Need smarter insights in your inbox? Join our weekly newsletters to get solely what issues to enterprise AI, information, and safety leaders. Subscribe Now
Previously decade, firms have spent billions on information infrastructure. Petabyte-scale warehouses. Actual-time pipelines. Machine studying (ML) platforms.
And but — ask your operations lead why churn elevated final week, and also you’ll doubtless get three conflicting dashboards. Ask finance to reconcile efficiency throughout attribution methods, and also you’ll hear, “It relies on who you ask.”
In a world drowning in dashboards, one fact retains surfacing: Information isn’t the issue — product considering is.
The quiet collapse of “data-as-a-service”
For years, information groups operated like inner consultancies — reactive, ticket-based, hero-driven. This “data-as-a-service” (DaaS) mannequin was positive when information requests had been small and stakes had been low. However as firms turned “data-driven,” this mannequin fractured beneath the load of its personal success.
Take Airbnb. Earlier than the launch of its metrics platform, product, finance and ops groups pulled their very own variations of metrics like:
- Nights booked
- Energetic consumer
- Accessible itemizing
Even easy KPIs diversified by filters, sources and who was asking. In management opinions, completely different groups introduced completely different numbers — leading to arguments over whose metric was “right” slightly than what motion to take.
These aren’t expertise failures. They’re product failures.
The results
- Information mistrust: Analysts are second-guessed. Dashboards are deserted.
- Human routers: Information scientists spend extra time explaining discrepancies than producing insights.
- Redundant pipelines: Engineers rebuild comparable datasets throughout groups.
- Resolution drag: Leaders delay or ignore motion on account of inconsistent inputs.
As a result of information belief is a product downside, not a technical one
Most information leaders assume they’ve an information high quality difficulty. However look nearer, and also you’ll discover a information belief difficulty:
- Your experimentation platform says a characteristic hurts retention — however product leaders don’t consider it.
- Ops sees a dashboard that contradicts their lived expertise.
- Two groups use the identical metric identify, however completely different logic.
The pipelines are working. The SQL is sound. However nobody trusts the outputs.
This can be a product failure, not an engineering one. As a result of the methods weren’t designed for usability, interpretability or decision-making.
Enter: The information product supervisor
A brand new function has emerged throughout prime firms — the information product supervisor (DPM). In contrast to generalist PMs, DPMs function throughout brittle, invisible, cross-functional terrain. Their job isn’t to ship dashboards. It’s to make sure the appropriate individuals have the appropriate perception on the proper time to decide.
However DPMs don’t cease at piping information into dashboards or curating tables. The most effective ones go additional: They ask, “Is that this truly serving to somebody do their job higher?” They outline success not by way of outputs, however outcomes. Not “Was this shipped?” however “Did this materially enhance somebody’s workflow or choice high quality?”
In observe, this implies:
- Don’t simply outline customers; observe them. Ask how they consider the product works. Sit beside them. Your job isn’t to ship a dataset — it’s to make your buyer more practical. Meaning deeply understanding how the product suits into the real-world context of their work.
- Personal canonical metrics and deal with them like APIs — versioned, documented, ruled — and guarantee they’re tied to consequential selections like $10 million price range unlocks or go/no-go product launches.
- Construct inner interfaces — like characteristic shops and clear room APIs — not as infrastructure, however as actual merchandise with contracts, SLAs, customers and suggestions loops.
- Say no to tasks that really feel refined however don’t matter. An information pipeline that no staff makes use of is technical debt, not progress.
- Design for sturdiness. Many information merchandise fail not from unhealthy modeling, however from brittle methods: undocumented logic, flaky pipelines, shadow possession. Construct with the belief that your future self — or your alternative — will thanks.
- Clear up horizontally. In contrast to domain-specific PMs, DPMs should always zoom out. One staff’s lifetime worth (LTV) logic is one other staff’s price range enter. A seemingly minor metric replace can have second-order penalties throughout advertising and marketing, finance and operations. Stewarding that complexity is the job.
At firms, DPMs are quietly redefining how inner information methods are constructed, ruled and adopted. They aren’t there to scrub information. They’re there to make organizations consider in it once more.
Why it took so lengthy
For years, we mistook exercise for progress. Information engineers constructed pipelines. Scientists constructed fashions. Analysts constructed dashboards. However nobody requested: “Will this perception truly change a enterprise choice?” Or worse: We requested, however nobody owned the reply.
As a result of govt selections are actually data-mediated
In right now’s enterprise, practically each main choice — price range shifts, new launches, org restructures — passes via an information layer first. However these layers are sometimes unowned:
- The metric model used final quarter has modified — however nobody is aware of when or why.
- Experimentation logic differs throughout groups.
- Attribution fashions contradict one another, every with believable logic.
DPMs don’t personal the choice — they personal the interface that makes the choice legible.
DPMs make sure that metrics are interpretable, assumptions are clear and instruments are aligned to actual workflows. With out them, choice paralysis turns into the norm.
Why this function will speed up within the AI period
AI received’t change DPMs. It is going to make them important:
- 80% of AI challenge effort nonetheless goes to information readiness (Forrester).
- As giant language fashions (LLMs) scale, the price of rubbish inputs compounds. AI doesn’t repair unhealthy information — it amplifies it.
- Regulatory strain (the EU AI Act, the California Client Privateness Act) is pushing orgs to deal with inner information methods with product rigor.
DPMs will not be visitors coordinators. They’re the architects of belief, interpretability, and accountable AI foundations.
So what now?
In the event you’re a CPO, CTO or head of information, ask:
- Who owns the information methods that energy our largest selections?
- Are our inner APIs and metrics versioned, discoverable and ruled?
- Do we all know which information merchandise are adopted — and that are quietly undermining belief?
In the event you can’t reply clearly, you don’t want extra dashboards.
You want an information product supervisor.
Seojoon Oh is an information product supervisor at Uber.
