
How to Measure Marketing Effectiveness
On this page
- Stage 1: Separate platform claims from real leads
- Stage 2: Repair UTM tags and use one naming convention
- Stage 3: Count submitted leads and wire the CRM
- Stage 4: Build a weekly channel scorecard
- Stage 5: Read self-reports and GA4 as two witnesses
- Stage 6: Test a Meta pause with a geographic holdout
- Stage 7: Make budget decisions and keep the review small
Marketing
To measure marketing effectiveness, start with real leads in your CRM, then compare their sources, costs and eventual sales. You do not need multi-touch attribution software or a data team to make the first useful decisions. In the example here, platforms claimed 101 conversions in four weeks while the CRM contained 64 real leads. Counting those claims as separate customers would make the marketing look much better than it was.
Brightwell Commercial Cleaning is fictional, and every number assigned to it is a hypothetical illustration. It has 45 staff, sells office cleaning contracts in two regions called North and South, and spends $5,800 on marketing per four-week period. Its average contract is $1,200 per month at a 30% gross margin, leaving $360 in monthly gross margin per customer.
Multi-touch attribution can help when a company has enough data and a team to maintain and interpret it. With 64 leads in a period, Brightwell has more immediate problems to solve: overlapping claims, inconsistent campaign tags and an event that counts people who started a form. GA4 and other platform documentation cited here was read on 29 October 2026.
Stage 1: Separate platform claims from real leads
Brightwell begins with reports that appear to show abundant demand. They do not agree on what a conversion means, and the CRM has a smaller count of actual inquiries.
Hypothetical table: four-week reports before cleanup
| Report | Figure reported | What it measures |
|---|---|---|
| Google Ads | 38 conversions | Conversions credited by Google Ads |
| Meta Ads Manager | 41 results | Results credited by Meta |
| SEO freelancer report | 22 organic conversions | Conversions credited to organic search |
| Sum of the three claims | 101 | Overlapping claims, not a count of distinct leads |
| CRM | 64 new leads | 52 submitted web forms and 12 phone calls |
| Email tool | 52% opens; 2.1% clicks | Reported email activity |
The three claims exceed the CRM count by 37, or about 58% of the 64 leads. That difference does not prove exactly which leads were counted twice. Each platform applies its own attribution rules, so one person who interacts with more than one channel can appear in more than one report.

The platform claims add up to 101, while the CRM has 64 real leads. Overlapping credit can make the combined platform total taller than the lead count.
Use each platform report to investigate its own campaigns. Use deduplicated CRM leads to count inquiries across the business, and check what each platform calls a conversion before comparing figures.
The CRM count also needs a definition. Brightwell counts a new inquiry once, whether it arrives through a submitted form or a phone call. A repeat inquiry from the same person needs a consistent treatment in the CRM; otherwise even the count used as the denominator can drift.
The GA4 report creates a separate problem. Brightwell has marked form_start as a key event, so the Traffic acquisition report counts starts rather than submitted inquiries.
Hypothetical table: GA4 Traffic acquisition before the event fix
| Session default channel group | form_start key events |
|---|---|
| Paid Search | 46 |
| Organic Search | 34 |
| Direct | 28 |
| Unassigned | 12 |
| Organic Social | 11 |
| Paid Social | 9 |
| Total | 140 |
Those 140 events are not 140 people asking for a quote. A person can start without submitting, and an event count can include repeated actions. Brightwell has 52 submitted forms in the CRM for the same period, so form_start is useful for diagnosing form behavior but is a poor headline lead count.
The email report needs a correction too. Apple says Mail Privacy Protection downloads remote content privately in the background. Our reading is that this can register an open even when nobody reads the message. Its protection does not extend to links. Brightwell keeps the 2.1% click rate and records replies, while removing the 52% open rate from its decision scorecard. The same distinction applies when you build an owned audience with email.
What does Traffic acquisition attribute?
Google says the Traffic acquisition report's default dimension is Session default channel group, and that the dimension uses paid and organic last click. Our reading of Google's definitions is that this makes it a session-level paid and organic last-click view. That remains the useful reading even when the property's reporting attribution model is data-driven.
