Short answer: A marketing attribution model is a rule for splitting credit for a lead or sale across the touchpoints that came before it. First-touch credits discovery, last-touch credits the final click, and data-driven attribution uses machine learning on your account's paths. Since November 2023, GA4 offers only data-driven and two last-click models, and Google Ads offers data-driven and last click. For Indian lead-generation businesses, the model matters less than connecting every lead to a CRM record and a closed-deal value.
Need this done for you? GrowthSparx offers a fixed-scope lead tracking and attribution setup: GA4, Google Ads, Meta, WhatsApp, calls and CRM wired into one dashboard, usually live in 2–3 weeks.
What is a marketing attribution model?
An attribution model is a credit-assignment rule. When a lead arrives after seeing a Meta ad, clicking a Google search ad and chatting on WhatsApp, the model decides how much of that lead (and its revenue) each channel "earned". It does not prove that any channel caused the sale. That needs a controlled test, such as a geo holdout or a campaign pause in matched cities.
"Performance attribution models" and "marketing attribution models" mean the same thing in a paid-media context. The useful question is not "which model is correct?" but "which model answers the decision I am making this week?" Budget between prospecting and retargeting is one decision. Cutting a keyword is another. They need different lenses.
Two definitions to fix before you compare any report:
- Conversion event. Is the thing being credited a form fill, a qualified lead, a site visit or a paid invoice? Crediting form fills tells you which channel produces enquiries, not which produces revenue.
- Lookback window. How far back a touchpoint can be and still receive credit. In GA4 the default is 30 days for acquisition events and 90 days for other key events, per Google Analytics attribution settings help.
How do attribution models split credit? One worked example
Example (illustrative, assumptions stated): a Noida B2B services firm sells a ₹2,00,000 annual contract. One buyer's journey looks like this:
- Day 1: clicks a Meta (Instagram) ad, reads a case study, leaves.
- Day 6: searches the service name on Google, clicks a Search ad, submits the enquiry form. The CRM creates the lead.
- Day 15: clicks a UTM-tagged link in a WhatsApp follow-up from the sales team, revisits the pricing page and books a call.
- Day 20: signs the contract. Sales marks the deal "Won" in the CRM.
Here is how each model credits the ₹2,00,000. Time decay uses a 7-day half-life measured back from Day 20; position-based uses the classic 40/20/40 split.
| Model | Meta ad (Day 1) | Google Search ad (Day 6) | WhatsApp link (Day 15) | What it rewards |
|---|---|---|---|---|
| First touch | 100% (₹2,00,000) | 0 | 0 | Discovery |
| Last touch / last non-direct click | 0 | 0 | 100% (₹2,00,000) | The final nudge |
| Google paid channels last click (GA4) | 0 | 100% (₹2,00,000) | 0 | Google Ads only; ignores non-Google channels |
| Linear | 33.3% (₹66,667) | 33.3% (₹66,667) | 33.3% (₹66,667) | Participation, equally |
| Time decay (7-day half-life) | 15% (₹30,100) | 25% (₹49,400) | 60% (₹1,20,500) | Recency |
| Position-based (40/20/40) | 40% (₹80,000) | 20% (₹40,000) | 40% (₹80,000) | Opening and closing |
| CRM "lead source" field | 0 | 100% (₹2,00,000) | 0 | The touch that created the lead |
| Data-driven | Modelled from your account's converting and non-converting paths; cannot be calculated by hand | Observed contribution patterns | ||
Time-decay maths: weight = 2^(−days before conversion ÷ 7). Meta 19 days → 0.152, Google 14 days → 0.250, WhatsApp 5 days → 0.610; each weight divided by the total (1.012). Rounded to the nearest ₹100.
Three things this example shows:
- Every model is "right" by its own rule. Meta looks worthless under last touch and essential under first touch. If you judge Meta prospecting on last-click GA4 reports, you will under-fund it.
