One of the most common pushbacks people use against data mesh is data ownership by domains will create siloes. Data marts are often mentioned in the same breath. If all data mesh said about data products was “package your data like a product” then yes, it would create siloes, agreed.
However, data product thinking includes making said product available/discoverable to potential consumers and even *gasp* known. A second order of data product thinking is making sure others inside your organization can find your product. Much like a product manager would do with an external facing product.
A simple way to prevent the likelihood of siloes is a proper data catalog setup (or a similar technology). That way, the data is easily discoverable by someone looking for it.
If I want to understand more about what is happening with orders, say to investigate extending shipping times, I would go and look for data products related to those topics. The data should be easy to understand because the Order Domain created documentation about their data products and offered a data output port so I can easily consume their data.
One reason a data product owner would *want* others to consume their data product is potentially better insights about their domain. If an external facing application is creating valuable information for analysis, then the application owner should be very excited to get valuable information to improve their application. Then, the team who did the analysis might create a data product for the original application owner to consume. This bi-directional consumption of data products is especially important for things like Machine Learning and Artificial Intelligence.
There’s another carrot (or possibly stick) some companies are using to encourage data product owners to push usage of their data products: These companies are tying consumption to the data product owner’s KPIs/OKRs/compensation in some way. Much like some companies do about external usage (sales) for an external facing product.
So yes, if a company does not 1) make data products easily discoverable and consumable and 2) does not provide incentivization for data product owners to push for consumption, data siloes are a likelihood. So don’t do either of those things 😉
There is some good commentary in this article by BIP (a consulting firm) about avoiding data siloes.