Lab · Analytics & Attribution

Analytics and attribution
in D2C is broken by design.

The same platforms spending your money are also grading their own performance. This page is about what measurement looks like when the business owns the customer journey instead of renting truth from ad dashboards.

Act 01 · A Founder's Story

The same platforms spending your money
are also grading their own performance.

This is how it usually goes.

High Low ROAS OVER TIME →
01
Dashboard
ROAS 3.8×
Meta performing. Google stable. Both numbers look healthy.
Founder
Things are working. Increase budget next month.
02
Dashboard
ROAS 5.1×
Strong month. Both platforms claiming revenue growth.
Founder
The budget increase worked. Scale further.
03
Dashboard
ROAS 5.4×
Peak performance. Meta and Google both reporting strong.
Founder
Best month yet. The product is working. The ads are working.
04
Dashboard
ROAS 4.2×
Numbers dipping. Agency says algorithm update. Seasonality. Give it time.
Founder
Something changed. The product that was selling has slowed. Nobody can explain why.
05
Dashboard
ROAS 3.1×
Meta claims ₹14L. Google claims ₹9L. Shopify shows ₹11L total.
Founder
Three numbers for the same month. Which one is true? Which platform do I cut? Which do I trust?
06
Dashboard
ROAS 2.3×
Spend is up. Revenue is flat. No clear signal on what to do next.
Founder
I cannot sanction more budget. I do not know which channel is working, who my customer is, or why the business has stalled.
This is not a performance problem.
It is a measurement problem.

And it is the default state of most D2C brands running paid media. The platforms are not lying. They are each reporting their own version of the truth — a version built to maximise their own credited contribution. The real damage is not just inflated ROAS. It is that when performance falls, the founder has no owned map of what broke, where customers dropped off, or which intervention should happen next.

The three sources you actually have
CRM / Shopify Hard anchor — revenue truth
GTM / GA4 Behavioural proxy — best available
Platform dashboards Claims layer — to be stress-tested
Free attribution check

Want to know how much your ROAS is being inflated?

Mawara runs a 40-minute check against your actual platform and Shopify numbers. It is a direct read on where the numbers diverge, where credit is being double-counted, and what decision the business should make next.

01Platform-claimed revenue vs Shopify revenue
02Likely zones of double-counting or cannibalisation
03One recommended tracking or budget decision
04Whether the issue is traffic, attribution, CRO, or retention
Act 02 · Attribution Conflict

Three scenarios.
Three versions of the same data.

What the dashboard reported. What actually happened. What right tracking revealed. All figures hypothetical. The situations are not.

Google PMax Cannibalisation
Email Attribution Inflation
Organic Contamination
Scenario · Google Performance Max
The campaign that looked efficient was taxing demand it didn't create.
Google PMax intercepting branded search customers — people who already knew the brand and would have bought through organic search at zero incremental cost. Same revenue recovered at 34% of prior Google PMax spend once branded terms were excluded.
Platform Reported ROAS
6.8×
Incremental ROAS (overlap removed)
2.4×
Wasted spend per month
₹2.8L
Scenario · Email Attribution
Two channels claimed the same customer. The one that closed got all the credit.
Three Meta ads built the awareness. Two emails arrived. The customer converted after clicking an email. Email claimed 100% last-click. Meta claimed view-through. Sequential attribution reveals Meta built the path — email only closed it. Budget decision reverses.
Email Last-Click Credit
100%
Actual — Meta 65% · Email 35%
Corrected
Misallocation corrected
₹1.4L back to Meta
Scenario · Organic Contamination
The paid channel was claiming customers who were already on their way.
18 months of paid built real brand recognition. Customers arriving via direct navigation and branded search. A geo holdout test — one city withheld from paid spend for four weeks — revealed how much of what was attributed to paid was actually organic demand.
Dashboard Reported CAC
₹480
True Incremental CAC
₹728
Monthly organic misattributed to paid
₹3.2L
Act 03 · Customer Journey Ownership

A measurement architecture that shows where demand is created,
where it leaks, and where it compounds.

Ad platforms can tell you who they think converted. They cannot tell you whether the customer journey is healthy. This architecture gives the business its own journey spine: from first session, to product scrutiny, to cart, to checkout, to purchase, to subscription, to reorder.

