Your ad platform, analytics dashboard and CRM can tell different stories about the same sale. A useful attribution model makes those differences understandable.
The goal is a repeatable way to assess marketing, with clear rules and honest limits. Start with the decision you need to make, then choose the simplest approach your data can support.
Decide what you need attribution to explain
Marketing attribution assigns credit to recorded interactions before an enquiry, sale or another defined outcome. A model is a set of rules for dividing that credit. It is useful when everyone understands the question it answers.
Start with one decision: which campaigns deserve investigation, which channels introduce suitable prospects, or where to test the next budget increase. Then define the outcome. A form submission, a qualified opportunity and a closed sale are different events, and they should not share an ambiguous “conversion” label.
Assigned credit is not proof of additional sales. Attribution describes a measured journey. To estimate what would have happened without advertising, you need a suitable experiment or another causal measurement approach.
Google’s Conversion Lift guidance describes comparisons between exposed and control groups. Such studies answer a different question from routine attribution reporting.
Choose a model your team can maintain
Match your reporting approach to the buying cycle and the interactions you can reliably record. A longer journey needs context about discovery and follow-up, but adding more weights cannot recover missing activity.
Common attribution approaches and their limits| Approach | How credit is assigned | Main limitation |
|---|---|---|
| First touch | All credit to the earliest eligible recorded interaction | Misses what helped the buyer progress later |
| Last touch | All credit to the final eligible recorded interaction | Can overlook earlier discovery and education |
| Linear | Equal credit across eligible recorded interactions | Assumes every interaction matters equally |
| Position-based | Defined shares for the first, last and middle interactions | Weights are chosen assumptions |
| Time-decay | More credit to interactions nearer the outcome | May undervalue work done early in a long cycle |
| Data-driven | A platform model allocates credit using available journey data | Depends on that platform’s coverage and methodology |
Check your tool before choosing. First-click, linear, time-decay and position-based models have not been available in GA4 attribution reports since November 2023. They remain useful concepts for separately implemented reporting, rather than current GA4 settings. See Google’s attribution models report documentation.
Google currently documents data-driven, paid and organic last click, and Google paid channels last click in its attribution reports. Eligible interactions and direct-traffic rules matter, so do not assume a platform’s last-click definition matches a simple spreadsheet. Read Google’s model definitions before comparing results.
See what changes in one example
Illustrative example: over a 60-day buying cycle, a customer clicks a paid social ad, later clicks a webinar invitation email, and finally clicks a paid search ad before making a $10,000 purchase. Assume all three interactions are recorded and eligible.
Same $10,000 sale, different allocation rules| Model | Paid social | Paid search | |
|---|---|---|---|
| First touch | $10,000 | $0 | $0 |
| Last touch | $0 | $0 | $10,000 |
| Linear | $3,333.33 | $3,333.33 | $3,333.34 |
| Position-based, 40/20/40 | $4,000 | $2,000 | $4,000 |
The business still earned $10,000. Only the allocation changed. The linear example includes a one-cent rounding adjustment; the 40/20/40 weights are an example, not a recommended universal split.
Use this kind of comparison to challenge a conclusion before moving money. A channel losing credit under a different rule has not suddenly stopped contributing. Conversely, being present on a journey does not prove it caused the sale.
Make the data dependable first
Write down the definitions your reports depend on. Give each one an owner who can explain how it is captured and what is missing.
- Campaign names: use a consistent convention for source, medium and campaign.
- Outcome definitions: separate enquiries, accepted opportunities and won customers.
- Record matching: link contacts and opportunities carefully, and avoid counting the same sale twice.
- Time windows: state how far back interactions are eligible and whether reports use interaction dates or outcome dates.
- Unknown sources: keep an explicit unknown category instead of guessing where a lead came from.
- Offline activity: document referrals, events and sales conversations that web analytics may not capture.
UTM parameters identify campaign traffic, but values are case-sensitive. A shared naming convention prevents labels such as “Email” and “email” from fragmenting reporting. Google’s campaign URL guidance explains the parameters and how to use them.
Test an enquiry from campaign link to CRM record. Check redirects, form submission and source fields. A model comparison is much less useful when the underlying handoff loses the source.
Give leadership a report they can act on
1. Start with business outcomes
Show qualified opportunities, won customers and the time it takes prospects to progress. If you use blended customer acquisition cost, define which sales and marketing costs are included and which customers belong in the denominator. Keep that definition stable.
2. Explain the model’s blind spots
State the reporting window and known gaps. A rise in branded search may follow earlier awareness activity. Missing partner tags may obscure referrals. Treat those patterns as questions to investigate, not automatic evidence that another channel deserves the credit.
3. End with a decision and a test
For each proposed budget change, show the evidence, the uncertainty and the result you will look for next. Do not add platform-attributed revenue together as though each platform counted a different set of customers.
Keep one baseline and review deliberately
Choose one primary reporting model for routine decisions and a second view for context where useful. Review on an agreed schedule, such as quarterly, rather than switching whenever the results are uncomfortable.
Before changing the model, preserve the baseline and record the switch date, reason and affected reports. Where possible, compare both models over the same historical period. Distinguish a credit reallocation from an actual change in performance.
Give marketing, sales and finance a way to question the assumptions with evidence. A stable reporting process should make disagreement easier to resolve, without pretending the data records every influence on a purchase.
ThinkSwift’s digital marketing team helps connect attribution choices with analytics, advertising and CRM reporting so the definitions remain understandable across teams.
Frequently asked questions
01Which attribution model should a small business start with?
Start with a model your tools support and your team can explain. A simple baseline with documented limits is useful while you improve data quality. For a longer sales cycle, add a view of earlier interactions before using last-touch results to reduce discovery activity.
02Can we run two models?
Yes. Keep one as the primary reporting baseline and use another to test how sensitive the conclusions are to the rules. Do not add their credited revenue together.
03Is data-driven attribution worth using?
It can provide a useful comparison when your platform supports it. Check which interactions and outcomes it covers, then compare its conclusions with CRM results and a simpler baseline. Availability alone does not establish that every budget recommendation is reliable.
04Can we use time-decay or position-based attribution in GA4?
Not in GA4’s current attribution reports. Google removed those models, along with first-click and linear attribution, in November 2023. A separate reporting system can implement its own rules if the necessary data is available.
05Will changing models break our targets?
It can change channel-level credit and make comparisons misleading. Preserve the prior baseline, document the change and agree how targets will be compared or reset. Overall sales should still reconcile to the same business records.
06Does attribution tell us which sales marketing caused?
An attribution report alone is not sufficient to establish incremental sales. Use it to form hypotheses, then use an appropriate controlled test or other causal method when the decision calls for that evidence.
Make your next budget review easier.
Talk to us about an attribution approach that fits your sales cycle, tools and available data.
Talk it through