If you run an operating store, you already live the problem attribution is supposed to solve. Meta claims a 4.2x return, Google claims 3.1x, your Shopify report crowns branded search, and a customer tells you they found you on a podcast where you spend nothing. Attribution is the discipline that decides who gets credit for each order. The useful question is not "which model is right" — it's "which questions can this actually answer, and which ones is it quietly guessing at."
The pages currently ranking for this keyword hand you a long bulleted list of question categories and stop there. This one goes further: real question-by-question, with worked numbers, the privacy caveat they skip, and the profit angle they never reach. For the full foundation, see our attribution and tracking guide.
Question 1: Which channels introduce new customers vs. which ones close the sale?
This is the question attribution answers best, because it maps directly onto the two oldest models. First-touch attribution gives 100% of the credit to the first interaction a buyer had with your brand, so it surfaces discovery channels — prospecting ads, organic social, content (Shopify). Last-touch gives all the credit to the final click before purchase, so it flatters closing channels like branded search, retargeting, and email (Kissmetrics).
Say your TikTok prospecting ads introduce a shopper who, nine days later, Googles your brand name and buys. Last-touch hands the sale to branded search; first-touch hands it to TikTok. Neither is lying — they are answering two different questions. The trap is budgeting off last-touch alone: it structurally over-credits the channel that harvests demand other channels created, which is why last-click has produced more bad budget calls than any other model (ProperExpression).
Question 2: How many touchpoints does it take to convert?
Attribution logs the path, so it can tell you journey length — and journey length is what decides whether model choice even matters. A commonly cited practitioner average is eight to ten touchpoints before a purchase (a heuristic, not a law — ReferralCandy). For a store like that, first-touch and last-touch will name completely different heroes.
But for a short-cycle impulse product — see the ad, click, buy in one session — the first touch and the last touch are frequently the same event, and swapping models barely moves the numbers. Before you obsess over which model to pick, let attribution answer the cheaper question first: how long are my journeys? If they're one session, stop model-shopping.
Question 3: Which marketing spend is wasting my budget?
This is the question that pays for the whole exercise, and it's where the profit angle the ranking pages skip actually lives. Attribution ranks channels by credited orders; your job is to layer cost and margin on top, because a channel with great attributed ROAS can still lose money per order.
Walk a real example. Say your store does 340 orders a month at a $31 average order value, with $2,800/month in Meta spend. Your product costs $11 from Printify, and Shopify plus payment fees run about $1.80 on a $31 order. That leaves roughly $18.20 of gross margin per order before ads. If Meta's pixel credits itself with 95 of those orders, its attributed revenue is 95 × $31 = $2,945 — a 1.05x ROAS that looks like break-even. But on margin, those 95 orders throw off 95 × $18.20 = $1,729 in gross profit against $2,800 in spend. That channel is losing about $1,071 a month even though the revenue number looked close to even.
Attribution surfaced the candidate; margin math convicted it. Shopify can't do the second half for you — it does not ingest ad spend, so ROAS cannot be computed inside native Shopify reports on any plan (Polar Analytics). You have to join spend to margin yourself, or use a tool that already does.
This is the exact gap PodVector AI fills. Victor sees every order, every ad dollar, and every supplier cost together, so he computes the true per-order profit on each product and campaign — and when a Meta ad set is burning margin like the one above, Victor pauses it inside your ad account and shifts the budget. He does the same across Google Ads, and he knows your Printify, Printful, and Gelato product costs before the first sale, so the margin math is never a guess.
Question 4: Which touchpoint deserves the credit — and how do the models split it?
Once journeys get long, "who gets credit" stops being yes/no and becomes a fraction. The multi-touch models each apply a different rule:
- Linear splits credit equally across every touch — fair, simple, and rarely actionable.
- Position-based (U-shaped) typically gives about 40% to the first touch, 40% to the last, and spreads 20% across the middle (ReferralCandy).
- Time-decay weights recent touches heaviest, which systematically talks you out of upper-funnel spend.
- Data-driven attribution (DDA), now GA4's default, uses machine learning to split credit based on your own account's winning conversion paths. It needs volume — a commonly cited threshold is roughly 300+ conversions a month before its fractional splits stop wobbling with noise (ReferralCandy).
Know your defaults, because they quietly answer the question for you. Shopify's default is last non-direct click over a 30-day lookback window (WeltPixel). GA4 removed first-click, linear, time-decay, and position-based models in November 2023 — you literally cannot pick them anymore (Google Analytics Help). So Shopify, GA4, and Meta are each answering the credit question with a different rule, and their channel rows will never match. Our guide to marketing attribution tracking walks through reconciling them.
Question 5: Which channels am I under-counting because they never emit a click?
Here's a question attribution can't answer from click data alone — and pretending otherwise is the most expensive mistake on this list. Podcasts, word of mouth, organic TikTok, an influencer shoutout watched on a different device: none of these fire a click, so last-click files those buyers under "Direct" or "Organic Search."
