How to Make Budget Decisions When Your Attribution Sources Disagree
A written budget rule for the period while platform, GA4 and CRM numbers still disagree: choose the governing source, pre-commit actions, record the decision.
This page is part of Zenko's guide to marketing attribution discrepancies.
There is plenty of advice on choosing an attribution model. There is less on what an operator should do with budget this month while the sources still disagree. This page offers a written process. It sets no thresholds and promises no outcomes. Your team and finance partner supply the numbers.
What this page covers, and what it leaves to others
This page covers the period of disagreement. Two existing Zenko pages cover neighbouring ground, so this page links to them rather than repeating them:
- Paid-growth saturation test: marginal versus blended CAC, incrementality and scale, hold, revise and stop rules.
- Pipeline Maturity Curve: comparing channels at the same cohort age.
Step 1: decide which source governs which decision
Do this before opening any report. Write a short table with a decision in one column and the governing source in the other. For example (hypothetical): "Shift budget between two campaigns in the same platform" might be governed by the platform report, while "Report pipeline to the board" might be governed by the CRM. Which source fits which decision is for your team to agree. The reasoning behind each choice goes in the ledger.
Why this works: if you pick the governing source after seeing the numbers, it is tempting to pick the one that supports the move you already wanted.
Step 2: log the disagreement in the ledger
Use the Attribution Disagreement Ledger. Name the cause of each gap, using the documented causes: credit rules, windows, date basis, modelled conversions, counting, visibility and lag. Mark any gap you cannot explain as unexplained.
Step 3: pre-commit an action for each gap band
Your team defines the bands. For each band, choose one of four actions and write it down before reading the result.
| Action | Meaning |
|---|---|
| Hold | Keep spend where it is until the review date |
| Cap | Keep spend but set a ceiling |
| Test | Run an experiment to estimate lift |
| Cut | Reduce or stop spend |
The table below shows the shape of the rule, using blank bands because Zenko supplies none.
| Gap band (set by your team) | Cohort mature? | Unexplained gap? | Pre-agreed action |
|---|---|---|---|
| Band 1: [team to define] | |||
| Band 2: [team to define] | |||
| Band 3: [team to define] |
Step 4: avoid large moves on immature cohorts
If the CRM read is immature, a deep cut based on it may be reversed once stages arrive. Hold or cap until the review date, then judge at a matured cohort age. See the Pipeline Maturity Curve for the method. A hold can also waste money. Keep tracking faults, invalid leads and breached spend ceilings actionable now; do not wait for cohort maturity to address a known problem.
Step 5: know when disagreement justifies an experiment
If an unexplained gap sits on a large share of spend, or the governing source cannot answer the question, the ledger points to a test action. Attribution assigns credit, whereas experiments estimate lift. A 2019 study by Gordon et al. found that observational estimates often differed from experimental results; see the Kellogg research abstract. This is an abstract-based summary, not a full-paper review. That was 15 US experiments on one platform, so it supports the principle, not a size for your account. Lift-test methods are in the paid-growth saturation test.
Step 6: write a decision record for finance
For every budget move, store a one-page record.
| Field | Entry |
|---|---|
| Decision | |
| Governing source and why | |
| Other sources and their gaps | |
| Named causes and evidence | |
| Pre-agreed action and band | |
| Cohort age at decision | |
| Owner | |
| Review date | |
| Restatement log |
If numbers are restated later, add a line instead of rewriting history. See also the weekly acquisition-economics scorecard.
Common mistakes
- Choosing the governing source after seeing the numbers.
- Adopting a published "normal gap" as your tolerance. Google's "often up to 20%" refers to expected discrepancies from different attribution models, not to a budget threshold (Google Ads Help).
- Comparing a period before and after a setting change as though they were like for like. See how lookback settings move your numbers.
- Treating a CRM figure as complete when it cannot see untagged touches. See why your CRM and ad platforms disagree.
What this page does not do
It does not guarantee better returns, supply thresholds, or replace finance review. It makes no claim about any Zenko product.
Sources checked on 18 October 2026
- https://support.google.com/google-ads/answer/6270625
- Gordon et al. (2019), Kellogg School page: https://www.kellogg.northwestern.edu/academics-research/research/detail/2019/a-comparison-of-approaches-to-advertising-measurement-evidence-from/ (the DOI page could not be retrieved)
- Zenko pages linked above, read on 18 October 2026