Two models of business intelligence keep getting argued as if one must win. In the first, a central team builds and owns every report, and everyone else requests changes through a queue. In the second, anyone can connect to data, build their own dashboards and share them. Both can be made to work. Both fail in predictable ways. I think the right answer is neither pure form, and I will say why.

The case for governed, centrally built reporting

The strength is trust. A small team that knows the source systems writes the logic once, tests it, documents it and controls who sees what. When the CFO’s pack and the board deck show the same number, it is because they come from the same governed place. For anything with audit, regulatory or contractual weight (financial statements, RBI-style regulatory returns, customer billing), this is not optional. Access control matters too: salary or customer-level data should not be a drag-and-drop away from every employee.

The weakness is the queue. A sales manager wants regional sales split by product family, which the standard report does not do. She raises a request, which waits behind forty others. Three weeks later it arrives, by which time the question has changed. Meanwhile the central team spends its time on small cosmetic changes instead of improving the data. The usual outcome: people export to Excel, build what they need, and the “governed” system becomes a data download service.

The case for self-service

The strength is speed and proximity. The person asking the question knows the business context, and answering it takes minutes. Analysts embedded in operations can try ten angles in an afternoon. Curiosity is cheap when the tool is open, and many of the most useful analyses start as somebody’s private experiment.

The weakness is what happens at scale. Within a year there are hundreds of dashboards, many abandoned, several with the same title and different numbers. Each author defines “active customer” or “net sales” slightly differently. Nobody knows which are safe to quote. Performance suffers when twenty people each pull full-year transaction data. And a sensitive field leaks because someone joined the wrong table and shared the link widely. The slogan “democratise data” meets “which of these numbers is right?”

Where I land: govern the foundation, free the surface

My position is that the disagreement is usually about the wrong layer. You do not have to choose who builds dashboards. You have to choose who owns definitions and data access. Those two can be separated.

Put the heavy lifting in a governed core: clean, documented tables or a shared semantic model that holds the agreed measures (net revenue, active customer, churn), the joins, and row-level security. A small central team owns it and is accountable for its quality. Then let people build on top of it however they like. A sales manager who drags “net revenue” and “product family” into a chart is using the governed definition automatically, so her quick dashboard cannot disagree with the finance pack on what revenue is.

Often people propose a tiering scheme on top of this, and it is worth considering, because it makes the rules visible:

TierWho buildsStatusExamples
CertifiedCentral or trained analystsTested, documented, owner namedBoard pack, regulatory and billing reports
TeamDomain analystsBuilt on governed data, peer-reviewedRegional sales review, ops scorecards
PersonalAnyoneUnreviewed, labelled as suchOne-off questions, prototypes

A personal dashboard that turns out to be useful gets promoted: someone reviews it, moves its logic into the core if needed, and certifies it. This gives the self-service crowd a route into the official world, and gives the central team a free pipeline of demand to learn from.

What makes the middle hard

It is boring work, which is part of why it gets skipped. Someone has to curate the core, resist adding every request, retire stale content, and label what is certified so that people can tell at a glance. Some honest costs:

  • Training. Self-service does not mean no skills needed. Users need to understand joins, filters and the grain of the data, or they will build confidently wrong things, such as multiplying rows with a careless join. Short, practical onboarding beats a long manual.
  • Clean-up. Without a periodic review of unused dashboards, the clutter returns. Many teams archive anything not opened in a set number of months.
  • Capacity. The core team becomes a product team. If it is staffed like a ticket desk, it will behave like one.
  • Culture. Some leaders prefer a polished central pack even when a quick view would do; others resent any central control. Neither habit changes through a policy memo alone.

I should admit where this view is weakest. For a company of thirty people with one analyst, the whole structure is overkill; a handful of well-built dashboards and a shared definitions page is enough. And in a very regulated setting, the personal tier may be restricted to sandbox data, which narrows the freedom considerably.

A way to test which way you lean

Count two things for your own organisation: how many days a typical new report request waits, and how many different values people can find for one headline metric. If the wait is long and the metric is stable, you are over-governed. If the wait is short and the numbers disagree, you are under-governed. If both are bad, fix the shared definitions first, because it makes everything else cheaper.

Figures and names in the examples are illustrative.

Leave a Reply

Trending