Why last click and TikTok's own column disagree
The TikTok measurement hub explains how TikTok counts a conversion. An event becomes a reported conversion when TikTok can tie it to an ad exposure within a window: click-through attribution, where the person clicked the ad before converting, and view-through attribution, where they saw it without clicking, each with selectable windows set in the attribution settings, and an engaged-view option for video that counts a longer watch as more than an impression. The TikTok Ads Help Center in the sources lists the current options and defaults, and this guide does not quote them, because they change.
A web analytics property works differently. It credits the session in which the conversion happened, usually to the last source it recognised, and it never saw the ad impression. A person who watched a TikTok ad, did not tap it, and later searched for the brand appears in analytics as a search visitor. The TikTok marketing pillar puts the consequence plainly: TikTok tends to sit at the top and middle of the funnel, feeding other channels, and its measured return looks worst when it is judged only on last-click sales.
It is tempting to conclude that TikTok's column is right and last click is wrong. It is not that simple. TikTok's count includes people who saw an ad and would have bought anyway, and platform reports on every network do the same, as the TikTok Ads versus Meta Ads comparison notes: each platform attributes conversions under its own windows, often counting views, and both claim people who would have bought anyway. Last click and platform attribution answer different questions, and neither answers the one that decides a budget. The attribution models framework sets out what each model assigns credit for, and the guide to choosing an attribution approach explains why the decision, not the model, should decide which view is used for what.
| View | What it counts | What it cannot tell you |
|---|---|---|
| Last click in web analytics | The last recognised source before the conversion | Any effect from an ad that was seen but not clicked |
| TikTok click-through | Conversions after a tap on a TikTok ad, within the chosen window | Whether the buyer would have come anyway |
| TikTok view-through and engaged view | Conversions after an ad was seen or watched without a tap, within the chosen window | How many of those buyers the ad influenced |
| Holdout test or lift study | The difference between people or places that saw ads and those that did not | Which ad, audience or creative caused it, without further tests |
| Mix model | The estimated contribution of each channel to total outcomes over time | Short-term or small changes, without enough history |
View-through credit kept as its own line
View-through credit is where the disagreement is largest and where the temptation to add numbers together is strongest. The measurement hub's advice is to treat view-through conversions as a separate line in every report rather than folding them into a single total, because they are the component most likely to be counted elsewhere as organic or direct. A View-Through ConversionA conversion credited to an ad that was shown but not clicked, within a set view window. is not wrong; it is unproven. Some people who saw the ad were moved by it, and some were already on their way, and the report cannot tell which.
TikTok provides views in Ads Manager that compare how results shift under different attribution windows, which the hub calls the quickest way to see how much of a campaign's reported value depends on view-through credit. Read that dependence before arguing about TikTok's value: a campaign whose reported return comes mostly from view-through is a campaign whose case rests on the part a test has to confirm. The comparison of view-through and click-through conversions explains the two kinds of credit for any platform, and the pixel and Events API tutorial is where the underlying events are made trustworthy, because no amount of careful reading rescues a pixel that double counts.
Three rules keep the line honest. Never add view-through conversions to click-through conversions or to analytics conversions in a single total. Keep the attribution setting written into the measurement plan with the date it was chosen, since changing it changes the numbers. And report the view-through line next to the evidence that tests it, so a reader sees the claim and the check side by side. The monthly TikTok report tutorial builds that layout into a report the business reads every month.
Geo holdouts and lift studies
The direct way to measure what TikTok adds is to withhold it from some people or some places and compare outcomes with those who saw it. The incrementality testing guide explains the designs, how to choose cells, size and duration, and how to read lift and its uncertainty for any channel; this guide does not restate any of that. What it adds is the choice between the two routes open to a TikTok advertiser.
