MEASUREMENT
The frame around the number
Different totals on two screens do not prove that one is wrong. Dates, record types and attribution rules change what becomes visible. Reliable measurement starts by understanding those frames and explaining the differences between them.
Ask what each source sees
An advertising platform shows conversions it attributes to its activity. Web analytics shows interactions it can measure. A CRM shows recorded contacts and outcomes. These scopes differ. In a hypothetical example, two platforms could each attribute the same sale to themselves; adding their reported revenue would double-count it. Agree which record establishes the commercial outcome. Payments or refunds may require finance data. A CRM can also be incomplete or updated late, so its name alone is no guarantee of accuracy.
Give each measure a full definition
A label such as sales conversion is insufficient. Document the numerator, denominator, included statuses, reporting date and observation window. Results grouped by sale date answer a different question from results grouped by enquiry date. Time zones, currencies, cancellations and refunds also belong in the definition. A requested appointment is different from an attended one. Assign an owner and retain definition changes, so apparent performance shifts can first be checked against changes in how the business records and reports them.
Know what one row represents
Before joining tables, ask what a row means. It could represent daily campaign spending, a person or a single conversation. If one person has three conversation records, an incorrect join can multiply the associated spending three times. Define each table’s grain and unique key, then test the relationship and compare totals before and after joining. Keep unmatched records visible. They may expose missing identifiers, incompatible dates or a genuine difference in scope, all of which matter when interpreting the result.
Explain the gap without forcing agreement
Reconciliation should explain differences. Align dates and definitions, then separate duplicate records, missing sources, reporting delays and attribution rules from the unexplained remainder. Attribution assigns credit to recorded interactions; it does not establish what would have happened without advertising. That question requires a suitable experimental design where feasible. A management report should carry the commercial result and the limits of measurement together. Agree how often reconciliation happens and who investigates a new discrepancy, making reliable measurement an ongoing practice.
The question to carry forward
Good measurement makes a number’s meaning clearer. Shared definitions, known record grain and visible limitations let the conversation move from competing screens toward the decision the evidence can support. Precision begins with knowing what is inside the frame.