WTF are data lakes?

So WTF are data lakes?
Data lakes are repositories where large chunks of multiple types of data are stored in their native format.

Why do they matter?
Most other data platforms will only store data once it has been formatted to fit a particular structure. For example, data warehouses will only accept data once it has been structured into rows and columns, so unstructured data such as phone call transcripts typically can’t be uploaded.

But data lakes ingest data in its raw form, regardless of how the data is structured, which means that a data lake can take in unstructured data like call transcripts, said Trevor Paulsen, a product manager at Adobe. By taking in all formats, data lakes have more data at their disposal.

This allows data-lake users to “keep all of your data, save it and then figure out what to use,” said Michael Hiskey, CMO of data-management firm Semarchy.

Who uses these data lakes?
Theoretically, anyone who wants to mesh together various types of data, like media and marketing companies but also universities, hospitals and businesses in general.

OK. So what’s an example of how they are used?
Imagine you are a CPG company trying to analyze data from many different sources. To determine if certain ads drove more sales at brick-and-mortar stores, you could combine various data formats like the user ID and browsing data from ad servers with your shipment data.

“If an analyst says, ‘I want to create a data set out of these things,’ I can more easily do that because the data lake can hold all of it,” said Max Knight, vp of marketing science services at demand-side platform Turn. “Before, that information existed in disparate systems. The data lake gives you a more universal access.”

Isn’t that what a data-management platform is for?
Sort of. But DMPs are structured around user and media-related data such as cookie IDs and audience segments. Data lakes are more encompassing and can include offline data related to topics like logistics and production, Knight said.

What makes this different from having a huge Dropbox folder?
From a pure storage standpoint, not a whole lot. But unlike consumer-facing file-hosting services, data lakes offer a lot of analytics tools like SQL, R, BigML and Python so that analysts can quickly run through millions of data points, Paulsen said.

Sounds good. Are there any drawbacks?
Cost. If you are storing 100 terabytes with a cloud service like Microsoft Azure, it will cost about $3,000 per month. Data lakes can also be built using free, open-source software like Apache Hadoop, but you still have to pay web developers to build out the data lake.

Another drawback is data overload. Big data doesn’t do any good if the people gathering the data don’t have an idea of what to do with it. And with so much data in one location, it can become difficult for analysts to quickly find the exact data set they need.

“If everything is stored, it’s easy for the true insights that can solve problems to get lost, or even worse, never discovered,” said Michael Collins, an independent ad tech consultant.