A geo holdout stops or withholds TikTok ads in a set of matched regions and compares their outcomes with regions where ads ran. It uses the business's own records, orders or leads by region, so it measures the outcome the business cares about, and it captures effects that no tracking sees, including the ad watched on TikTok and the purchase made later in a shop or through search. It needs regions the business can target and measure separately, and TikTok's location targeting, documented in the help centre, is what makes the withheld regions possible. It suits advertisers whose sales can be read by region and who can live with a region going without TikTok for the length of the test.
A lift study run by TikTok splits people into exposed and held-out groups inside the platform. TikTok offers lift studies for eligible advertisers, as the TikTok against Meta comparison and the TikTok setup for brand awareness both note, and the help centre documents how a study is set up and what qualifies. The awareness setup adds the practical point: a study has to be booked before launch and has conditions of its own, so a team that decides to measure lift halfway through a flight will usually find it too late. The comparison of geo holdouts and conversion lift studies sets out the trade between the two in general: the platform study is easier to run and reads platform-tracked outcomes, while the geo test is harder to run and reads the business's own outcomes. Whichever route is chosen, the holdout test readiness checklist checks the booking, the withheld regions and the reading date before the test starts.
Which outcome the study measures matters too. A Brand LiftThe difference in brand measures between people exposed to a campaign and a comparable unexposed group. study asks exposed and held-out people about recall or awareness; a conversion lift study counts actions. The comparison of brand lift and conversion lift explains when each answers the question, and for TikTok the rule follows the objective: an awareness flight is judged on brand lift, a sales campaign on conversions, and a business that wants to know whether awareness on TikTok became sales later needs a conversion test or a geo holdout over a longer period. In every design the Holdout GroupA randomly chosen part of an audience kept from exposure to measure what exposure caused. is the point; a test without one is a report. The academy checkpoint on evidence for what social channels add tests last click against the platform's own column, view-through credit and the choice between a geo holdout and a platform lift study.
Signals outside the platform: branded search and how-did-you-hear answers
Some evidence costs nothing to collect and should be running before any test. Neither kind proves cause, but both see things a pixel cannot, and both make a later test easier to plan and to believe.
Branded SearchSearch queries that contain a company's or product's brand name. is people searching for the brand by name. If TikTok makes people aware of the brand, some of that awareness will show up as branded searches and direct visits, which last click credits to search and direct. Record a baseline before a TikTok push, read it during and after, and compare against the same period in earlier years to allow for the season. The brand awareness and lift guide describes branded search, direct traffic and recall surveys as outside measures of awareness, and the awareness setup reads them at the end of each flight against the baseline taken before it. Branded search rises for many reasons, so it is a signal to explain, not a result to claim.
How-did-you-hear answers ask the customer. A single question at checkout or sign-up, with TikTok as one of the options and a free-text field, records the exposures customers remember, including ads they watched without tapping. The self-reported attribution guide covers wording the question, coding the answers, comparing them with tracked sources and reading around their biases; people remember some channels better than others and name the last thing they noticed. Read against tracked attribution, a TikTok share in the answers well above its share of last-click orders is a reason to test, not a verdict.
Mix models for larger budgets
A marketing mix model, or MMMStatistical model of channel impact., estimates each channel's contribution to total outcomes from aggregated spend, outcomes and controls over time, without tracking individuals. The marketing mix modelling basics guide explains the inputs, adstock and saturation, the outputs and the data requirements. For TikTok its strength is that it does not depend on clicks or pixels at all, so it can see an effect that reaches sales through search, direct visits or shops.
Its limits decide when it is worth building. A model needs enough history and enough variation in each channel's spend to separate TikTok from everything else that changed at the same time, and an advertiser whose TikTok spend has been small or flat gives it little to work with. The comparison of mix modelling and incrementality testing explains how the two work together: tests are sharper but narrow, models are broad but uncertain, and a model calibrated with test results is more credible than either alone. For most TikTok advertisers, a mix model is the third step, after view-through has been separated and a test has been run, not the first.
