Three dashboards disagree because they use different definitions, grains, filters, or refresh windows for the same label. Pick a single source of truth analytics setup by naming one canonical dataset, one owner, and one published definition every report must reuse.
Why three dashboards can all look right
Sales, finance, and operations often each built a report that was locally honest. One counts booked pipeline, another invoiced revenue, a third cash collected. All three may still be titled "revenue."
Grain is the next silent split. A daily snapshot and a month-to-date rollup will not match in the afternoon. Filters hide the rest: one tile drops test accounts, another includes partners, a third ignores refunds.
None of this is a charting bug. It is a modelling and ownership problem that a prettier dashboard cannot fix.
Diagnose the disagreement before you pick a winner
Start with the metric name, not the dashboard. Write the label down and ask each team what event creates a row, when that row is considered final, and which systems can rewrite it later.
Then compare four things, in this order.
Definition
What business event is being counted? A "customer" might be a billed legal entity, a logged-in user, or a shipping address. Until that sentence is written down, every dashboard is improvising.
Grain
Is the row a person, an account, an order, or a day? Mixing grains is how a unique-users tile disagrees with an active-accounts tile that leadership treats as the same number.
Time window and freshness
Does the tile mean yesterday, month-to-date, or a trailing window? Is it last night's batch or a streaming approximation? Teams often argue about numbers that are simply not contemporaneous.
Filters and late-arriving facts
Refunds, chargebacks, test tenants, and delayed warehouse loads change history. If one dashboard is a snapshot and another is restated, they will never stay aligned.
Walk those four questions on the three dashboards that started the fight. You will usually find the split in the first two.
What a source of truth actually is
A source of truth is not "the warehouse" or "the dashboard everyone likes." It is a published contract: a named table or semantic metric, an owner, a definition, a grain, a freshness target, and a rule for how late data is handled.
Everything else is a view, a department cut, or an experiment. Those can exist. They must not reuse the same label as the canonical metric.
Citi Global Wealth spent eighteen months moving off five regional datacentres onto one cloud platform. Computer Weekly reported on 8 September 2026 that reporting compressed from weeks to days, and that their medallion design treated silver as the unified data model. That conformed layer is the source of truth you want, not a gold dashboard that silently redefines the metric.
Field notes on unifying data, governance and getting analytics teams to one source of truth.
Every Monday, five minutes.
How to pick the canonical source
Use this order. Do not skip steps to please the loudest stakeholder.
Follow the system of record for the event
The source of truth for "invoice issued" is the billing system, not the CRM opportunity that predicted it. The source of truth for headcount is HRIS, not last Friday's finance export.
Prefer the cleaned, conformed layer over raw and over the chart
Raw extracts still carry source quirks. Dashboard calculations hide logic in a click path nobody versions. The conformed model — often silver in a medallion lakehouse — is where keys, types, and business names should stabilize.
Require a named owner who can change the definition
If nobody can edit the metric without a committee, it is not a source of truth. It is folklore. The owner publishes the definition, the grain, and the exceptions list.
Publish one label, and rename the rest
The losing dashboards do not get deleted on day one. They get renamed: "Revenue (cash, ops cut)" instead of "Revenue." Label collision is what created the fight.
A working ritual for the next month
Pick the three metrics that cause the most chat arguments. For each one, write a one-page definition: event, grain, time, filters, owner, source table.
Wire every executive dashboard to that table or semantic metric. If a team needs a different cut, they fork the name, not the definition.
When a number still disagrees, treat it as an incident: which input moved, which filter diverged, which load was late. Do not reopen the definition in the same meeting as the incident.
LensAIght is a unified data intelligence platform so chat, notebooks, and dashboards can point at the same modelled datasets instead of each growing a private extract. The product does not magically create a source of truth. It makes the chosen one harder to accidentally fork.
What not to do
Do not average the three dashboards. A blended number that nobody owns is worse than a wrong number with a named owner.
Do not declare the warehouse the source of truth without naming the table. "It is in the lake" is how you get three lakes.
Do not let gold-layer marts redefine silver metrics. If finance needs a restated view, it is a new metric with a new name.
Frequently asked questions
Why do my three dashboards show different revenue?
They almost always count different events, grains, or time windows under the same label. Align definition and grain first, then compare freshness and filters.
Is the data warehouse always the single source of truth?
No. The warehouse is a place. A source of truth is a named dataset plus a definition and an owner. Pick the table, not the platform.
Should we delete the dashboards that lost?
Not immediately. Rename them so they cannot collide with the canonical label. Retire them only after the canonical metric has been reused without incident.
How does single source of truth analytics handle late data?
Publish the rule. Either the metric restates history on a schedule, or it is a snapshot as-of a close. Mixing restated and snapshot tiles is a common weekly fight.
Can operational systems stay sources of truth?
Yes, for the events they originate. Analytics should conform and combine those events, not overwrite them with a dashboard calculation.
How do we stop arguing about the number?
Arguments drop once the label only appears on the canonical metric and mismatches are debugged as pipeline issues, not definition debates. The calendar depends on how many colliding labels you still allow.
Book a demo to see a managed data platform running on your own sources
LensAIght