When ROAS falls, these are different failures
TOF
Wrong traffic

Cheap clicks, weak source quality, or creative that attracts curiosity without buying intent.

TOF → MOF
Weak product conviction

Visitors arrive, but do not examine ingredients, pricing, proof, subscription, or comparison content.

MOF → BOF
Intent decay

People show buying signals, then drop at cart, shipping, offer clarity, trust, or checkout friction.

BOF → Retention
Bad customer quality

The first order lands, but customers do not subscribe, reorder, or generate enough LTV to justify acquisition.

A platform dashboard shows revenue after the fact. A journey architecture shows where revenue was lost before it happened.
What this lets the founder decide
Which channel deserves more budget.Not which channel wrote the prettiest self-report.
Why ROAS fell.Traffic quality, product proof, checkout friction, or retention decay.
Whether paid is creating demand.Or merely harvesting demand organic already created.
Which cohort is worth acquiring again.First-order revenue is cute. Repeat value pays the bills.
Hypothetical journey spine
One month of paid traffic, broken into decision points
The point is not the exact number. The point is knowing which leak deserves action.
24,000Paid visitors
8,400Product sessions
2,100Proof engaged
740Added to cart
310Checkout started
126Purchased
38Subscribed
21Reordered
Did paid social fail? Did the landing page fail? Did product proof fail? Did checkout fail? Or did the platform find buyers once without building customers?
The tag system behind the journey
Google Tag Manager turns funnel stages into observable customer movement.

The architecture below is not just a naming system for events. Google Tag Manager becomes the control layer that listens for meaningful behaviours, enriches them with context, and sends clean signals to GA4, ad platforms, CRM, Shopify, and eventually the warehouse. Without this layer, TOF, MOF, BOF, and retention remain consulting labels pasted over a blind storefront rather than operational views the business can act on.

01 · Listen
Triggers detect behaviour

Page depth, PDP scrutiny, quiz starts, subscription toggles, cart actions, checkout starts, purchases, cancellations, and reorders become stage-specific signals.

02 · Describe
Variables carry meaning

Source, campaign, goal segment, SKU, order type, subscription frequency, coupon, customer type, and device context explain what the event actually means.

03 · Send
Tags route the signal

GA4 receives behavioural events, Meta and Google receive conversion signals, CRM receives lifecycle context, and BigQuery receives the joined business record.

04 · Govern
Rules prevent data rot

Consistent names, deduplication keys, consent checks, test modes, and release discipline keep the measurement system documented, queryable, and trusted.

GTM is the operating nervous system: it watches how customers move from awareness to intent, from intent to conversion, and from conversion to repeat value.
The honest constraint
This improves measurement. It does not defeat privacy, browsers, or physics.

A serious architecture must admit what it cannot see. iOS privacy changes, browser cookie restrictions, ad blockers, consent banners, cross-device behaviour, and stricter privacy expectations in markets such as the EU can reduce signal quality. Server-side tagging, Consent Mode, CAPI, deduplication, and first-party data help recover part of the view — but they do not restore perfect attribution. Anyone promising perfect visibility is either confused, selling software, or both.

iOS privacyAttribution windows shrink, user-level visibility drops, and platform-reported conversions become more modelled.
Consent regimesIn privacy-sensitive countries, tags must respect consent choices before firing or storing marketing identifiers.
Cookie lossBrowsers and ad blockers interrupt client-side tracking, especially for returning-user and assisted-conversion analysis.
Modelled gapsEven good systems still infer parts of the journey. The goal is less distortion, not omniscience.
Compound Hypothetical · Protein Supplements · D2C India

Below is the instrumentation layer behind the plain-language journey map. The technical pieces are there, but they exist for one purpose: making better business decisions when growth stalls.

