If you run real ad spend, you have already lived this. Meta says your prospecting campaign returned 4.2x. Google claims 3.1x. Shopify crowns branded search. Your post-purchase survey says "TikTok," where you run no ads at all.
None of them is lying. They are measuring different things. Understanding why they diverge is the whole game, and it is the part most attribution articles skip. This guide walks the real challenges with operating numbers, then gives you a way to budget despite them. For the full picture, start with our attribution and tracking guide.
Challenge 1: every platform grades its own homework
The single biggest attribution challenge is structural. Your tools are not configured wrong — they are built to answer different questions about the same order.
Shopify's default is last non-direct click over a 30-day window: when a customer converts, it walks back through their sessions and credits the most recent non-direct source (WeltPixel, Order Legend). GA4 now uses data-driven attribution by default, after it removed first-click, linear, time-decay and position-based models in November 2023 (Google Analytics Help). Meta and Google Ads, meanwhile, each report their own self-credited conversions on their own windows.
Here is why the rows can never reconcile. Say you sell 340 orders a month at a $31 average order value — that is $10,540 in revenue. You spend $2,800 on Meta and $1,200 on Google. If Meta claims 4.2x and Google claims 3.1x, they are jointly taking credit for $11,760 plus $3,720, or $15,480 in sales. That is more revenue than your store actually made. Both platforms counted buyers the other one also touched.
It gets worse: Shopify does not ingest ad spend on any plan, so you cannot even compute ROAS inside native Shopify reports (Polar Analytics). The numbers live in separate systems that were never designed to agree.
Challenge 2: privacy quietly broke click tracking
The second challenge is that the click-based tracking all these tools depend on has been degrading for years, in waves most operators feel as "tracking got worse again" without knowing which layer broke.
Apple's App Tracking Transparency landed in 2021, and roughly 96% of US iPhone users opted out at launch per Flurry; the industry-wide opt-in average had only climbed to around 35% by Q2 2025 per Adjust. That is what gutted the Meta pixel's view of iOS buyers.
Then iOS 17 added Link Tracking Protection in 2023, which strips click IDs like fbclid and gclid from links opened in Apple Mail, Messages and Safari Private Browsing — though your UTM tags survive (Klaviyo). Safari's Intelligent Tracking Prevention also caps most client-side cookies at seven days (Stape), so a shopper who clicks an ad and buys on day nine looks like a brand-new "direct" visitor.
This is not an edge case. Safari accounts for roughly 55% of US mobile sessions (TechnologyChecker). For a US store, about half your mobile traffic lives in the browser that fragments tracking the hardest.
Challenge 3: every model is a biased lens
Operators go model shopping hoping one is "correct." None is. Each attribution model is a lens with a known bias, and the bias is the thing that drives bad budget calls.
Last-click — the most common default — gives all the credit to the final touch before purchase. That structurally over-credits branded search and retargeting, which close demand that other channels created. Someone had to make the customer search your brand name in the first place.
Data-driven attribution feels more objective because it is machine learning, but it is trained only on observable click paths, so it inherits every blind spot of click tracking. It also needs volume: a commonly cited practitioner threshold is around 300 or more conversions per month before its fractional splits stabilize (ReferralCandy). Below that, you are reading noise. Switching models never creates revenue — it only reshuffles credit. If a channel suddenly "improves" because you changed the model, nothing real happened.
Challenge 4: your best channels may emit no click at all
Here is the challenge last-click can never see: zero-click influence. Podcasts, word of mouth, organic TikTok and Instagram mentions, and influencer shoutouts watched on another device drive real buyers but emit no click for any model to credit. Those sales get filed under Direct or Organic Search.
A one-question post-purchase survey — "How did you hear about us?" on the order-confirmation page — asks the buyer instead of inferring from clicks. Single-question confirmation-page surveys consistently see 40–60% response rates, with KnoCommerce reporting a platform-wide average near 45% (KnoCommerce). When a channel shows up heavily in survey responses but barely in your pixel-credited orders, that is the fingerprint of a zero-click channel your tracking is starving. Surveys are directional, not an order-level replacement — buyers misremember — but they see influence your tools structurally cannot.
Challenge 5: attribution can't tell you what your ads caused
The deepest challenge is conceptual. Attribution assigns credit for conversions that happened. It cannot see the counterfactual — which of those buyers would have purchased anyway.
Retargeting is the classic trap: you are showing ads to people already on your site, whose base purchase rate is high with or without the ad, which is why platform-reported retargeting ROAS is typically the most inflated relative to its true causal contribution (Haus). Only a holdout test measures causation. Meta's Conversion Lift splits the audience into a test cell that sees ads and a control cell that does not, requiring at least a seven-day window and 10% of the audience per cell (Haus).