GA4 Attribution reports offer three models: Data-driven attribution, Paid and organic last click, and Google paid channels last click. Google removed the first click, linear, time decay and position-based models in November 2023. Google's attribution model documentation explains the remaining models and their treatment of credit.
Data-driven attribution is the property default. It distributes credit using data for each key event, with a model specific to the property and its events; Google's documentation does not state a minimum data requirement. That does not mean a small set of leads can settle every budget question. A model can allocate observed credit without proving how many leads an advertisement caused.
Paid and organic last click is also called last non-direct click. All three attribution models exclude direct visits from credit unless the entire path is direct. Direct is therefore not a reliable synonym for someone typing the company URL: an untagged link or missing referral information can also leave a visit classified as Direct.
The dimension name matters. Default channel group uses the property's attribution model, while Session default channel group and First user default channel group use paid and organic last click. A report grouped by one dimension cannot be assumed to answer the same question as a report grouped by another. Google's default channel group definitions specify those differences.
The default lookback window is 30 days for the acquisition key events first_open and first_visit, and 90 days for other key events. Credit can be reattributed for up to seven days after an event, so a very recent report can move. A user needs the Marketer role or above to change attribution settings. See GA4 attribution settings before changing a window.
Stage 2: Repair UTM tags and use one naming convention
Brightwell's campaign links contain enough information to confuse both a person and GA4. The company first puts every campaign link in one shared sheet, then assigns a consistent source, medium and campaign to each non-Google Ads link it controls.
Hypothetical table: campaign tags before and after
| Link | Before | Likely channel before | After | Channel after |
|---|---|---|---|---|
| Newsletter | utm_source=mailchimp&utm_medium=newsletter | Unassigned | utm_source=newsletter&utm_medium=email&utm_campaign=monthly-tips | |
| Meta ad | utm_source=Facebook&utm_medium=Paid_Social, with facebook used elsewhere | Paid Social, split into two source rows | utm_source=facebook&utm_medium=paid_social&utm_campaign=office-deep-clean&utm_content=video-a | Paid Social |
| LinkedIn sponsored post | utm_source=linkedin&utm_medium=sponsored | Organic Social | utm_source=linkedin&utm_medium=paid_social&utm_campaign=office-deep-clean | Paid Social |
| LinkedIn company post | No tags | Organic Social, or Direct if the app drops the referrer | utm_source=linkedin&utm_medium=social&utm_campaign=monthly-tips | Organic Social |
| Partner page | No tags | Referral | utm_source=partner-name&utm_medium=referral&utm_campaign=partner-program | Referral |
| Google Ads | Manual UTMs | Varies with setup | Auto-tagging, no manual UTMs | Paid Search (Search campaigns) |
The “likely channel before” column follows Google's published rules. mailchimp as the source and newsletter as the medium match none of the stated Email rules, so the expected result is Unassigned. Email matches a source or medium of email, e-mail, e_mail or e mail.
Paid Social requires a source on Google's social-site list and a medium matching its paid-medium pattern, ^(.*cp.*|ppc|retargeting|paid.*)$. paid_social qualifies; sponsored does not. With linkedin as the source, the sponsored post can then fall through to Organic Social. Organic Social also recognizes mediums such as social, social-network, social-media and sm.
Referral recognizes mediums including referral, app and link. Direct has source (direct) and medium (none) or (not set). These labels describe how GA4 classified the session; they do not certify what the buyer was thinking when the visit began.
Google defines AI Assistant as “the channel by which users arrive at your site from sources like ChatGPT, Gemini, Deepseek, Copilot, or Grok.” A medium of exactly ai-assistant also qualifies. Google does not publish its full referrer list, and its AI Overviews and AI Mode count in Organic Search.
Brightwell's convention is deliberately dull:
- Use lowercase values everywhere.
- Use hyphens inside campaign, source and content values, such as
office-deep-clean,partner-nameandvideo-a. - Take mediums exactly as written from a fixed list:
email,paid_social,social,referralandcpconly outside Google Ads.paid_socialkeeps its underscore because it is a fixed list value that matches Google'spaid.*medium pattern. - Name campaigns for the offer or theme; use
utm_contentto identify a creative or link variant. - Record the finished link, owner and destination in one shared sheet before publishing it.