Triggers
What fires the event
Variables
What data it carries
Tags / GA4 Events
What gets recorded
TOF
Awareness &
Acquisition
Where did this session come from — and what goal brought them in?
All Pages — Page View
Session Source Classifier
Goal Quiz Start
If this fails: every downstream event loses source and goal context
Session Identity Spine
Session Attribution Taxonomy
Goal Segment
Technical context: session identity, consent state, traffic type, intent, and claim-risk scoring.
session_start
page_view
quiz_start
Decision context: source, medium, paid/organic flag, and goal segment.
MOF
Formulation
& Subscription Intent
Did they scrutinise the formulation — and did they choose to subscribe?
Supplement Facts Panel View
Flavour Selected
Subscribe vs One-Time Toggle
Add to Cart — Observer
Begin Checkout
Facts panel view = scroll-to + 4s dwell on the nutrition/ingredient block
Product & Formulation Context
Subscription Intent Flag
Returning Customer Flag
Technical context: SKU, flavour, subscription intent, frequency, and new-versus-returning status.
facts_panel_view
select_subscription
add_to_cart
begin_checkout
Every event keeps the original goal and traffic context attached.
BOF
Conversion,
Subscription & Retention
One-time or subscription — what is the real LTV — did they reorder before churning?
Purchase — Client-side
Purchase — Server Webhook
Subscription Start
Skip / Pause / Cancel
Replenishment Reorder
Server + client deduped by order_id — closes the iOS / privacy-browser gap
Transaction Identity Spine
Revenue, Order Type & Quality
LTV & Churn Cohort
First-Touch Attribution Bridge
Technical context: order identity, order quality, cycle count, churn risk, reorder timing, and first-touch source.
purchase
subscription_start
subscription_cancel
reorder
return_visit
Purchase records net revenue and order type. Reorder reveals whether acquisition compounded.
Decision it drives
What broken looks like
✕  Close
Mawara POV · Architecture before attribution
There is no universal funnel architecture. There is only the architecture your market deserves.

A supplement brand, a luxury furniture studio, a clinic, a B2B service firm, and a creator-led commerce brand do not have the same buying journey. So they should not inherit the same measurement system. Mawara does not force every client into a prebuilt TOF/MOF/BOF template and then pretend the work is strategic because the labels are tidy. Neat dashboards are how bad thinking gets framed nicely.

01 · Time
Consideration windows change the spine

A ₹799 impulse product may convert in one session. A ₹2L service, clinic plan, course, or luxury purchase may need weeks of proof, reassurance, comparison, and follow-up.

02 · Cohorts
Buying power changes intent

High-income, bargain-sensitive, first-time, returning, gift-buying, and subscription cohorts do not hesitate at the same point. Their drop-offs need different signals.

03 · Geography
Markets distort measurement differently

Payment habits, COD behaviour, privacy expectations, language, delivery friction, and platform usage vary by region. A clean funnel in one geography can lie in another.

04 · Current motion
Existing campaigns shape the build

The architecture must account for what the brand is already doing: paid search, Meta, influencers, email, WhatsApp, retail, organic content, sales calls, and CRM flows.

Mawara builds measurement systems around how customers actually decide, not around how platforms prefer to report.
How the system is designed

The work begins with the client’s actual market knowledge: customer objections, sales-cycle length, price sensitivity, buying triggers, repeat-purchase behaviour, acquisition channels, and the founder’s current doubts about what is really working.

  • Map the real decision journey before naming events.
  • Decide which stages matter for this brand, not every brand.
  • Modify common funnel models when the customer journey demands it.
  • Separate platform attribution from business-owned behavioural truth.
  • Design signals for intervention, not just reporting.
  • Preserve room for privacy limits, missing data, and uncertainty.
One month, three different answers — attribution reconciled against the journey
Every order checked back against the real Shopify record
What the platforms claimed
₹23.8L
Meta ₹14.2L + Google ₹9.6L.
Both took credit for the same sales.
What honest tracking showed
₹18.4L
Every order matched to Shopify.
The platforms over-claimed by ₹5.4L.
What actually hit the bank
₹18.4L
Shopify. The one number
no ad platform can inflate.

Meta claimed ₹14.2L. Google claimed ₹9.6L. Together the dashboards report ₹23.8L — but only ₹18.4L reached the bank. The missing ₹5.4L is the same orders counted twice: each platform credits itself whenever its ad was touched anywhere in the journey, so one sale can show up in full on both. Every ROAS figure on both dashboards is inflated by this overlap — here, by 29%.

Reconciled against Shopify, the true blended ROAS is lower than either dashboard claims — and the inflation is not even. It concentrates on the channels that overlap most, like branded search and retargeting, which usually join a journey another channel already started. That is where wasted budget hides. The architecture above adds the missing layer: not only who purchased, but where customers were created, where they hesitated, where they dropped, and who came back. Meta receives repeat purchases through the Pixel and Conversions API; it just does not give the founder a business-owned journey view inside Ads Manager. That view has to be built outside the platform.

