This article just popped into my inbox this morning. But I see from the comments it's at least 2 years old. In any case I've heard of data lakes but did not know what they were or how to use them. Note of disclosure: I haven't fully read the article YET. But I take issue with the opening paragraph.
If you are designing a data warehouse around a set of user questions, you're doing it wrong. All software and software/database solutions are a model of some reality and/or process. The data models, be they operational or BI, should model that reality. Naturally, a DW designed around the initial questions will quickly become obsolete as questions beget more questions. Most users don't really even understand their own questions.
I design my data warehouses as a model of the business,not the initial questions. I work in the transportation industry. My data warehouses typically have 2 main fact tables: moves and financial transactions. Dimensions include the ever present date as well as location, customer, hauler (driver), freight type, etc. With the details of every move and every financial transaction (invoice line item, voucher line item, commission line item, etc.) there is literally no question the business can ask that the DW can't answer.
At my current employer my DW replaces something called the "cost file". This was their original "data warehouse" from the mainframe computer days. It fell into the trap of being designed around a question: "what are our costs?" It has a bad reputation because it is difficult to maintain and is not timely. It's a kluge of trying to shoehorn different attributes to answer different questions.
My data warehouse, designed around the core of our business, is much more robust. The business may change. A big change I see coming is driverless vehicles. I believe a well designed DW will be impervious to even that kind of paradigm shift. We are a transportation company. The hauler (driver) dimension may fall away, but we will always move freight from point A to point B. We'll always have moves and we'll always have to pay for those moves (financial transactions). A data warehouse (or any BI solution) designed around this reality will be far more robust than one designed around initial user questions.