The fix is to ask the buyer. A one-question "How did you hear about us?" survey on the order-confirmation page consistently pulls 40–60% response rates per the major survey vendors, with KnoCommerce reporting a platform-wide average near 45% (KnoCommerce). When a channel shows up at 30% of survey responses but 10% of pixel-credited orders, that's your flag that click tracking is starving it. Self-report is noisy — people say "Instagram" when they mean an Instagram ad — so treat it as directional evidence of influence, not an order-level replacement.
Privacy changes made this blind spot worse. Since iOS 17's Link Tracking Protection (September 2023), Apple strips fbclid and gclid click IDs from links opened in Mail, Messages, and Safari Private Browsing — though UTM parameters survive (Klaviyo). And Safari's tracking prevention caps JavaScript-set cookies at seven days (Stape), so a Safari shopper who buys on day nine looks brand-new. With Safari running roughly 55% of US mobile sessions (Backlinko), that's not an edge case — it's half your mobile traffic quietly drifting into "Direct."
Question 6: Did my ads actually cause these sales?
This is the question merchants think attribution answers and it never does. Attribution assigns credit among touchpoints that happened; it cannot see the counterfactual — which of those buyers would have purchased anyway. That's incrementality, and it's a different measurement entirely.
Retargeting is the classic trap: the audience is people already on your site, with a high base purchase rate with or without the ad, which is why platform-reported retargeting ROAS is typically the most inflated relative to its true causal contribution (Haus). The ground-truth check is a holdout test. Meta's Conversion Lift splits the audience into an exposed cell and a control cell; it requires at least a seven-day test window and at least 10% of the audience per cell (Haus). If Meta claims 500 conversions and the lift test finds 300 incremental, your Incrementality Factor is 0.6 — meaning a 4.0x platform-reported campaign is really a 2.4x incremental one. Reserve these tests for the big budget claims; they cost paused sales and take weeks.
The honest playbook: triangulate, then trust blended math
Here's the sequence that beats model-shopping when your numbers disagree:
- Fix your UTMs first. Most "mystery Direct" traffic is a tagging problem — all-lowercase, a documented vocabulary, and never tag internal links, which resets the session's real source (utm.io). This is free. Our Klaviyo UTM tracking guide shows the discipline for email specifically.
- Give each tool one honest job. Shopify = how many orders and how much revenue. Platform dashboards = directional in-channel comparison only. GA4 = journey shape. Survey = unclickable influence. Never compare one platform's self-credited number against another's — each grades its own homework. The Conversions API vs. pixel breakdown explains why Meta's own number drifts.
- Use blended reality as the guardrail. Total ad spend ÷ total new-customer orders (blended CAC) and total revenue ÷ total ad spend (blended MER) come straight from Shopify plus your ad invoices, and no attribution model can game them. If platform ROAS climbs while blended MER stays flat, credit got reshuffled — nothing real happened.
That third step is the one no attribution model can fake, and it's where Victor operates. He pulls your real orders, your real ad invoices, and your real supplier costs into one view, so the number he acts on is blended profit — not a dashboard's self-flattering guess. When you want an AI employee doing the margin math and the budget moves instead of you, put Victor to work on your store.
If you're shopping tooling rather than models, our rundown of the best AI-powered marketing attribution tool compares the options.
FAQs
What is the single most useful question attribution answers?
"Which channels introduce new buyers versus which ones merely close them?" Separating discovery from harvesting is what stops you from over-funding branded search and retargeting — the two channels last-click structurally over-credits. Everything else builds on getting that split right.
Can marketing attribution tell me my real profit?
No, not on its own. Attribution assigns credit for revenue; it says nothing about product cost, fees, or shipping. You have to layer margin on top — and Shopify can't help, since it doesn't ingest ad spend (Polar Analytics). PodVector AI's Victor joins orders, ad spend, and supplier costs to compute true per-order profit.
Why do Meta, Google, Shopify, and GA4 all report different numbers?
Because they're answering different questions. Shopify uses last non-direct click over 30 days (WeltPixel), GA4 uses data-driven attribution (Google Analytics Help), and Meta and Google each credit themselves within their own windows. There is no setting that makes them match — the rows measure different things.
Why is so much of my traffic showing as "Direct"?
Direct is mostly a catch-all for lost attribution, not people typing your URL. Stripped click IDs, expired Safari cookies, and untagged links all land there — Safari alone caps tracking cookies at seven days (Stape) and runs about 55% of US mobile sessions (Backlinko). Rising Direct share is usually a tracking-decay symptom.
Does attribution prove my ads caused the sales?
No. Attribution credits touchpoints that occurred; it can't see the counterfactual. Only holdout experiments — like Meta's Conversion Lift, which needs a seven-day minimum and 10% of the audience per cell (Haus) — measure whether sales were actually caused. A channel can win every attribution model and still be non-incremental.
Which attribution model should an operating store use?
The pros don't pick one — they triangulate across pixel, GA4, and a post-purchase survey, then anchor budget decisions to blended CAC and MER, which no model can distort. If switching models "improves" a channel, nothing real changed; model choice reallocates credit, not revenue.