The lowercase rule prevents split reports. Google's channel definitions are not case sensitive, but UTM values are: Facebook and facebook can land in the same channel while appearing as separate source rows.
Google lists utm_source, utm_medium and utm_campaign as required parameters. It strongly recommends utm_id and utm_source_platform; utm_term and utm_content are optional. The Google URL builder documentation also says utm_creative_format and utm_marketing_tactic are not currently reported in GA4 properties.
For linked Google Ads accounts, use auto-tagging for Google Ads clicks. After publishing a tagged link, test it and inspect the resulting source, medium and channel in GA4. A naming sheet is useful only if the links actually follow it. Put the channel definitions and ownership into a measurement plan so the next campaign does not start a second naming system.
Stage 3: Count submitted leads and wire the CRM
A GA4 key event is an event marked as important to the business. For Brightwell, the website lead event should represent a successful form submission, while the CRM should record the actual inquiry and what sales did with it.
Hypothetical table: event and CRM setup
| Item | Brightwell's rule | Observable check |
|---|---|---|
form_start | Keep it for form analysis; unmark it as a key event | Starting and abandoning a form does not create a submitted lead |
generate_lead | Fire on the successful submission or thank-you state; mark as a key event | A test submission creates one event and one CRM lead |
| Hidden UTM fields | Copy landing-session utm_source, utm_medium and utm_campaign into the lead record | A tagged test visit produces the expected values on the record |
| Self-reported source | Require a “How did you hear about us?” picklist with free text | The submitted record contains an answer and any explanation |
| CRM lead source | Use a fixed picklist with the scorecard's channel names | Each new lead has one source or is explicitly Unknown |
| Phone inquiries | Ask callers how they heard about Brightwell and enter the answer | Phone leads appear in the CRM alongside form leads |
| Qualified lead | Office is at least 2,000 square feet and inside the service area | Sales sets the qualification status within two working days |
| Later lead events | Optionally send qualify_lead and close_convert_lead | GA4 receives those events when the corresponding status is set |
In a standard GA4 property you can mark up to 30 key events. Go to Admin > Data display > Events and use the star beside the event. Marking an event does not change historic data, so Brightwell starts its clean comparison on the day the change is made. Google's key event instructions describe the limit and the marking process.
A click-to-call interaction can be tracked with a custom event such as phone_click, but a click is not proof that a call connected or produced a lead. The CRM record and the caller's answer provide the business count. Give staff a consistent way to enter calls so the 12 phone inquiries do not disappear when the web form is repaired.
GA4's Reports > Acquisition > Lead acquisition report has its own definitions. New leads counts unique users with generate_lead, Qualified leads counts users with qualify_lead, and Converted leads counts users with close_convert_lead. Those are user-based GA4 metrics, so they need not match a CRM's count of lead records exactly. The Lead acquisition report documentation lists the event-to-metric mapping.
Keep GA4 key events and Google Ads conversions distinct when configuring measurement: key events support GA4 analysis, while Google Ads conversions support bidding and cross-channel measurement. Before changing a bidding conversion, check which campaigns depend on it.
The form, event and hidden fields all need to work together on the site. If that capture needs rebuilding, Iron Goo's website foundation service is relevant to that implementation. Whatever the site setup, submit a test lead from a tagged link and inspect both the GA4 event and the CRM record before trusting the numbers.
Stage 4: Build a weekly channel scorecard
Brightwell assigns each CRM lead one lead source and reviews spend, lead quality and customers together. The table covers the first hypothetical four-week period after the Stage 3 fixes, with sales outcomes shown for the leads in that period.