The operating loop
01Measure the journey

Capture behaviour by stage, not only by channel.

02Locate the leak

Separate traffic problems from offer, product-page, checkout, and retention problems.

03Intervene precisely

Change the creative, landing page, proof, bundle, checkout flow, or lifecycle sequence.

04Reallocate budget

Move spend toward journeys that produce customers, not just attributed orders.

Act 04 · Measurement to Media to CRO Loop

Knowing where the money goes is step one.
Deploying it better is step two.

Most founders treat measurement and media as separate conversations — one for the analytics setup, one for the ads account. They are the same conversation. What the attribution architecture reveals directly determines where paid budget should move, what organic needs to build, and which pages deserve a conversion test. The sequence matters.

01
Measure

Build the owned view of what is actually happening.

Attribution reconciled against Shopify. Journey leaks identified by stage. Channel contribution separated from channel credit. The measurement layer produces a map — not a dashboard that flatters the platforms.

GA4 + GTM CAPI / server-side BigQuery reconciliation
02
Deploy smarter

Move budget toward what the data shows is actually working.

Reallocate paid spend away from channels overclaiming credit toward channels producing subscribers and reorders. Build organic content that answers the questions the measurement shows visitors are asking before they buy. Use paid to amplify what organic has already proven. The map is now a brief.

Meta · Google Ads audience segmentation organic × paid interlock
03
Experiment on conversion

Test variants against each other to learn what customers actually need before they commit.

Once the traffic is cleaner, CRO stops being surface-level button-colour changes. Mawara turns each exposed friction point into a testable hypothesis: one version speaks to price anxiety, another to trust, another to product clarity, another to checkout resistance. The winning variant does not just lift conversion. It reveals what customers value, fear, misunderstand, or need to see sooner.

A/B variant testing choice and pain-point signals friction-led learning loop
Mawara operates across all three stages. The paid media work informs the CRO brief. The CRO results feed back into the media brief. Running only one of these in isolation produces a ceiling — usually hit in months two or three, when the easy gains from the first budget reallocation stop compounding on their own.
Act 05 · Conversion Research

Most pages are not converting at their ceiling.
The question is whether you can see why.

At its simplest: you have a page. Some visitors buy, most don't. CRO is the discipline of figuring out why — and then testing a fix. You run two versions of the page at the same time: one with the change, one without. Whichever version gets more of the right visitors to buy, wins. You keep the winner, build the next hypothesis from what you learned, and run the next test. Over six months that loop compounds into meaningful revenue movement — from the same traffic you were already paying for.

What CRO actually is

The typical D2C brand runs ads, watches the ROAS number, and tweaks creative when it drops. That is not CRO. CRO is asking a different question: given the traffic that is already arriving, how much of it should be converting — and what, specifically, is stopping the rest?

Most founders cannot answer that question because they do not have the behavioural data that would let them. They know traffic and revenue. They do not know where in the journey each type of visitor hesitates, what they looked at before they left, or whether the checkout drop is a trust problem or a shipping-cost problem. Those are different problems with different fixes.

Once you can see where each type of visitor hesitates — which step, which cohort, which source — you have a test brief. That is when CRO stops being a gut call and becomes a programme: a specific change, tested against a control, with a clear answer at the end.

Without measurement
Checkout rate dropped. Change the button colour. Ask the agency to refresh the creatives. No one knows which step the drop is happening at.
Measurement shows → CRO tests

Measurement shows
More visitors start checkout than complete it. That gap is 3.1× wider for paid-social first-time visitors than for organic ones — concentrated at the payment step, not earlier in the flow.

Hypothesis
Paid-social traffic arrives with lower brand familiarity. At the payment step, unfamiliarity becomes hesitation. This is a trust problem, not a price or UX problem.

Test
Surface a third-party tested badge and payment security mark at the payment step for paid-social traffic only. Success metric: checkout completion rate for this cohort vs control.

Without measurement
Subscription rate is low. Add a discount to the subscribe toggle. Wait two months. Nothing changed.
Measurement shows → CRO tests

Measurement shows
The subscribe toggle is being selected — the intent is there. But on 68% of organic visitor purchases, order_type at checkout reverts to one-time. The visitor chose subscription, then quietly reversed it before buying.

