The Short List
Data Mining. ETL. Business Intelligence. Data Lake. Every one of those was, in its moment, the thing that would finally fix how organizations use their own data. Every one of those got quietly absorbed into "how things are done" once the next bandwagon arrived. None of them actually disappeared. They just stopped being interesting enough to put on a conference agenda.
The full parade is longer than four entries, and it goes back further than most people in it were working.
The Hall of Fame
| Era | The New Bandwagon | What Was Actually Underneath |
|---|---|---|
| 1970s–80s | Management Information Systems / Decision Support Systems | Reports, printed overnight, on paper with holes down the sides |
| 1990s | Data Warehousing (Inmon vs. Kimball, a genuine religious war) | A second database you copied the first database into, on a schedule |
| 1990s | OLAP Cubes | SUM() and GROUP BY, pre-computed so nobody had to wait for it live |
| Late 1990s | Data Mining | Statistics, run on a bigger table than your professor ever had, with worse documentation |
| 2000s | ETL as its own discipline (Informatica, DataStage) | Copying data from one place to another, with a mapping document nobody kept current |
| Mid-2000s | Business Intelligence (Cognos, BusinessObjects, MicroStrategy) | Dashboards showing numbers that Finance disputed anyway |
| Mid-2000s | Service-Oriented Architecture / Enterprise Service Bus | Systems calling each other, now with more XML and a consulting retainer |
| 2000s | Master Data Management, "one single source of truth" | A new database that also needed reconciling against the other seven |
| 2008–2013 | Big Data, Hadoop, NoSQL | A bigger, slower way to do what the data warehouse already did, minus the schema |
| 2012+ | Data Science, "the sexiest job of the 21st century" | A statistician with a GitHub profile and a Python notebook |
| 2013+ | Self-Service BI (Tableau, QlikView, Power BI) | Business users building their own reports, and their own version of the truth, in parallel |
| 2015+ | Data Lake | A folder, usually in S3 or ADLS, usually undocumented within eighteen months |
| 2015–2018 | IoT Analytics | Time-series data at a scale where the sensors were more reliable than the pipeline |
| 2017–2019 | Blockchain for Enterprise Data | A slower, append-only database nobody could query, quietly retired |
| 2018+ | Citizen Data Scientist, Data Democratization | Handing the statistics to whoever asked for the dashboard first |
| 2019+ | Data Mesh | Giving every department its own data warehouse and calling the resulting chaos "domain ownership" |
| 2020+ | Data Fabric | A vendor diagram promising to connect everything nobody had connected in twenty years |
| 2020+ | Lakehouse | The Data Lake and the Data Warehouse, merged back into one thing, which is what a data warehouse already was |
| 2020+ | MLOps / DataOps | DevOps, for the pipeline that used to just be called "the ETL job" |
| 2022+ | Generative AI, "AI-powered insights" | A chatbot bolted onto the same dashboard, now with a natural-language box on top |
| 2024–2026 | Agentic AI | The chatbot now runs the ETL job too, and nobody has fully reviewed what it changed |
Nobody sat through every keynote on that list, and nobody was expected to. That's what the analyst reports and the consultants were for.
Five Reasons There's Always a New Bandwagon
1. Vendors Need a New SKU
A product that has been quietly doing its job since 2011 does not renew a licensing contract on its own. A product that has been "reinvented for the AI era" does. Rebranding is cheaper than rebuilding, and it resets the sales conversation to a point where the incumbent has no advantage.
2. Analysts Need a New Quadrant
An industry analyst firm cannot sell the same report two years running. A new category, Data Fabric this year, Agentic Data Platforms next, creates a fresh axis to rank vendors on, a fresh reason to commission the research, and a fresh reason for vendors to buy placement in it.
3. Résumés Need Refreshing
"ETL Developer" and "Data Engineer" describe, in most job postings, the identical set of responsibilities. Only one of them reads as current on a profile. The underlying skill, moving and shaping data correctly, doesn't expire. The label attached to it has a shelf life of about five years before it starts working against you in a job search.
4. Consultancies Need a New Keynote
Nobody buys a three-day workshop titled "Same Data Warehousing Practices as Last Decade, Executed Slightly Better." They buy the one titled "Data Mesh: Rearchitecting Your Organization for the Domain-Driven Future." The billable hours are often nearly identical work. The pitch deck is not.
5. The Real Problem Is Boring, and Fixing It Doesn't Announce Itself
Getting clean, well-documented, trustworthy, timely data into the hands of the people who need it is genuinely hard, and has been the actual problem, unchanged, since the first MIS report ran overnight. Declaring a new paradigm is far easier and far more fundable than doing the unglamorous work of fixing data quality, ownership, and documentation. Every bandwagon promises to finally solve the boring problem. None of them are actually about the boring problem.
What Doesn't Change
Underneath every entry in that table is the same short list of things that were true before the rebrand and stayed true after it: understand your source systems, model the data honestly, document what you built, and don't let the pipeline outlive the person who understood it. None of that is exciting enough for a keynote. All of it is what actually determines whether the next bandwagon, Agentic AI, does anything an organization can trust, or just runs the same undocumented ETL job faster and with less supervision.
The safest bet in this industry has never been picking the right bandwagon. It's building the fact table well enough that it survives the next one.