Hypothetical table: four-week scorecard by CRM lead source
| Channel | Spend | Leads | Qualified | Customers | CPL | Cost per qualified lead | CAC | Payback, months |
|---|---|---|---|---|---|---|---|---|
| Google Ads | $3,000 | 24 | 12 | 4 | $125.00 | $250.00 | $750.00 | 2.08 |
| SEO content | $1,500 | 14 | 8 | 3 | $107.14 | $187.50 | $500.00 | 1.39 |
| Meta Ads | $1,200 | 10 | 2 | 0 | $120.00 | $600.00 | Not calculable | Not calculable |
| Email newsletter | $100 | 4 | 3 | 1 | $25.00 | $33.33 | $100.00 | 0.28 |
| Referrals and partners | $0 | 7 | 6 | 3 | $0 | $0 | $0 | 0 |
| Unknown | Not assigned | 5 | 1 | 0 | Not calculable | Not calculable | Not calculable | Not calculable |
| Total | $5,800 | 64 | 32 | 11 | $90.63 | $181.25 | $527.27 | 1.46 |
CPL = channel spend / leads cost per qualified lead = channel spend / qualified leads CAC = channel spend / customers payback months = CAC / monthly gross margin per customer
For Google Ads, $3,000 divided by 24 leads gives a $125 CPL. Dividing the same spend by 12 qualified leads gives $250 per qualified lead; dividing by four customers gives a $750 customer acquisition cost, or CAC. At $360 monthly gross margin per customer, $750 divided by $360 gives about 2.08 months of marketing spend payback.
That payback calculation uses the stated gross margin and marketing spend. It is not a cash-flow forecast, and it does not include costs that Brightwell has not put in the table. If contract margins differ materially, use the relevant margin for the customers in each channel.
Meta's $120 CPL is close to Google Ads' $125, but only two of its 10 leads meet the qualification rule. Its cost per qualified lead is $600, more than twice Google Ads' $250. With zero customers, dividing by zero cannot produce a meaningful CAC or payback; do not put $0 in that cell.
The referral row has zero recorded marketing spend for this period, so the arithmetic produces zero acquisition cost. That does not mean referrals require no staff time or will stay free after a reward is introduced. Unknown contains five leads because forcing an unsupported source onto a record would make the table look more complete while making it less accurate.
Four Google Ads customers are a small sample. Review the scorecard weekly to catch broken tracking and slow follow-up, but base budget decisions on rolling 12-week figures where possible. Check whether later-closing leads from the period have been updated before comparing CAC.

A 20-minute review puts spend, qualified leads and customer cost side by side. Reading the columns together gives the team a basis for deciding where money goes.
A low CPL alone does not establish whether a campaign can pay for itself. Use gross margin and the rate at which leads become customers to set a break-even threshold; the worked method for paid acquisition on a small budget helps with that calculation and budget setting.
Stage 5: Read self-reports and GA4 as two witnesses
Brightwell asks every lead how they heard about the company and keeps that answer beside its GA4 session view. The two counts have different populations: self-reports include all 64 leads, including calls, while the GA4 column describes 52 web leads.
Hypothetical table: self-reported source and GA4 session channel
| Self-reported source, 64 leads | Count | GA4 Session default channel group, 52 web leads | Count |
|---|---|---|---|
| Google search | 20 | Paid Search | 18 |
| A colleague or friend | 15 | Organic Search | 12 |
| Facebook or Instagram | 6 | Direct | 10 |
| 6 | Paid Social | 6 | |
| ChatGPT or another AI assistant | 5 | 3 | |
| Newsletter | 3 | Referral | 2 |
| Other or blank | 9 | AI Assistant | 1 |
The rows sit beside each other for space; they are not matched pairs. A person who says “Google search” may have clicked an ad or an organic result, while GA4 separates those sessions into Paid Search and Organic Search. A person who names a colleague may later search for Brightwell or arrive through an untagged link.
GA4 cannot directly see a conversation between colleagues. Some of those 15 word-of-mouth mentions may later appear as Direct sessions, but the table does not establish which ones. Direct can also contain other visits whose source was unavailable.
The five AI-assistant answers and one AI Assistant session are similarly different observations. An assistant can recommend a business without sending a click, and the person may later find the site through search or a direct visit. That is a plausible explanation for the gap, not proof that all four missing sessions followed that path.
Self-reports have their own weakness: they rely on memory and on the options presented in the question. Give people a free-text field, record their answer without coaching them, and avoid treating a remembered source as a complete path. If five leads mention an assistant, start the monthly prompt log described in marketing in AI search and assistants and compare what appears over time.
The CRM source picklist is a practical classification for the scorecard, not a verdict that one touch caused the sale. Keep the raw self-reported answer and captured UTM values on the record so someone can inspect the classification later.
