldn’t afford an expensive IBM tool.
Now how can you use Big Data in the restaurant industry? Can you use it to determine what promotions are working? Can you see how well promotions work at other restaurants and if they’ll work for you? Can you find the best times to run promotions, and what promotions to run? Can you use it to forecast sales better? Can you understand how other restaurants are performing to figure out how well you’ll do? Can you optimize labor and inventory? Can you find theft? These are all topics we’ll be discussing over the next few weeks as we lay out the importance of data in running a business.
You’ve probably been to the airport and seen personnel in nondescript vests pushing around the elderly and disabled to the security checkpoint. These people are ushers, and are contractors to the major air carriers falling under the banner of PrimeFlight, G2 or AirServe. Normally, I would never have noticed their activities as out-of-norm. But on a recent trip to San Francisco International Airport, they were doing something quite interesting.
I arrived for my departure 2 hours before take off at Terminal 1, gates 40-48. The queue was a respectable 20 minutes, but then these personnel in nondescript vests started asking everyone in line for their departure time. To my surprise there were people around me who had to catch a flight in 30 minutes. But even MORE surprising, the personnel in nondescript vests moved these travelers to the front of the queue! Soon my wait burgeoned from a mere 20 minutes to 38. There were over 200 passengers that were placed in the queue in front of me.
After digging into some research I learned these people were PrimeFlight personnel. Their job was to move travelers devoid of any personal responsibility to the front of the line. So I posit this to the airlines: what’s my incentive to follow rules and get to the airport on time?
Several airports have tried using hardware solutions to determine TSA security wait times. We’re still not convinced real-time waits are as important as understanding what the wait time will be like when you get to the airport. Nonetheless, there are parties who spend hundreds of thousands of dollars implementing these solutions.
Bluetooth technology is designed to pick up signals from devices that have Bluetooth enabled. Each device has a unique signal so the device can be tracked. Airports set up the Bluetooth receivers to identify when a phone first enters a security line, and when it exits. The problem has been two fold. First, not many devices have Bluetooth enabled. Second, determining if a unique device is actually in line or just somewhere near the line has proven problematic.
The below chart shows the wait time results from a Bluetooth installation at Indianapolis International Airport. Bluetooth would tell us that there were multiple waits over 1 hour in duration for a given day. This is entirely inaccurate. Don’t believe it? Let’s look at some sample data from Bush Intercontinental Airport, where we can see how many gaps exist in Bluetooth data.
We can see the large gaps in device signals from the Bluetooth data from Bush International. This particular image was generated from Terminal D on May 31st, 2013. Many times the Bluetooth system picks up ghost devices: devices where there’s an entry or exit signal, but not the corresponding converse.
IAH Bluetooth Data Terminal D
The quality of Bluetooth is suspect. It misses many passengers who may have their Bluetooth devices off and the system has a difficult time pinpointing device location accurately. The good news is that the Houston Airport System spent less than a million dollars on this Bluetooth system – we’ve seen tax dollars spent more poorly before.
You can see the full purchase order here: HAS Bluetooth costs
Are there certain days and times that are better than others when it comes to airport wait times? We looked at the top US airports over the past month and pulled out the day of week and time of day where security wait times were the highest.
When looking for the worst airport wait times, we really had a long list of candidates. Nearly every airport had some egregious waits. But instead of publishing a novel we decided to choose the top 3 airports with the worst wait times on a consistent basis. These airports had more than 20 minute waits several times a day, and some with waits over an hour.
Your winners are…
Los Angeles International Airport (LAX)
LaGuardia International Airport (LGA)
Kansas City International Airport (MCI)
We often get asked: what’s the average wait time at the airport? It’s a bit of a misleading question, really. What do you mean by average? Wait times are dependent on flight schedules, and one vacant checkpoint might greatly skew the average for the airport. Regardless, here’s a chart of the average wait times for the top 10 airports by time of day. This airport wait time data was taken by normalizing a month’s worth of wait times for all the airport’s checkpoints. But take it with a grain of salt – we’ve seen wait times balloon, that make these averages inconsequential.
Average Wait Time by Time of Day at Top 10 US Airports
Over the past year we’ve spent a good chunk of time working with various stakeholders to build a giant algorithm that forecasts airport security wait times. But how does it work?
Think about a security line as a bucket of water. There are droplets of water falling in the bucket, and a little hole in the bottom of the bucket where water drips out. The drops into the bucket are the passengers arriving at security and the drops out are the passengers processed by TSA. So how do you know when and how many drops fall into the bucket, and when and how many drops drip out?
Passengers arrive for a departure in a bell curve shape. The first passenger might show up 2 hours before his flight. The next person arrives 4 minutes later and then five people arrive 2 minutes after that. Overall you’ll see a nice bell curve. The bell curve, however, can change! Things like weather, times of year, and even the type of traveler (i.e. business traveler v.s. budget airline traveler) can influence what the bell curve looks like. Luckily we have years of data that help us understand how these external factors influence the bell curve.
Airport Passenger Arrival Bell Curve
When you understand how passengers arrive, you can next turn to how they are being processed. In 2004 the TSA contracted IBM to build SAM, or Staffing Allocation Model. This model tells TSA how to staff to process passengers to minimize airport security wait times. You can learn how much this cost the TSA here.
The net output is an algorithm that understands how passengers arrive at the airport, and how they’re processed. All you need is a consistent feed of flight departures and you can determine the wait times at each checkpoint where TSA staffs. For our particular UI, we give the maximum wait for a 30 minute period. Over time we will fine tune this to 15 minute periods, but as of now it’s a great place to start.
And that’s how you can put math to work!