Why this matters
When three dashboards show three numbers for one word, the argument is about the definition, not the business. The guide to metrics layers and definitions lists the sources of drift and the fields a definition must settle; the guide to data flow and integrations shows where in the stack each number is born; a definition is only enforceable once you know which system can produce it.
The provider and its conflicting numbers
Syntaxfern Software Training is an invented provider of online and classroom software courses for businesses. Individuals buy online courses through a checkout; companies book classroom seats through a sales adviser, often for courses starting weeks later.
Syntaxfern Software Training: one month, three dashboards
| Metric | Marketing dashboard | Sales dashboard | Finance report |
|---|---|---|---|
| Leads | 640 | 480 | n/a |
| Enrolments | 210 | 260 | 245 |
| Revenue | £98,000 | £131,000 | £112,000 |
- Marketing: Looker Studio on the analytics connector. Leads are lead key events, including brochure downloads and repeat submissions. Enrolments are online checkout orders. Revenue is online purchase revenue including VAT, on the event date.
- Sales: CRM reports. Leads are new person records, deduplicated by email, excluding existing customers. Enrolments are seats on won deals, by close date. Revenue is won deal value excluding VAT, by close date.
- Finance: the ledger. Enrolments are seats on courses delivered in the month with paid invoices, after cancellations. Revenue is recognised on the course delivery date, net of VAT and refunds.
Data flow: web forms feed the analytics property and the CRM; the online checkout writes to the booking system, which fires the analytics purchase tag; sales-led bookings go from the CRM to the booking system by hand; the booking system raises invoices into the ledger. The marketing dashboard blends three connectors and is slow to open.
Tracing the gaps:
- Leads: 640 − 110 brochure downloads − 30 repeat submissions − 20 existing customers = 480.
- Enrolments: 260 seats closed − 25 for next month's courses − 5 cancelled + 15 sold last month for this month = 245. The 210 are online orders, some with several seats, and exclude sales-led bookings.
- Revenue: marketing's £98,000 includes VAT and covers online only; sales' £131,000 includes deals for future courses.
Anatomy of a governed definition
A governed definition has a stable key, one plain sentence, a formula built from other defined metrics, a grain and its roll-up rule, the filters always baked in, the date and zone a row is counted on, a source of truth, an owner, a version and a known-differences note. When one word legitimately carries two meanings, such as revenue booked and revenue recognised, it becomes two metrics with two names, never one metric with a toggle. The source of truth is the system closest to the event, chosen from the data flow rather than from the loudest team.
Task
- Decide which disputed words need one definition and which need splitting into two named metrics.
- Write each definition with key, meaning, formula, grain, filters, time rule, source, owner and known differences.
- Say what happens to the marketing dashboard's current lead and enrolment figures.
- Plan versioning and enforcement, including how the dashboards read the definitions.
- Answer a vendor who says a customer data platform would fix the problem.
Worked answer
Every team counts correctly by its own rules; the fix is naming and ownership.
| Key | Meaning | Grain and time rule | Source of truth | Owner | Known differences |
|---|---|---|---|---|---|
form_submissions | Every web form completion, any type | Submission, by event date | Analytics property | Head of marketing | Higher than leads: includes downloads, repeats and customers |
leads | New prospective customers from any enquiry route | Person, by CRM creation date | CRM | Head of marketing, agreed with sales | Lower than form submissions; deduplicated by email |
seats_booked | Learner seats on confirmed bookings, net of cancellations | Seat, by booking date | Booking system | Sales director | Runs ahead of finance for courses starting later |
seats_delivered | Seats on courses delivered and invoiced | Seat, by delivery date | Booking system and ledger | Finance director | Lags bookings by the gap to the course date |
revenue_booked | Value of confirmed bookings, excluding VAT | Booking, by booking date | Booking system | Sales director | Includes future courses; not recognised revenue |
revenue_recognised | Revenue on delivered courses, net of VAT and refunds | Seat, by delivery date | Ledger | Finance director | The only figure for financial reporting |
The marketing dashboard. Its lead figure is renamed form submissions and stays as a channel signal. Its online orders become an online slice of seats booked, read from the booking system rather than from analytics (GA4Google's event-based analytics platform.), so classroom seats are no longer invisible to marketing.
Versioning and enforcement. Record every definition in a KPI dictionary with version and date, and mark each company KPIA metric tied directly to an objective. as certified once its owner signs it off. Build one modelled table per metric from the named source and point every dashboard at those tables rather than at raw connectors. That also fixes the slow dashboard, as the Looker Studio performance guide explains for blends. A change needs the owner's approval, a new version, restated history and a dashboard note. Each month, check that seats delivered reconcile to the ledger and that leads reconcile to CRM records.
The vendor. A customer data platform (CDPUnified first-party customer profiles.) resolves identity and activates segments; it does not decide what revenue means, and added now it would produce a fourth number.
Common mistakes
- Choosing a winner instead of naming meanings. Finance's revenue and sales' revenue are both needed.
- Defining on raw columns. Seats booked should feed both cost per seat and revenue booked.
- Letting dashboards compute their own versions. One path from source to chart, or the drift returns.
- Buying a tool to settle a definition. Agreement comes first; enforcement follows.
Next, bring this together with warehouse data and causal tests in the measurement strategy capstone, or revisit dashboard design at the intermediate analytics level.