The output is an incrementality factor: incremental conversions divided by platform-reported ones. If Meta reports 500 conversions and the lift test finds 300 were truly caused, the factor is 0.6 — so a campaign Meta calls 4.0x is really closer to 2.4x. A channel can win every attribution model and still be non-incremental.
The challenge every other article skips: profit, not credit
Every point above is about which channel gets credit. None of it tells you whether an order made money. That is the gap.
Say your $31 shirt costs $12 from Printify, with about $1.50 in Shopify and payment fees. That is $17.50 of gross profit per order before ads. At 340 orders a month, your Meta spend of $2,800 works out to roughly $8.24 of ad cost per order — leaving about $9.26 in actual per-order profit. Now the attribution fight matters in dollars: if last-click is over-crediting a retargeting campaign whose real incrementality is low, you are not just misreading a chart, you are paying $8 per order to harvest sales you would have gotten for free.
This is where connecting your numbers pays off. PodVector AI's AI employee, Victor, connects to Shopify, Meta Ads, Google Ads, Printify, Printful, Gelato and Klaviyo, so he sees every order, every ad dollar and every supplier cost together and computes the true per-order profit on each product. Victor pauses the losing ad set, shifts budget toward the campaigns that actually clear profit, builds the Klaviyo flow that recaptures the carts your tracking lost, and delivers the reconciled report to your Google Drive. He is not a dashboard you have to read — he does the work on the numbers he can see across all six platforms.
A playbook for budgeting when the numbers disagree
You will never make these tools match. You can make them useful. This five-step shape beats model shopping.
Fix the self-inflicted layer first. Most "mystery Direct" and "Unassigned" traffic is a tagging problem — lowercase, consistent UTMs, no untagged paid links, never UTM-tag internal links. Our UTM tracking guide and the Klaviyo UTM setup cover the discipline that survives Apple's privacy changes.
Give each tool one honest job. Shopify tells you how many orders and how much revenue. Platform dashboards are only for comparing campaigns inside the same platform. GA4 shows cross-channel journey shape. The survey shows influence clicks miss.
Triangulate. Where pixel, GA4 and survey rank a channel the same way, trust it. Where they diverge hard, suspect a zero-click channel last-click is starving.
Anchor to blended reality. Total ad spend divided by total new orders (blended CAC) and total revenue divided by total ad spend (MER) come straight from Shopify plus your ad invoices, and no attribution model can game them. If platform numbers improve while blended numbers stall, credit is being reshuffled, not created. Our media mix modeling overview and attribution tools comparison go deeper on blended measurement.
Reserve causal tests for the big claims. When being wrong about one channel would materially change your budget, run a lift test or geo holdout on that channel — not everything, and not continuously.
FAQs
Why don't Meta, Google, and Shopify ever show the same number?
Because they answer different questions about the same order. Shopify credits the last non-direct click over a 30-day window (WeltPixel), GA4 uses a data-driven model (Google Analytics Help), and Meta and Google each self-credit on their own windows. There is no setting that makes them agree, because they are not measuring the same thing.
Why is so much of my traffic showing as "Direct"?
Rising Direct is usually tracking decay, not brand love. Stripped click IDs, expired Safari cookies, and untagged links all dump sales into Direct. Since Safari caps most client-side cookies at seven days (Stape) and drives about 55% of US mobile sessions (TechnologyChecker), considered-purchase journeys fragment into Direct constantly.
Did iOS 17 kill UTM tracking?
No — it is backwards. Link Tracking Protection strips click IDs like fbclid and gclid in Apple contexts, but UTM parameters survive because they identify campaigns, not individuals (Klaviyo). That makes disciplined manual UTM tagging more valuable than before, not obsolete.
Which attribution model should a POD store use?
There is no single correct model — each is a biased lens. For short-cycle impulse products where the shopper sees the ad and buys in one session, first-touch and last-touch are often the same event and model choice barely matters. The pro move is to triangulate across tools and anchor to blended CAC and MER rather than trusting any one model.
Is my platform-reported ROAS real?
Partly. Platforms credit themselves for conversions they merely touched, including buyers already headed to purchase, so retargeting ROAS is typically the most inflated (Haus). To find the real number, run a holdout test and multiply the reported ROAS by the measured incrementality factor.
Do I need an attribution tool, or can I DIY this?
The hygiene steps — clean UTMs, blended CAC and MER from Shopify plus invoices — are free and mechanical, and you should do them regardless. The hard part is seeing orders, ad spend and supplier costs in one place to compute real per-order profit, which is exactly what connecting your accounts to a tool like Victor solves without you stitching spreadsheets together.