Want Smarter Spights in your Inbox? Sign up for our weekly newsletters to get what items on business leaders, data, and security leaders. Subscribe now
In the last decade, companies spend billions Data infrastructure. Petabyte-scale warehouses. Prizes of Time Time. Machine Learning Platforms (ML).
And yet – ask your operations leading why churn increases last week, and you probably get three conflicting dashboards. Ask the finance to restore the performance of nature’s identification systems, and hear you, “it depends on what you ask.”
In a world drowning in dashboards, a fact continues to surf: data is not the problem – product thinking.
The quiet collapse of “data-as-a-service”
For many years, Data Tims Works like internal consultancies – reactive, ticket based, hero-driven. This “Data-as-A-Service” (Daas) model is good if data requests are small and stakes below. But as “Data-driven companies,” this model is broken under the weight of one’s own success.
Get airbnb. Before launching its platform, product, finance and ops team pulls their own versions of metrics such as:
- Booked nights
- Active User
- Available list
Even simple KPIs are different from filters, sources and who ask. In leadership reviews, various teams present different numbers – resulting in arguments of whose metric “is correct” instead of what to do.
These are not technology failures. They are the product of the product.
The consequences
- Data mistrust: Analysts are the second regarded. Dashboards left.
- People: Data scientists spent a lot of time explanation of differences than making insights.
- Tubes: engineers build the same datas across the teams.
- Dissision Drix: Leaders can delay or ignore action due to inconsistent inputs.
Because data dependence is a product problem, not a technical one
Most data leaders think they have an issue with quality data. But see closer, and you’ll find the issue of dependent data:
- Your experimental excorce says that one part is hurting maintenance – but product leaders do not believe it.
- Ops see a dashboard as opposed to their lives.
- Two teams use the same metric name, but different logic.
The pipes work. SQL sounds. But no one trusts outputs.
This is a product failure, not an engineering. Because systems are not designed for anything, translating or making decisions.
Enter: Data Product Manager
A new paper has emerged with top companies – Data Product Manager (DPM). Unlike Generalist PMS, DPMS works throughout abundant, invisible, cross-functional soil. Their work does not send dashboards. This is to ensure that the right people have the right sight at the right time in make a decision.
But DPMs don’t stop piping data into dashboards or tables pushing. The best people go further: they asked, “What really helps someone get their job better?” They explain success not in terms of outputs, but results. Not “it sent it?” But “is this material way of improving work quality or quality of decision?”
In practice, this means:
- Don’t mean users; Note them. Ask how they believe the product works. Sit beside them. Your job will not send a dataset – this is to make your customer effective. That means so much how the product fits the real world context of their job.
- Own canonical metrics and treat them like apis – prescribed, decided, administered – and make sure the product is like $ 10 million product in the product.
- Building internal interfaces – like shop stores and clean room items – not as infrastructure, but as real products with contracts, slas, users and feedback loops.
- Say not on projects to feel sophisticated but not important. A data pipeline with no team use is technical debt, not progress.
- Design for whole. Multiple data products failed from bad modeling, but from British systems: No document logic, flaky pipelines, shadow possession. Building with the mind that your future self – or your replacement – thank you.
- Resolve the horizontal. Unlike specific PMS in domain, DPMs should always zoom. A Life Life of a Team (LTV) logic is another post input to the team budget. A small minor metric update may have the consequences of the second order of total marketing, finance and operations. The stewardly complex is to work.
In companies, DPMS is quiet again to change what internal data systems, handled and adopted. They are not there to clean the data. They are there to make organizations also believe it.
Why long
For many years, we are wrong activities for progress. Data engineers build pipelines. Scientists have built models. The analysts are built on the dashboards. But no one asks: “Is this sharp understanding of a business decision?” Or worse: we asked, but no owner of the answer.
Because executive decisions are now appointed-of-data
In today’s business, almost every major decision – budget transfer, New LaunchesORG changes – pass through a data layer. But these layers are always not turned back:
- The metric version used in the previous quarter has changed – but no one knows when or why.
- Experiment logic is different from teams.
- The models of recognition will conflict with each other, each one has a significant logic.
DPMS does not take care of – they own the interface that makes the decision to read.
DPMS insures that metrics translated, minds are transparent and tools in accordance with real workflows. Without them, the decision paralysis becomes behavior.
Why is this role facilitate at AI Era
AI cannot replace DPMS. It will make them important:
- 80% of AI’s AI efforts still go to data readiness (Forrester).
- As many language models (LLMS) scale, the cost of compounds in the waste input. AI doesn’t fix bad data – it increases.
- Pressure Pressure (the EU AI Act, the California Consumer Privacy Act) pushes orgs to treat internal data systems with product rigor.
DPMS are not traffic authors. They are the architects of trust, translation, and responsible foundations of AI.
So what now?
If you are a CPO, CTO or data head, question:
- Who owns the power systems of power over our greatest decisions?
- Is our internal APIs and dimensions, discovered and managed?
- Do we know which data products are adopted – and that is quietly withholding trust?
If you can’t answer clearly, you don’t need a lot of dashboards.
You need a product manager of data.
Seojoon Oh a product manager of Uber data.
