Restaurant data is dirty. Like insanely dirty. And some of this is understandable: restaurants are, in effect, manufacturing facilities where chefs put their own spin on common items. It’s why there are probably no less than 10,000 different types of hamburgers across the country even though they all have the same basic ingredients of a bun and ground beef.
Tracking this is an absolute nightmare in a low margin industry dealing with perishable products, so you shouldn’t be surprised that more sophisticated food operators (ie grocery stores) standardized on UPCs (universal product codes) and retail has standardize on SKUs (stock keeping units) as much as possible. But restaurants? Still behind everyone else by decades (maybe millennia).
Not to fear, though: data science can get us through.
If you want some technical details for how item data cleansing happens, you can read about what Square / Caviar did to cleanse online ordering data.
It’s a fun data science task and definitely a necessity when ensuring accuracy for multi-billion dollar online ordering companies… but hard to justify the scientific rigor when restaurants aren’t inclined to pay much for anything. So when we took our approach we started with the 80/20 perspective; that is, fully cleansing and classifying data would be on the order of $5M. Chump change to an UberEats, but very material for our company which doesn’t earn 30% fees on delivery orders (nor makes make hundreds in monthly EBTIDA per merchant via payments processing).
Thus we used a combination of word embedding and cosine distance. These are statistical techniques which use some training data and neural networks to assign similarities to words based on characters. You don’t really need to read too much into it but know that this is probably the first time data scientists have bothered to to provide merchants this value.
The result is localized search that displays the items that are selling in your area, the popularity of the item, and the price ranges of the item.
For example, if you’re in New York City we look at what’s selling across POS systems in your area and do our best to group the items together into similar categories. Ideally we’d recommend what items to sell based upon your type of restaurant (which is a whole other problem because there are no clean categories for independent restaurants who often serve multiple types of cuisine), but the purist in us feels really uncomfortable recommending items without knowing COGS of the items. It’s the same reason why menu engineering requires COGS data: otherwise price changes could lose the business money.
But we know that 85% of restaurants don’t know their own COGS which makes the potential market for that relatively small and relegated to enterprise merchants. SMBs get boned again, but this might be their own fault.
Regardless, below is a screenshot for how this works today. If you’re an SMB you’ll see someone else’s logo because we white label this with channel partners.
If you do want recommendations you have to settle for our other variety across labor and theft optimization, where we quantify the necessary changes by employee, shift, and revenue center, and then audit the results to see if anyone actually took the action to generate the benefit. ROI us 5-8% of revenue and this has been proven in countless case studies even before the application of data science… it’s just that data science makes it way easier to understand, and way clearer to measure the ROI.
Again, you can’t do this without a lot of PhD investment, which nobody makes because, you know, payments = free money. Seriously, why don’t Square, Toast, Shopify, and countless others who claim to be about creating value do this? Because it’s infinitely harder than pulling on the payments crank. Oh, and we guess nobody has POS data from 100K restaurants either. Only took us ~9 years to get (it’s 3x what Toast has for context).
Wait, so how do I get this?
If you’re a merchant and you want this tool, send us an email (jordan @ whatsbusy [dot] com) including your number of stores and your POS system and we’ll try to track down your POS partner; it’s possible they may already have this tool available and are just doing a really bad job of explaining value.
If you provide value to restaurants (don’t worry, this will be available for retailers before the end of the year) and don’t yet offer this you should know that we automatically sell this for you on ROI so there’s no manual sales work required… it’s the only way to scale sales in SMB, in fact. So once we have the POS data from your customers, all the sales are handled automatically just as how the aforementioned article explains. #Software.
This tool kills it in enterprise and we’re proving that data science DOES make life easier for merchants of all sizes and IS affordable. And no, there are no gotchas or duplicitous payments contracts: you either get the ROI or you can dump it. Listen to what enterprise industry executives have said:
Many restauranteurs attempt to harvest insights from their data or bring in expensive platforms that over promise and under deliver…a common theme in our space… I have seen countless presentations on “consumer insights” and “AI tools” and besides being very expensive, offer little actionable insights for the operators and require a team of data scientists to actively work the data sets. This tool fixes all of that.
CIO 1000-unit brand
This tool covers a significant portion of what we would like to achieve. Specifically, predictive analytics for forecasting and prescriptive analytics for operation and marketing decisions. This information will have a significant impact on top line sales by helping us better utilize our marketing dollars, and improve profitability through precise store-level operational improvements.
VP Marketing 1400-unit brand
I have known Jordan and been familiar with his data science endeavors for roughly five years. Recently, Jordan shared that he would be making available his prescriptive tools to enterprise restaurants. I have seen what Jordan has produced, and it is substantially better than anything I have found available, even in the enterprise segment of the restaurant vertical, which is typically where more advanced solutions make themselves known. This tool would not only provide specific, quantifiable recommendations for our operations teams that we cannot provide today, but the tool has the potential to drastically improve our marketing spend allocation with more demonstrable, and higher, ROAS.
CDO 1500-unit brand
We will make some more announcements over the next month as we roll out an industry-wide benchmark on top of POS data from 100K restaurants (in addition to data we’ve exclusively secured from the card networks), but as far as we’re aware this is the first time anyone has bothered investing in data science to do anything prescriptive for the restaurant industry… and we’ve sure as hell seen a lot of disappointments over 9 years.
Can you focus on delivering quantifiable value and succeed in brick and mortar? It will be a case study for the ages.
Add comment