Could you elaborate on how this works a little more?
It seems to me that this is a really dangerous state of affairs - and I'll set aside the obvious problem of different users creating different models with different semantics.
Sure. We are in a new age where we have a number of visualization (i.e.: PowerBI, Tableau, ETC) tools that allow us to remodel the existing data. We are able to take the original model, transform it without impacting the physical model, and transpose data in new ways within the front-end tools. This is what I mean by shifting that modeling from the back-end to the front-end. It means that instead of having the DBA, ETL Developer, whomever do the work, you can now have Report Developers, Visualization experts and so forth take on similar responsibilities with managing their own models/reports.
Over my career there have been countless occasions on which someone in the business has come over to IT looking for "help with a query". You listen to their problem, and you start to wonder what their actual end goal might be
This is true, but now tools are generating SQL code and more (i.e.: NoSQL tools) based on point-and-click models. While you can argue it's not the best optimized code, it does get the job done where you don't need to poke IT for a query. You can depend on the front-end application to do that for you and populate the model you created and or go directly to a report. This becomes a self-serving model, which is good.
I could list off dozens of examples of this
This is ignorance and has nothing to do with the technology, model and so forth. In my experience, the analyst have a good understanding of the data. They know the difference between the attributes and metrics. They have a understanding of the grains and how to ask questions of the data. When you run into cases where a user is averaging and average, they are completely oblivious to the data. They really have no right touching it. Yes, I know this is common, but you can either take your approach to your data seriously, or you can just randomly pull data without knowing what you're doing.
My concern with this is that you make things REALLY easy for the people who write the data and they may do this only once. But you make it REALLY hard for the people who read data who may do it many times.
It is an over-simplification but writing data incurs cost, reading data generates revenue.
I became a DBA because my work as a data analyst was severely hampered by the structure and quality of the data and that I was receiving. A vast amount of time was spent doing stuff to data rather than doing stuff with data.
Vast amounts of time were spent getting data into a form where I could derive insight from it and then I was up against the clock, having burned most of my time, in getting something of value out by the proscribed deadlines.
This is very true. If you can find a consistent model that always answers the same questions then a predefined model will always win. But, there are plenty of instances where flexibility is needed. Having the power to choose is what makes many products, regardless if they deal with data, succeed. Giving the end user the power to choose a different path is pretty powerful. Most analyst want that power, but they struggle with harnessing and using that power because of the limited technology skill-set they have (hence the prior poster referencing constantly getting poked for a new query). This is why in my experiences, a predefined model is not everything. Most analyst are fine with building a case to answer their data questions as long as they have the power to do so. But, it really depends on the situation and business.
For example, I work with a lot of data scientist and data analyst who all know the data and love to have flexibility with said data. Even with predefined models, questions of the data are constantly changing. This means I'm constantly having to keep up with that change.
With the proper business process--not technology--you can still maintain a flexible model on the front-end that's in the hands of the visualization/reporting team. And before anyone says it, not every analyst or business user has good understanding of data. This is what I focus most of my time on; education. I'm training the end users to be self-serving. I'm making them into mini-SQL developers, data modelers and more. I'm giving them the power to build their own worlds where I can just sit back and eat everyones lunch. 😀