- The WhatsApp touch is a sales activity, not a marketing channel. A last-click view hands the whole deal to a follow-up message your sales team sent. We usually tag sales-sent links with
utm_medium=sales_followupand exclude them from channel-budget decisions. - Each ad platform claims the deal for itself. If the GCLID from Day 6 is imported, Google Ads reports a ₹2,00,000 conversion. Meta may also count it if Day 1 falls within its attribution setting. Added together, platform dashboards can show more revenue than you actually banked. That is why the CRM, not any ad platform, is the revenue source of truth.
For a deeper comparison of the two single-touch models, see our guide to first-touch vs last-touch attribution.
Which attribution models does GA4 offer now?
Google removed four rule-based models in 2023. Google's GA4 attribution help page states that "the first click, linear, time decay, and position-based attribution models are no longer available as of November 2023." GA4 now offers three reporting attribution models:
| GA4 model | How it assigns credit | When we use it |
|---|---|---|
| Data-driven (default) | Machine learning on converting and non-converting paths. Factors include time from the key event, device type, number of ad interactions, order of exposure and creative type. Each model is specific to the advertiser and key event. | Default reporting and cross-channel budget discussions, once the property has steady conversion volume. |
| Paid and organic last click | 100% to the last channel clicked, ignoring direct traffic. | Sanity-checking data-driven results; comparing with older reports. |
| Google paid channels last click | 100% to the last Google Ads click; non-Google channels are ignored. | Reconciling GA4 with the Google Ads interface. |
Practical notes for GA4:
- Data-driven credit can move after the fact. Google says conversions can be re-attributed for up to 7 days after they happen, so don't lock weekly reports on Monday morning.
- First-touch still exists, as a dimension. "First user source / medium" in acquisition reports shows where a user first came from. It is a user-scope dimension, not an attribution model, but it answers the discovery question.
- Linear and position-based views now live outside GA4. If you need them, compute them in BigQuery from the GA4 export or in your CRM from stored touchpoints.
- Lookback windows matter for long cycles. A real-estate buyer or a study-abroad applicant often takes longer than 30 days to decide. For lead events, keep the 90-day key-event window (options: 30, 60 or 90 days).
Setting up the events themselves is covered in our GA4 lead generation tracking guide.
How does attribution work in Google Ads?
Google Ads attributes conversions only across Google ad interactions, and it now supports two models: last click and data-driven. Google's sunset notice set the timeline: the change was announced on 6 April 2023, new selection of the old models stopped in June 2023, and conversion actions still on them were switched to data-driven in September 2023.
Three implications for lead-generation accounts:
- Smart Bidding learns from whatever the conversion action counts. If the primary conversion is "form submit", tCPA will find people who fill forms, including spam and students looking for a job. Moving the primary action to a qualified-lead or closed-deal event from the CRM changes what the algorithm chases.
- Data-driven in Google Ads only redistributes credit among Google clicks. It cannot tell you that Meta started the journey. That cross-channel question belongs in GA4 or the CRM.
- Imported offline conversions are attributed to the click. When you upload a won deal with its GCLID, Google Ads credits it to the click's date and campaign, not the deal date. Your "conversions by day" chart will move backwards in time as imports arrive.
Which attribution model should you choose?
Choose by the decision, not by preference. This is the selection table we use with clients:
| Decision you are making | Lens to use | Where to get it |
|---|---|---|
| Which channels bring new prospects into the funnel? | First touch | CRM "original source" field; GA4 first user source |
| Which campaigns produce enquiries efficiently? | Last non-direct click on the lead event | GA4 paid and organic last click; CRM lead source |
| How should Google Ads bid? | Data-driven on a qualified or won conversion | Google Ads with offline imports |
| How should budget split between Meta and Google? | Data-driven plus a first-touch cross-check | GA4 and CRM cohort report |
| Does a channel actually cause extra sales? | Experiment, not a model | Geo or time-based holdout test |
When an event records only a handful of conversions a month, any model's output swings with each new deal. In that situation, report qualified leads and revenue by original source and latest source, and make decisions on quarter-long cohorts rather than weekly swings.
How do you track attribution through the CRM?
