Restaurants deal with a lot of data; structured, unstructured, even streaming. Most of which - if stored at all - would live in its own database, probably in a data warehouse. Want to know transaction value on a given day? Pull that information out of the transaction database. Want to know how many millennials have signed up for your loyalty program? Check the CRM. Once upon a time this was enough, but the drive towards personalization has introduced new data requirements, and new challenges. Because a lot of this data tends to sit in its own database, it’s formatted in its own unique way and can be difficult to combine with data from other sources. So if you’re a QSR sifting through 5 separate databases to discover, for example, which demographic is most likely to purchase a certain combo at the downtown store at 4pm on a sunny weekday, and whether free sundaes or apple pies are the better inducement – well, you might be out of luck.
The Data Lake in food service
A Data Lake may sound slightly esoteric, but it’s basically just a strategy for storing data. The whole point is that you can throw all your data in together without needing to standardize anything first (some organization is generally a good idea, but data doesn’t all have to be of the same type or in the same format). So you can have your POS transaction data swimming around with customer demographics, reviews and loyalty reward status, inventory and supplier data – each in its native format. In addition to the data you ‘own’, you can also call in data from other sources via API: data that may be fast moving or quickly outdated (eg Twitter feeds), or the contextual time-sensitive data used in some predictive analyses – like weather or traffic conditions.
Then, at any time, anyone in the business can dip into the data lake for automated standard reports or ad-hoc analysis – a QSR might need to find out which items are selling in meals or alone, to identify which high value customers are in danger of churning, to explain why certain transactions were declined on certain days, or to analyze current brand sentiment according to social feedback and Yelp scores. You can draw out and analyze anything really, because all the data you’ve collected is right there waiting to be mined for insights; the Lake is specifically designed to make data recall and analysis much easier than traditional storage.
Future-proofed data collection
As more data is added to the Lake, it grows. It’s massively scalable, which is just as well, because IDC predicts a massive 40ZB of data will be created by 2020. And while no one QSR brand is likely to crack the zettabyte anytime soon, newly created data (including data generated via Machine Learning – like the data Plexure generates with its predictive algorithms) is expected to account for the bulk of it. So the more cool personalization and intent-based targeting you do, the more (and the greater variety of) data you’re generating, and the more likely the flexibility of a data lake is going to be increasingly relevant vs traditional data storage designed for structured data.
Fishing up meaning from the masses
Any time we start talking about Big Data, we really mean ‘enough relevant, meaningful data to make your decision making and personalization processes easier’. So it doesn’t make sense to have a giant data repository that’s basically the electronic equivalent ofHangar 51, or your Grandparents’ attic. Without the ability to pull meaning from ‘big’ data, it’s basically an accumulation of dusty junk. Luckily there are more, increasingly user-friendly, tools available to find the gold in all that data you’ve got stored away: check out David Inggs’ recent post and AzureCon presentation to see how McDonald’s is benefitting from Plexure’s use of Azure Data Lake and associated tools.
If you’re not averse to a buzz phrase or two, we’re entering at an era of ‘democratized analytics’. Data is more accessible than ever before, and the new wave of tools – including stream analytics and visualization software – make drawing insights from enormous and previousl yimpenetrable data sets not only easy, but easily used for business decision making (people love dashboards...)