What it means
Attribution is the answer to "which marketing actually worked?" There's no perfect model — every approach (first-touch, last-touch, linear, time-decay, position-based, data-driven) makes a trade-off. Modern teams typically combine a platform model for in-channel optimisation, a self-reported "how did you hear about us?" field for directional truth, and incrementality tests to validate the channels that matter most.
Worked example
Last-touch credits all closed deals to branded search. Switching to a position-based model (40% first, 40% last, 20% middle) reveals that LinkedIn ads start two-thirds of the deals that branded search closes. The team doesn't cut LinkedIn — they double it.
Why it matters
Bad attribution funds the channels that close deals and starves the channels that create them. Over a few quarters, that compounds into a brand nobody's heard of and a pipeline that's shrinking.
Common mistakes
- Treating any single model as ground truth.
- Trusting platform-reported conversions in isolation — every platform overclaims.
- Skipping incrementality tests, the only way to know what actually moved revenue.
Example — Marketing Attribution in practice
Suppose a Noon shopper sees an Instagram ad, later clicks a Google search ad, and finally buys after opening an email reminder about an abandoned cart. Last-click attribution gives 100% of the credit to email. A multi-touch model instead splits credit, say 30% to Instagram, 30% to search, and 40% to email, giving Noon's team a fairer view of which channels actually moved the sale.
لنفترض أن متسوقًا على نون يشاهد إعلانًا على إنستغرام، ثم ينقر لاحقًا على إعلان بحث في جوجل، وأخيرًا يشتري بعد فتح رسالة تذكير بالبريد الإلكتروني عن سلة متروكة. يمنح نموذج النقرة الأخيرة كل الفضل للبريد الإلكتروني. بينما يوزّع نموذج الإسناد متعدد اللمسات الفضل، مثلًا 30% لإنستغرام و30% للبحث و40% للبريد، ليمنح فريق نون رؤية أكثر إنصافًا لما حرّك عملية البيع فعليًا.
Marketing Attribution, properly understood
Attribution assigns credit to touchpoints along a customer's path using a chosen model — last-click gives 100% to the final touch, first-click gives it to the first, linear splits it equally, time-decay weights touches closer to conversion more heavily, and data-driven or algorithmic models weight touches based on patterns observed across many paths. The inputs are the full touchpoint history per converting customer, which requires stitching ad-platform click and view data, website or app analytics, and CRM conversion events into a single identity — the hard part of attribution is this identity-resolution plumbing across devices and sessions, not the choice of model itself.
Cross-device and cross-app attribution is especially lossy in Gulf markets, where a purchase journey routinely spans Instagram or TikTok discovery, a WhatsApp conversation with a sales rep or a friend's recommendation, and a separate browser or app session for the actual purchase. WhatsApp-influenced conversions are close to invisible to standard pixel-based tracking, so multi-touch models built purely on trackable digital touchpoints systematically understate the real influence of word-of-mouth and WhatsApp-driven channels.
No attribution model is objectively "correct" — each one embeds an assumption about which touchpoints matter, and every ad platform reports its own attributed conversions using its own model and lookback window. That's exactly why the same sale routinely appears as "caused" in three different platforms' dashboards simultaneously. A quick way to spot the problem: sum attributed conversions across all platforms and compare that total to actual total sales — the overshoot is over-attribution made visible.
Sense-check attribution against incrementality testing, which measures causation rather than correlation, and against MMM for a channel-level reality check that doesn't depend on individual-level tracking at all.
Attribution model choice should also match the decision it's meant to support — a linear or time-decay model is often more useful for budget-allocation conversations across channels, since it credits the full path rather than one touch, while last-click remains useful for narrower, bottom-funnel optimization questions like which specific ad creative closed a sale. Picking one model as the default and switching models only deliberately, with everyone aware of the change, avoids the common trap of quietly re-running the same report with a different model until the numbers say what a stakeholder wants to hear.
Put it to work
- Pick one attribution model as the primary decision-making view, not each platform's own number.
- Sum attributed conversions across platforms and compare to total actual sales to spot over-attribution.
- Invest in identity resolution across devices and sessions before trusting cross-channel attribution.
- Treat WhatsApp- and referral-influenced conversions as understated by default in the region.
- Sense-check attribution output against incrementality tests and MMM periodically.
- Match the attribution model to the decision at hand, and keep the default model consistent.
Turn the theory into real pipeline.
Get a free 60-second growth audit of your site, or talk to a strategist about your funnel.