Analytics tools see sessions; the CRM sees people and money. Attribution only becomes reliable when the two are joined on a stable lead ID. The sequence we implement:
- Govern UTMs. Fixed values for source, medium and campaign; lower case; no free-text.
- Capture identifiers on the landing page. Store GCLID (or GBRAID/WBRAID),
fbclid, UTMs, landing page and referrer in first-party storage, then write them to hidden form fields on submit. - Write two source fields in the CRM. "Original source" is set once and never overwritten. "Latest source" updates on each new meaningful touch.
- Log non-form touches. Calls (via tracking numbers) and WhatsApp chats need their own source capture; our guides on call tracking and WhatsApp lead tracking cover both.
- Timestamp every stage. New → contacted → qualified → site visit or demo → won/lost, each with date, owner and loss reason.
- Send outcomes back. Import qualified and won stages to Google Ads (Google now recommends enhanced conversions for leads via Data Manager) and to Meta through the Conversions API.
- Reconcile monthly. Compare CRM revenue by original source with what each platform claims. The gap is your over-attribution.
The full field mapping and report design is in our CRM revenue attribution setup guide, and the Google Ads import steps are in how to track qualified leads from Google Ads to CRM.
Why this matters in practice: on lead-generation accounts we judge campaigns on downstream outcomes, not form volume. For ILeadGuru's real-estate buyer leads, managed in-house by our founder before GrowthSparx, the number reported against is a 30% lead-to-deal conversion (case study). A figure like that only exists if the CRM records the outcome against the original lead.
What fields should every lead record carry?
| Layer | Fields | Owner |
|---|---|---|
| Acquisition | original_source, original_medium, original_campaign, latest_source, landing_page, gclid / gbraid / wbraid, fbclid, first_touch_at | Marketing |
| Enquiry | lead_id, created_at (with IST offset), channel (form/call/WhatsApp/portal), consent flags | Marketing ops |
| Qualification | status, qualified_at, qualification reason, rejection reason, owner, first_contact_at | Sales |
| Revenue | deal_value (₹), won_at, product/project, invoice or booking reference | Sales / finance |
Collect only what has an operational use. Phone and email are needed for matching; date of birth or income usually is not.
Portal leads (99acres, MagicBricks, Housing.com, IndiaMART, Justdial) arrive without any click ID. Give them a fixed original_source value through the portal's CRM integration or email parser, so they appear in the same funnel report instead of as "unknown".
What are the most common attribution mistakes?
| Mistake | Symptom | Fix |
|---|---|---|
| Optimising Google Ads to raw form fills | CPL falls, sales complain about junk | Make a qualified-lead import the primary conversion; keep form submit as secondary |
| Overwriting original source | Every lead looks like "direct" or "WhatsApp" | Write-once original source field; separate latest source |
| Summing platform-reported revenue | Platforms claim more revenue than finance booked | Use CRM revenue as the total; treat platform numbers as each platform's view |
| Untagged sales WhatsApp and email links | Inflated "referral" or "direct" in GA4 | Tag with a sales medium and exclude from channel ROI |
| Comparing GA4 to Google Ads and expecting a match | Endless reconciliation threads | Document model, window and conversion-date differences; compare trends |
| Duplicate conversions from page reloads or repeat uploads | Conversion rate jumps without a sales increase | Use one lead ID per conversion and flag rows once uploaded |
No setup produces perfect person-level attribution. Consent choices, cross-device journeys and offline word-of-mouth leave gaps. Write those gaps down next to the dashboard so nobody reads a modelled number as a fact.
Sources
- Google Analytics Help: Get started with attribution: current GA4 models; removal of first click, linear, time decay and position-based as of November 2023; data-driven factors and 7-day re-attribution.
- Google Analytics Help: Attribution settings: lookback window defaults (30 days acquisition, 90 days other key events) and options.
- Google Ads Help: About attribution models: Google Ads supports last click and data-driven; old models upgraded to data-driven.
- Google Ads Help: First click, linear, time decay and position-based models are going away: April, June and September 2023 timeline.
- Google Ads Help: About offline conversion imports: enhanced conversions for leads and Data Manager as the recommended route.