Hypothesis
The reversal is not about price. It is about not knowing what happens if the subscription doesn't work out. The commitment feels irreversible.

Test
Show the cancellation and skip policy inline at the subscribe toggle — not buried in the FAQ. Success metric: whether subscribe_selected at the toggle holds through to order_type on the purchase event.

Without measurement
Cart abandonment is high. Send a discount code email. Run retargeting. No one knows where in the session the decision to leave was made.
Measurement shows → CRO tests

Measurement shows
Abandonment is highest on sessions where the visitor selected a flavour that was out of stock. These sessions exit almost immediately — they are not reaching checkout and dropping there.

Hypothesis
The out-of-stock moment is a dead end. There is no next option offered, so the visitor leaves rather than substituting.

Test
Surface an alternate flavour recommendation at the moment of out-of-stock selection, before cart entry. Success metric: add_to_cart rate on out-of-stock SKU sessions vs current exit rate.

Simulation · small tests, serious movement

Funnel gains rarely arrive as one miracle redesign.
They arrive as small frictions removed in sequence.

A trust badge at the payment step. A clearer subscription promise. An alternate recommendation when a SKU is out of stock. Each experiment looks almost too small to matter in isolation. Together, they increase the number of qualified visitors who move from one stage of the funnel to the next.

Illustrative baseline
Monthly sessions100,000
Average order value₹2,400
Starting conversion rate1.80%
Starting monthly revenue₹43.2L
Six-month revenue path
Baseline
1.80% CVR held constant
₹2.59Cr
No intervention
same leaks, same ceiling
₹2.59Cr
CRO programme
1.80% → 2.22% CVR
₹2.93Cr
The gain is not magic. It is compounding motion: more visitors trust the payment step, more subscription intent survives to purchase, fewer out-of-stock sessions die before cart. Same traffic. Better journey. Approximate uplift: ₹33L+ over six months.
Illustrative model only. The point is not the exact rupee number. The point is that small controlled tests can move revenue more reliably than another round of vague creative refreshes.
The same data that exposes attribution inflation also shows you that someone who clicked an ad drops at the payment step at 2.4× the rate of someone who found you through search. That gap costs more per month than most founders realise — and you cannot fix it by changing the creative. The visitor made it to checkout. The creative already worked. What failed was the moment they had to commit real money to a brand they had known for eleven seconds.
01
Surface

Find the friction from the data, not the opinion.

Tracking tells you what visitors did: which pages they read, whether they added to cart, where they stopped. From that, you build a ranked map of where your funnel leaks — which step has the highest drop rate, which type of visitor loses momentum earliest, where someone chose subscription and then quietly reversed it before buying. That map is the brief. The opinion of what is wrong does not get a vote.

Signals from the stack
Visitors reading product details but not reaching checkout — by who sent themfacts_panel_view → begin_checkout drop by traffic_type
Visitors choosing subscription, then switching to one-time before buyingsubscribe_selected → purchase order_type mismatch
Sessions ending immediately when a product variant is out of stockout-of-stock abandonment by sku
The exact checkout step where paid-traffic visitors drop mostcheckout-step exit rate by traffic_type
02
Segment

Different cohorts have entirely different barriers.

Someone who clicked a paid social ad and landed on your product page for the first time is not the same customer as a subscriber who has already bought twice and is coming back directly. They have different reasons to hesitate, different information gaps, different doubts about committing to a subscription. Running one experiment on both groups wastes the test — the result ends up being true for neither.

Segmentation variables
What the visitor came to buy — fitness goal, use case, intentgoal_segment · traffic_type · intent_type
First-time buyer or returning customercustomer_type
Whether they chose subscription before reaching checkoutsubscribe_selected at cart entry
How long since their last order — for retention targetingdays_since_last_order
03
Compound

Tests that learn from each other, not tests that reset.

Each experiment is designed to answer a specific question about a specific group — and its result feeds the next hypothesis instead of being archived as an isolated win or loss. If showing ingredient sourcing on the product page lifts subscription intent among first-time buyers, the next test is whether changing the default subscription frequency for that same group reduces early cancellations. Each test builds on the last. The programme accumulates knowledge, not just results.