Stage 6: Test a Meta pause with a geographic holdout
Brightwell wants to know whether Meta is producing leads that would otherwise disappear. It pauses Meta in South for four weeks and leaves it running in North, then counts leads from all sources in each region.
Hypothetical table: four-week geographic holdout
| Region | Prior four weeks | Test four weeks | Change |
|---|---|---|---|
| North, Meta on | 34 | 35 | +1 |
| South, Meta off | 30 | 28 | -2 |
The difference-in-differences calculation is South's change minus North's change: -2 - (+1) = -3 leads. Under the test's assumptions, the observed result is three fewer South leads after the pause, adjusted for the one-lead increase in North. Counting all leads matters because a person may see an ad and later arrive through another channel.
The arithmetic is much noisier than the -3 suggests. Using a Poisson approximation, the standard error is roughly sqrt(34 + 35 + 30 + 28), or sqrt(127), about 11.3 leads. Multiplying 11.3 by 1.96 gives about 22.1, making a rough 95% band around the estimate approximately minus 25 to plus 19 leads.
Meta claimed 41 results across both regions in the earlier four-week report. If those claims were evenly split, which is an assumption, that would be about 20 per region. The observed -3 is about 1.5 standard errors away from -20: it leans against an effect of that size, but the test does not rule it out or prove that Meta caused no leads.
The approximation treats the counts as independent Poisson observations. Real lead counts can depart from that assumption, and the wide band is already a warning against a confident result. The test is mainly capable of detecting a large effect at this volume.
Brightwell also chose only two regions. In the Google geo experiments paper by Vaver and Koehler, non-overlapping regions are randomly assigned to treatment or control; a hand-picked pair provides much weaker protection against regional differences. Seasonality can affect North and South differently, and people in South may still see ads targeted to North.
A long pause can change the campaign as well as the measurement. Meta counts a pause of seven days or longer as a significant edit, which sends the ad set back into learning. Record the pause dates, campaign settings and other major changes so a later comparison has context.
Stage 7: Make budget decisions and keep the review small
Brightwell treats the holdout as weak evidence alongside the CRM scorecard. Its initial decisions keep the four-week budget at $5,800.
Hypothetical table: next four-week budget and decision
| Channel or action | Previous budget | New budget | Reason to review |
|---|---|---|---|
| Google Ads | $3,000 | $3,000 | Four customers, $750 CAC and about 2.1 months payback |
| Meta Ads | $1,200 | $400 | Two qualified leads from 10; retain retargeting and recheck after 12 weeks |
| SEO content | $1,500 | $1,800 | Eight qualified leads and three customers in the scorecard |
| Email tool | $100 | $100 | Three qualified leads from four; review clicks and replies |
| Referral rewards | $0 | $500 | Seven referral leads produced six qualified leads and three customers |
| Total | $5,800 | $5,800 | The $800 removed from Meta funds $500 for rewards and $300 for content |
The referral budget assumes a $250 reward per signed referred contract and about two rewards per period. Brightwell should record the actual rewards paid against referred customers as they occur. The past referral CAC was based on $0 recorded spend and will change when rewards are paid.
The Meta decision does not rest on the holdout alone. The scorecard shows $600 per qualified lead and zero customers in this small period; the holdout's estimated difference is minus three leads with a wide uncertainty band. Brightwell cuts prospecting, retains $400 for retargeting and checks the rolling 12-week results before drawing a firmer conclusion.
Google Ads stays at $3,000 because its four customers imply a $750 CAC and about 2.1 months of gross-margin payback under the stated assumptions. Three SEO customers and three referral customers support testing more investment, though those counts are also small. The five AI-assistant self-reports justify a monthly prompt log, not a claim that five leads arrived through tracked assistant links.
Keep the weekly meeting to 20 minutes: confirm that new leads entered the CRM, inspect Unknown sources, check qualification and sales updates, then read spend and customer cost together. This week, export the last four weeks of new CRM leads and count distinct form submissions and phone inquiries. Put that count beside the platform claims before changing a budget.
Put your next steps in order
Build on what you have just read with a free ebook on semantic SEO and AI visibility. Get worked examples, decision rules, checklists and a do-this-week action in every chapter, plus a six-week plan for your site.