Choosing the evidence by budget and stage
The right evidence depends on what is at stake and what the business can measure. A business new to TikTok does not need a mix model; a business spending heavily across many channels cannot rely on a how-did-you-hear question. The order below is general practice, not a TikTok rule.
| Situation | Evidence to use | Why |
|---|---|---|
| TikTok is new and spend is modest | View-through on its own line; branded search baseline; a how-did-you-hear question | Cheap, always on, and shows whether a test would be worth paying for |
| An awareness flight is planned | A brand lift study where TikTok offers one, booked before launch; recall and branded search against a baseline | Awareness is judged on awareness, and the study cannot be added afterwards |
| Spend is growing and finance asks whether it works | A geo holdout, or a TikTok lift study on conversions where offered | The budget question needs a counterfactual, and these supply one |
| Sales happen offline or through other channels as well | A geo holdout read on the business's own records | It captures outcomes no pixel sees |
| TikTok is one of many channels with years of varied spend | A mix model, calibrated with holdout or lift results | Separates channels over time and puts TikTok in the context of the whole budget |
| Results from different methods disagree | Report them side by side with what each measures | Each answers a different question; none is added to another |
Whatever is chosen, the result has to reach a budget decision. Once a test gives an estimate of the orders or revenue TikTok added, divide by spend to get an incremental return and compare it with the return the business needs; the ROAS calculator does the arithmetic. The incrementality guide's section on moving from lift to budget decisions explains how to set budgets around a measured return and when to test again. The academy exercise on building evidence of what TikTok and YouTube add for a fishing tackle retailer practises this choice for a business whose finance lead wants to cut TikTok on its last-click return, beside the report an organic YouTube channel can honestly support.
A sleepwear brand tests what TikTok adds (illustrative)
Glorrentz Sleepwear runs a geo holdout before deciding on its TikTok budget
Glorrentz Sleepwear, an invented direct-to-consumer brand selling pyjamas and robes online, spends £18,000 a month on TikTok. On last click in its analytics, TikTok is credited with 140 orders a month, a return of 0.9. TikTok's own column, on the account's chosen windows, reports 520 conversions, 330 of them view-through. The finance director proposes cutting the budget by half.
The marketing lead keeps the view-through count on its own line in the monthly report, adds a how-did-you-hear question at checkout, and proposes a geo holdout before any cut. With an analyst, the team splits its 12 sales regions into matched pairs on past orders and pauses TikTok in one region from each pair for 6 weeks, keeping every other channel unchanged.
| Measure over the test | Regions with TikTok | Holdout regions |
|---|---|---|
| Orders, indexed to the pre-test period | 104 | 91 |
| Branded searches, indexed to the pre-test period | 112 | 97 |
| Checkout answers naming TikTok, share of answers | 17% | 6% |
The team estimates that TikTok added about 280 orders a month at full spend, roughly double the last-click figure and well below TikTok's own count, giving an incremental return of 1.8 against the 1.5 the business needs. The brand keeps the budget at its tested level, books a repeat holdout for the following year, and reports the three views side by side rather than choosing one.
Common mistakes
- Treating last click as the answer. It never sees an ad watched but not tapped, so it cannot say what TikTok adds.
- Treating TikTok's column as the answer. It credits people who would have bought anyway, as every platform report does.
- Adding view-through to a total. Keep it on its own line until a test says how much of it is real.
- Changing the attribution setting mid-period. The numbers change with it, and the trend becomes meaningless.
- Deciding on lift after launch. Lift studies are booked before a flight; a holdout is planned before the budget changes.
- Testing without a holdout. A before-and-after comparison is a report, not a test.
- Claiming branded search as proof. It is a signal to explain, moved by seasons, press and other channels.
- Building a mix model too early. Without history and varied spend, it cannot separate TikTok from everything else.
- Picking the method that flatters. Choose the evidence before seeing the results, and report what each method measures.
