On this page12 sections
Growth Operations ResearchA sample-based consistency test, not an attribution benchmark.

Testing Whether Lead Source Values Stay Consistent Across the Form, CRM, and Billing Record

A bounded test for comparing the lead source recorded by the form, the CRM, and the billing record for the same customer.

Published ••6 min read
Testing Whether Lead Source Values Stay Consistent Across the Form, CRM, and Billing Record research illustration

Why source consistency deserves a test

Lead source is often treated as a single fact, but the same customer can carry several source values in different systems. The form may store a campaign parameter, the CRM may store an original source property, and the billing record may store a free-text reference. Each system answers a slightly different question, and none of them is automatically the truth. A consistency test does not decide which value is correct. It shows where the values diverge, how often, and whether the divergence follows a pattern that a team can repair.

Define the three records under comparison

Start by naming the exact objects being compared. For the form, identify the submission record and the field that captured the source. For the CRM, identify the contact or company record and the property used for original source. For billing, identify the invoice, subscription, or customer record and the field that carries the source or campaign reference. Write down whether each value is a controlled list, free text, or a derived label. Comparing a controlled list with free text will produce mismatches that are not really data errors, so the definition has to come first.

Build a comparison key that survives handoffs

The three systems rarely share one identifier. Use a stable key such as a normalized email address or an internal customer number, and document the normalization rule. Lowercase, trim, and handle plus-addresses consistently, but do not merge distinct people who share a company inbox. Where the CRM stores multiple contacts for one account, decide in advance whether the comparison happens at the contact or account level. A weak key creates false mismatches and can make a healthy process look broken.

Capture the raw form value

Extract the source value exactly as the form recorded it, including parameters that look empty or generic. Save the submission timestamp and the landing page URL alongside it. These two fields often explain a value that seems wrong: a direct visit with a campaign parameter still carries that parameter, while a paid click that lost its parameter may appear as direct. Do not rewrite the raw value during extraction. Store a cleaned version in a separate column so the original evidence remains available for review.

Record the CRM original source

The CRM's original source property usually records the first known source and does not change when later activity occurs. That behavior is useful for a consistency test because it should remain stable. Capture the property value and the date it was set, then note whether the system derives it from a hidden field or from a manual entry. If the property is editable, include who last changed it when that history is available. A source value that changes without a documented reason is itself a finding worth recording.

Compare billing and invoice fields

Billing records are created later and often by a different team, so they are where source values drift most visibly. Capture the field that carries the source or campaign reference and whether it is required at invoice creation. If billing uses a free-text note, record the note rather than a guessed category. Compare the billing value with the CRM value using an explicit mapping table, and show unmapped values as their own category instead of forcing them into a match. The goal is an honest account of agreement, not a flattering percentage.

Classify mismatches before counting them

Create a small set of mismatch types before reviewing the sample. Useful types include a missing value in one system, a controlled value that maps to a different category, a free-text variant of the same source, and a genuinely different source. Add a type for records where the customer had more than one meaningful touch and the systems captured different ones. Classification turns a pile of differences into a diagnosis. It also prevents a reviewer from treating every spelling variation as a serious attribution failure.

Test a sample end to end

Select a sample that covers the main forms, campaigns, and time periods in scope. For each record, trace the customer from the form submission through the CRM record to the billing record, and record each value with its timestamp. Do not stop at the first match or mismatch; complete the chain so the comparison is consistent. Keep a short note for any record that required judgment, such as a merged contact or a corrected invoice. Those notes are the most useful part of the evidence when someone questions the result later.

Quantify agreement without inventing a benchmark

Report the share of records that agree across all three systems, the share with a missing value, and the share with a classified mismatch. Report the counts alongside the percentages so a small sample cannot look more precise than it is. Do not describe the result as an industry benchmark or a target. It is a description of this sample at this time. If the team wants a target, it should set one after seeing the baseline and after agreeing which mismatch types are acceptable.

Investigate the pattern, not just the total

A single agreement number hides where the problem lives. Break the result down by form, campaign, CRM entry path, and billing team. Look for a form that never passes a campaign parameter, a CRM import that overwrites original source, or a billing template that invites free text. These patterns point to specific repairs. A team that only tracks the total will keep re-cleaning records without changing the process that creates the differences.

Choose a repair and retest

Pick one mismatch type with a clear cause and fix it at the source. That might mean adding a hidden field, locking a CRM property, or replacing a free-text billing field with a controlled list. Retest with the same key, the same sample rules, and the same classification. Compare the new result with the baseline and report what changed. A repair that reduces one mismatch type without creating another is a real improvement, even if the overall agreement number moves only slightly.

Document limits and data handling

This test cannot establish which source caused a sale, and it cannot reconcile every customer touch. It also depends on the quality of the comparison key and on access to billing records that may sit outside the marketing team. State those limits in the method note. Handle customer data carefully: extract only the fields needed, restrict access, and delete the working sample when the review is done. A consistency test is only repeatable if the people who run it can do so without exposing personal data or depending on one person's memory.

Sources

  1. https://support.google.com/analytics/answer/1033863
  2. https://knowledge.hubspot.com/records/understand-original-source-properties
  3. https://www.w3.org/TR/NOTE-datetime
  4. https://www.census.gov/data/developers/guidance/api-user-guide.html
  5. https://www.ftc.gov/business-guidance/privacy-security