BI is dead! Long live Data as a service!
July 26, 2021
We all want to "make Redbull"
When I moved into the marketing world, thoughtful advertising concepts and Redbull campaigns were models for all marketers. At each marketing or social media conference, you met at least one case study that gave you wings. And every Czech marketer said: "They have a ten times bigger budget for one video than I have for the whole year." And when you talked to them about it, you quickly realized that a "perfect" campaign is just part of a comprehensive and very thoughtful marketing concept. And whether you're using 4P or 7P marketing, one "P" won't save it when the other "P's" lags behind. And then, logically, it's not even the budget.
It's the same with data. We all want to do machine learning, neural networks, speech processing, etc. But if you look at the data science market, for example, you will find that a lot of data science teams have either broken up, been bought by their most significant customer or started developing their product. Why? When a company wants to take the application of artificial intelligence seriously, it finds that sooner or later, it has to build its team, which for the first 1-3 years does nothing but clean up data and straighten processes in the company so that artificial intelligence can work. It is challenging to build a pricing engine that should dynamically set the prices of your products based on competition monitoring when 95% of your products have only one price in history. The one he got from the category manager at the time of his listing. So even the data has its "care," without which it is simply not possible.
What am I talking about? Apply such a concept about the maturity of the company at all. One "old and good" graduation model looks like this:
A lousy human factor
When you think about it a bit, talk to teams where you can build a data-driven culture, you will find that behind the success or failure of all projects is how well it is possible to adapt the people themselves. At first, we had (and still have) immortal excels. Then came BI tools, including better or worse self-analytics concepts like Tableau or Power BI. But that is still not enough. Why? In data-driven teams, they are increasingly finding that just "looking at data" does not affect. They need to be actively involved in the process. It is not enough to have just one "smart place" where the magical AI "invents" what to sell to whom. It is not enough to buy a clever marketing automation tool that "arranges everything." You need to be able to integrate different parts of the data architecture throughout the company. What am I talking about?
“Data as a Service”
In short and well, it's a logical concept to build a service architecture and an entire company so that it can be truly data-driven. A school example might look like this.
It is still true that without consolidating data in one "logical place," this process cannot be done. So clearly, we'll build a data warehouse, or better yet, a data lake. Ok. What's next? For the lake not to become a swamp, we need to clean and enrich the data. Then you can build AI models on top of them. However, what is crucial in this part is that this "data center" must also include an analytical engine initially used in BI tools. Why? Because without centralized analytical logic, metrics, dimensions, and functions, you will constantly struggle with the fact that "in BI it works differently than the buyer sees it," but it doesn't look at the charts. He orders according to how the shopping application offers/recommends/advises him.
Now imagine that you need all this to work in real-time to react dynamically according to what is happening in real-time, the whole data solution. On-time, reliably.
Today, many sub-concepts fit into this paradigm, such as
- Streaming data in real-time to your data warehouse (CDI continuous data integration) that support all the current leading players in the field of data warehousing, incl. Snowflake or BigQuery.
- ELT (unlike traditional ETL) concept, where you transform data only when you need it. On the go. In real-time. A lot of usable implementations are available both in dbt and in the competing Dataform.
- Training, testing, and deployment of many partial ML / AI models and algorithms and their concatenation.
- New concepts of self-service analytics and visualization, such as analytical notebooks, such as Count.
What complements the whole concept of Data as a Service and makes it complete is removing analytical logic from the BI tool and its access to other components via API. GoodData has recently introduced this solution with its GoodData.CN solution, about which we will hear a lot more.
So what is Data as a Service? The concept of building the architecture of services, products, and companies so that Data is available and usable anytime, by anyone, in any way. Data correct, current, reliable.