Compounding metrics
Subscription rate per buyer group, by where they came fromsub_rate by cohort and traffic_type
Whether buyers from each test variant are still ordering 90 days later90-day reorder rate vs control
Whether reading the product details correlates with subscribing downstreamfacts_panel_dwell vs downstream sub_rate
Which checkout step loses each type of visitorcheckout-step exit by intent_type
Compound — four buyer cohorts · each with a distinct barrier
Cohort 01 · New · Paid
First-time prospecting buyer
Arrives from paid social. Has never bought a supplement from this brand. High intent, low trust.
Primary barrier
Formulation credibility — ingredients without context. Decides before the facts panel.
Cohort 02 · New · Organic
Research-led first visit
Arrives from organic search or direct. Has read something. Already past the awareness phase.
Primary barrier
Subscription commitment — willing to buy once, hesitant to lock in a cadence before tasting.
Cohort 03 · Returning · Active
Subscriber approaching renewal
On cycle 2 or 3. Satisfied enough to reorder but not yet habitual. Skip risk rising.
Primary barrier
Monotony — same flavour, same frequency, no signal that the brand is paying attention.
Cohort 04 · Returning · At-risk
Lapsed one-time buyer
Bought once, did not subscribe, has not returned in 60+ days. Price-sensitive or unimpressed.
Primary barrier
No reason to return — no personalisation, no acknowledgement of the first purchase.
Cohort detail
Close ×
Active experiment queue — Compound · hypothetical
ID
Hypothesis
Cohort
Status
Lift
CRO-04
Showing a third-party tested badge and FSSAI certification number next to the subscribe option increases the rate at which first-time paid-traffic visitors choose to subscribe.
Cohort 01
Concluded
+18% sub rate
CRO-05
Replacing the default 30-day subscription frequency with a 45-day default for first-time subscribers reduces skip rate at cycle 2 without reducing subscribe_selected rate at checkout.
Cohort 01 → 03
Running
CRO-06
A single-flavour trial pack offer shown only to organic first-time visitors — before the full product page — increases how many research-led visitors start the checkout.
Cohort 02
Running
CRO-07
A personalised flavour recommendation on the reorder confirmation email — based on the buyer's stated fitness goal — increases return visit rate within 30 days for lapsed one-time buyers.
Cohort 04
Queued
CRO-08
A mid-cycle check-in notification at day 22 of a 30-day subscription — acknowledging usage and asking one question about the experience — reduces cancel rate at cycle 2–3 boundary.
Cohort 03
Queued
"
Epilogue · What measurement can and cannot do

Every analytics architecture, however precisely built, is an approximation of reality. Attribution models distribute credit across touchpoints according to rules — rules that are useful but not true. The last click did not cause the purchase. The first click did not either. The customer's decision was the result of a sequence of experiences, some of which your tracking captured and many of which it did not.

Incrementality testing — holdout experiments, geo-based lift studies, matched market tests — brings you closer to the truth than attribution alone. But even the best-designed incrementality test measures the average effect across a population, not the specific contribution to a specific customer's decision. No level of advanced statistics or tracking can surgically isolate incrementality.

There is also the privacy layer. iOS privacy controls, cookie restrictions, consent rules, ad blockers, and cross-device behaviour create blind spots that no tag system can fully remove. A good setup makes those blind spots smaller, documented, and less dangerous. It does not pretend they disappeared because a dashboard rendered a decimal point.

What a well-built measurement infrastructure gives you is not certainty. It gives you a version of reality that is less distorted than the version your platforms produce on their own. It gives you decisions that are closer to right than decisions made from platform dashboards alone. That is not nothing. In a market where most brands are making budget decisions based on the most optimistic possible interpretation of their data, being less wrong is a significant advantage.

Mawara builds measurement systems that are honest about what they measure and honest about what they don't. We will never tell you a number means more than it does.

If any of this sounds like the conversation you've been trying to have with your current analytics setup, it probably is.

The 40-minute attribution check
Give us 40 minutes, on the house. We will compare your platform-reported revenue against Shopify, trace the major overlaps, and tell you what the numbers are probably hiding. No proposal. No pitch. Just an honest read.
You leave with
  • Your platform-claimed revenue versus actual Shopify revenue.
  • The likely zones of double-counting, cannibalisation, or retargeting inflation.
  • One recommended tracking or budget decision.
  • A clearer read on whether the problem is traffic, attribution, CRO, or retention.
Start the attribution check →