The marketing leadership pack shows MQLs (Marketing Qualified Lead) up 12% quarter on quarter. The CFO’s pack, distributed the same morning, shows them down 3%. Both numbers come from the same source system. Both have been signed off by the analyst who built the report. Neither is wrong in the sense that either can be reproduced from the underlying data. They are answering slightly different questions, and the headers do not say so.
This is not a story about lying dashboards. It is a story about dashboards that are technically truthful, individually defensible, and collectively useless. The CMO walks out of the meeting unable to decide whether MQLs are up or down. The CFO walks out unsure whether marketing is hitting its commitments. The analyst goes back to their desk and tries to figure out which version to defend in the follow-up.
Marketing dashboards drift for five engineering reasons. None of them are anyone’s fault in the moral sense. All of them are fixable in the engineering sense. The first job for a marketing leader is to know which one is happening, so the conversation with the data team can move from “the dashboard is wrong” to “the freshness window is six hours, the campaign team is doing sends every two”.
The pack discrepancy
Look closely at any two marketing reports from the same week and the differences start to make sense. The leadership pack pulled data on Tuesday morning from a warehouse view that refreshes nightly. The CFO pack pulled data on Tuesday afternoon from a finance-side replica that refreshes hourly but applies a different lead-stage filter. The leadership pack counts an MQL the moment a contact crosses the score threshold. The finance pack counts an MQL only after the SDR (Sales Development Representative) has logged a discovery call.
These are not bugs. They are different definitions, applied to different snapshots, served through different pipelines, with no shared header that says so. The chart in the leadership pack says “MQLs”. The chart in the CFO pack also says “MQLs”. The reader has no way to tell them apart.
Multiply this across twenty dashboards, three audiences, and five quarters of trend data, and the result is what most marketing leaders actually experience: a collection of charts they have stopped fully trusting, used mostly to confirm decisions already made for other reasons. The dashboards have not failed. They have done exactly what they were built to do, which is each answer a slightly different version of a slightly different question.
The five reasons dashboards lie
Freshness
Every dashboard sits behind a pipeline that updates at some cadence. The cadence is rarely what the dashboard suggests. A “real-time” dashboard usually refreshes every fifteen minutes. A “daily” dashboard usually refreshes overnight, twelve to fourteen hours after the events it shows. A “monthly” dashboard usually closes the previous month three to five working days after month-end, depending on how stable the upstream finance reconciliation is. None of this is unreasonable. All of it is invisible to the reader unless explicitly surfaced. “Sales last week” depends entirely on which Tuesday afternoon the pipeline ran.
Identity drift
The set of customers in any segment changes between dashboard refreshes. A customer who unsubscribed yesterday is still in last week’s “active email contactable” count. A duplicate record merged on Monday changes the historical count for every snapshot before Monday, unless the merge is handled with proper history-preserving logic in the warehouse, which it usually is not. Trend lines move because the underlying definition of who counts has shifted, not because anything in the real world changed.
Transformation lag
Between the raw event and the dashboard number there are transformations: deduplication, joining, attribution model application, segment classification, currency conversion, exchange rate snapshotting. Each transformation has its own logic and its own update cadence. A change to the attribution model on Monday produces different historical numbers on Tuesday’s dashboard than Friday’s, because the model now applies retroactively to past events. The dashboard reader has no signal that the underlying logic moved.
Definition drift
The most common cause of dashboard discrepancy is that the same word means slightly different things in different places. “MQL” in the marketing automation platform is a score threshold. “MQL” in the CRM is a lifecycle stage set by a human. “MQL” in the warehouse is whichever of the two the data team mapped to the canonical field, with rules that have probably been revised quietly since the dashboard was first built. Three dashboards, three definitions, one label.
Source-of-truth ambiguity
When the same metric exists in multiple systems, dashboard builders pick the source that is easiest to query, not necessarily the source that is authoritative. Email opens in the email platform differ from email opens in the warehouse, because the platform deduplicates one way and the warehouse another. Revenue in the CRM differs from revenue in the billing system, because one is invoiced and the other is collected. Picking the wrong source is invisible until someone notices that two charts disagree.
What good looks like
Marketing dashboards that do not lie share a small set of properties. The metric definition is captured in code, version-controlled, and the dashboard surfaces a link to the current definition. The freshness timestamp is visible on the chart, not buried in a tooltip. The source system is named. When the definition changes, the change shows up in a changelog, and historical numbers are either recomputed consistently or fenced off behind a “definition changed” line on the trend.
These are unglamorous properties. They are also the difference between a dashboard the CMO can use in a board meeting and a dashboard the CMO has to apologise for in a board meeting. The teams that produce trustworthy dashboards have usually done the harder work upstream: a single warehouse model that every dashboard pulls from, a small canonical metrics layer that defines MQL once and only once, and an explicit refresh schedule that is communicated rather than discovered.
A useful shorthand: when two dashboards disagree, the question to ask is not “which is right” but “which definition, which freshness, which source”. When the data team can answer those three questions for any chart in under five minutes, the dashboards have probably stopped lying. When the answer involves a Confluence search and a Slack thread from 2024, they have not.
The unsexy work of marketing observability
The infrastructure world solved a version of this problem a decade ago and called it observability. The same idea applies to marketing data, and almost no marketing team has it. Observability for marketing data means three things: the pipelines are monitored, so when a feed stops or runs late someone is paged before the CMO is; the definitions are testable, so a change to the MQL rule fails a test before it ships to production; and the lineage is visible, so any number on any dashboard can be traced back to the raw event it came from.
This is not a tool purchase. It is a modest amount of platform engineering, a willingness to write tests for business logic, and an operating model where someone owns the pipeline the same way someone owns a production service. Keeping it running quietly after launch is the day-job of an operations team, which is the role Sakura’s Managed Services play for a number of marketing data platforms in production. It is also where most marketing data programmes underinvest, because the work does not produce a new campaign or a new persona. It produces a quieter Friday afternoon and a CMO who does not have to apologise for the numbers.
Marketing dashboards that tell the truth are an engineering artefact, not a tooling outcome, and building the substrate underneath them is the kind of work Sakura’s Data & AI practice is set up to do.