Omnichannel attribution: foundations & core concepts

A customer sees a product in a TikTok video on Sunday night and does not click. On Tuesday she searches your brand name and lands on the site. Thursday she opens a promotional email. Saturday she buys the item in a store. One sale, four contacts, three systems that each recorded a fragment, and no system that recorded the whole thing. Deciding which of those contacts earned the revenue is attributionattributionA framework for assigning credit to the touchpoints that contributed to a conversion, so you can measure which channels and interactions actually drive results.View full definition →. The decision is not academic bookkeeping: the number you publish determines which channels get funded next quarter and which get cut. This lesson sets out the object itself, the models that split credit, the time windows that bound the question, and the identity limits that make every answer an approximation.

What attribution actually means

Attribution assigns credit for a conversion to the marketing touchpoints that preceded it. A touchpoint is any recorded interaction between a person and your marketing: a paid click, an ad impressionad impressionThe total number of times an ad or piece of content is displayed, regardless of clicks. Each display counts as one impression, even to the same person.View full definition → served but never clicked, an email open, a store visit scanned against a loyalty card, an inboundinboundA strategy that attracts prospects organically via valuable content (blog, SEO, social) rather than interrupting them.View full definition → call. A conversion is whatever event you have decided carries value, usually a purchase, a qualified lead, or a subscription start. Attribution answers two questions about a set of touchpoints: which ones get credit, and how much each one gets.

Three words get used loosely and are worth keeping apart:

• Touchpoint: one recorded interaction, with a timestamp and a channel.
• Path: the ordered sequence of touchpoints tied to a single identity, ending either in a conversion or in nothing.
• Credit: the fractional share of a conversion's value assigned to a touchpoint.

That third definition carries the whole subject. Credit is produced by a model. It is never observed. Your logs show that a person was exposed to some things and later bought something. They do not show what the person would have done otherwise. Every attribution figure you will ever read is a model's opinion about causal share, dressed as a number.

OmnichannelOmnichannelAn integrated approach connecting all customer touchpoints (physical, digital, mobile) into a seamless experience, with shared data and consistent context across channels.View full definition → attribution is attribution across touchpoints that span digital and physical environments, and across devices and platforms that do not share an identifier. It is harder than digital-only attribution for a practical reason more than a conceptual one: offline and app-store events arrive late, in batches, keyed to something other than a browser cookie, and sometimes not at all.

Sub-concept 1: attribution models

An attribution model is the rule or algorithm that distributes a conversion's credit across the touchpoints on the path. Two families exist.

Rule-based models apply a fixed formula regardless of what the data says:

• Last click gives 100% of the credit to the final touchpoint. Google Analytics reported this way by default for years, which is why so much of the industry's intuition is built on it. It flatters brand search, retargetingretargetingShowing ads to users who have previously visited your site or interacted with your brand, to bring them back and drive conversion.View full definition → and coupon affiliates, because those sit closest to the till.
• First click gives 100% to the opening touchpoint, inflating prospecting and display.
• Linear splits credit equally, treating a skipped pre-roll and a product page visit as equals.
• Time decay weights touchpoints nearer the conversion more heavily, on a decay curve you choose.
• Position-based, often 40/40/20, gives 40% to the first touch, 40% to the last, and spreads 20% across the middle.

None of these is derived from evidence. They are conventions, and their popularity comes from being cheap to compute and easy to explain in a meeting.

Data-drivenData-drivenAn approach where decisions are systematically informed by data analysis rather than intuition alone.View full definition → attribution (DDA) estimates credit from your own conversion data, comparing paths that converted with paths that did not and assigning each touchpoint a share based on its measured marginal contribution. Google made DDA the default in Google Ads in 2021 and by 2023 had retired most rule-based options from its reporting entirely. DDA is a better answer than 40/40/20, but it remains a model: it needs conversion volume to be stable, and it can only weigh touchpoints its own tracking can see. A channel that is invisible to the tracker scores zero, which is not the same as contributing nothing.

Sub-concept 2: attribution windows

An attribution window (or lookback window) is the period before a conversion in which a touchpoint stays eligible for credit. Outside the window, the touchpoint is treated as if it never happened.

Windows come in two kinds. A click-through window counts touchpoints the person actively clicked. A view-through window counts impressions served without a click, and it is almost always much shorter. TikTok Ads Manager, for example, defaults to a 7-day click and 1-day view window, with click options running out to 28 days.

Two consequences matter. First, changing the window changes your reported conversions without anything changing in the real world. Widening from 7 to 28 days will raise a channel's apparent performance overnight. Second, a window that is shorter than your actual consideration cycle silently deletes the top of the funnel. If people take five weeks to buy a mattress and your window is seven days, the awareness work will look worthless in every report you run.

One term to hold onto here: a self-attributed conversion is one that a platform claims using its own tracking and its own window, inside its own interface. Each platform applies this only to touchpoints it can see, which is why platform-reported conversions, added up, routinely exceed the number of sales the business actually made.

Sub-concept 3: the identity resolution problem

Before you can attribute anything across channels, you have to establish that the person who opened the email and the person who paid at the till are the same person. That is identity resolution, and without it omnichannel attribution is arithmetic on unrelated rows.

Deterministic matching links touchpoints through a known identifier the person supplied: a logged-in account, a loyalty number, a hashed email address, an order confirmation. Match rates here are high, commonly above 90% for the traffic that is actually logged in. Probabilistic matching infers that two anonymous records belong to one person from shared signals such as IP address, device characteristics and behavioural timing. Accuracy is materially lower, typically in the 60 to 80% range, and errors are not random: they cluster in shared households and shared networks.

The constraint has tightened from both ends. Third-party cookies have been restricted in Safari and Firefox for years, and Chrome's timetable has shifted repeatedly. On mobile, Apple's App Tracking Transparency, shipped with iOS 14.5 in April 2021, made access to the IDFA an explicit opt-in prompt, and most users declined it. The practical effect is that a large share of your paths are not paths at all. They are unlinked fragments, and the model quietly attributes each fragment as if it were a whole customer.

Sub-concept 4: user-level and aggregate measurement

Two levels of measurement sit under the same word, and mixing them up causes real confusion.

Multi-touch attributionMulti-touch attributionA method that distributes conversion credit across all marketing touchpoints in the customer journey, rather than crediting only the first or last interaction.View full definition → (MTA) works at user level. It needs a path per person, so it inherits every identity limit above, and it is blind to media it cannot tag.

Marketing mix modelling (MMMMMMA statistical approach that estimates how each marketing channel and other factors drive sales, guiding budget allocation.View full definition →) works at aggregate level. It regresses outcomes such as weekly sales against spend, price, seasonality, distribution and other drivers. It needs no personal data, so it can include TV, out-of-home, sponsorship and price effects, but it needs years of history and gives you nothing about an individual.

Both are estimates of the same underlying quantity from opposite directions. Choosing between them, and validating whichever you choose, is method work that comes later. For now, note only that MTA and MMM answer differently shaped questions and will not agree.

Real-world cases

Apple's ATT rollout is the clearest example of identity constraints redefining what attribution can be. Once IDFA access became opt-in, iOS app advertisers moved to SKAdNetwork, which returns campaign-level install data with a coarse conversion value, a randomised delay, and suppression of anything below a crowd-size threshold. User-level paths simply stopped existing on that inventory, and measurement shifted from counting individual journeys to reading aggregates. Worth noting that Apple sells advertising through Apple Search Ads and defines the measurement rules for everyone else on the platform, so it is not a neutral party in the design.

TikTok shows the window problem in its purest form. Much of the platform's influence is view-through: people watch, do not click, and come back days later through a brand search or by typing the URL. Under a 1-day view window and last-click logic, that sale is credited to search. TikTok's own attribution reporting counts what its window can see, and like every platform selling both the media and the measurement, its figures are self-attributed rather than independently verified.

CMO action items

• Write down, on one page, the model and the windows your reports currently use, per channel. If nobody can name them, your budget is being allocated by a configuration default someone accepted years ago.
• Track your deterministic match rate as a monthly metric: the share of revenue that lands against a known customer ID. That single percentage caps how much omnichannel attribution is available to you.
• Map the two or three most common real purchase paths for your top product, including offline steps, before you argue about credit. You cannot model a path you have never described.

Common mistakes that kill results

• Reading model output as causal fact. A channel appearing on many converting paths has demonstrated presence, not influence.
• Comparing or adding numbers built on different windows and different identity graphs. Two platforms claiming the same sale is the normal case, not an error to be reconciled by summing.
• Setting the window to fit the reporting calendar rather than the buying cycle, which makes considered purchases look like impulse purchases.
• Treating model choice as a technical setting. It is a budget decision, and it is usually made in a settings screen by someone who was never told that.