What is influencer marketing attribution?
Influencer marketing attribution is the practice of assigning sales and other outcomes to the creator posts that caused them. It answers two different questions that are easy to mix up: which creator gets credit for an order you can see (attribution), and how many orders would not have happened at all without the creators (incrementality). Tracked attribution is fast and per creator; incrementality is slower and per program. A working measurement setup needs both.
The gap between them is large for creator content. People watch a post on TikTok, search the brand two days later, and buy on Amazon or in a store. None of that carries a code or a link. In Haus’s analysis of hundreds of incrementality experiments, TikTok advertising drove an average of 34% additional sales on Amazon and 57% in physical retail, and measuring across channels showed 1.9x the value of measuring the brand’s own site alone10. Haus measured TikTok ads in general, not creator posts specifically, but creator content travels the same path.
This guide compares the six attribution methods, shows what each one misses, and ends with a stack by budget. For the ROI formula and KPIs, start with the influencer marketing ROI guide.
Six influencer attribution methods side by side
Each method sees a different slice of the return. The table is the short version; the sections below explain how to set each one up.
| Method | What it counts | What it misses | Cost and effort | Use it for |
|---|---|---|---|---|
| Discount codes | Orders where the buyer typed the creator’s code | Buyers who skip the code, Amazon and retail sales; over-credits when codes leak to coupon sites | Low | Per-creator floor on site sales |
| UTM and affiliate links | Clicks that lead to a session and order inside the window | Mobile viewers who search later, cross-device buyers, view-only influence | Low | Per-creator site traffic and orders |
| Platform data (TikTok Shop, Spark Ads, partnership ads) | In-app orders and ad-attributed conversions in the platform’s window | Sales off the platform; the platform grades its own ads | Low, built in | Daily optimization on TikTok and Meta |
| Post-purchase survey | Self-reported “where did you hear about us” | Recall bias; buyers who skip the question | Low | Untracked site orders, directionally |
| Holdout tests (geo or audience) | Sales that would not have happened without the activity, across all channels | Per-creator detail; needs scale and a clean control | Medium to high | Calibrating everything above |
| Marketing mix model (MMM) | Each channel’s share of total sales over time | Per-creator and per-post detail; needs years of data | High | Budget split across channels |
Discount codes
A discount code is a unique code per creator that the buyer enters at checkout, which credits that creator in your store’s order data. Codes catch buyers who watch on a phone and type the URL later, which links cannot. They miss everyone who buys without the code, and they spread: once a code reaches a coupon-aggregator site, it credits the creator for buyers who never saw the post. Use single-purpose codes per creator, watch for sudden spikes from unknown referrers, and treat code sales as a floor.
UTM and affiliate links
A UTM link is your product URL with tags (utm_source, utm_medium, utm_campaign, utm_content) that analytics tools read to file the session under that creator9. In Google Analytics 4, credit then follows the property’s attribution settings: the first-click, linear, time-decay and position-based models were removed in November 2023, and the default lookback window is 30 days for acquisition events and 90 days for other key events8. That matters because many “influencer attribution model” articles still recommend time-decay or position-based splits that GA4 no longer offers. Affiliate links work the same way inside an affiliate platform, with that platform’s own cookie window, so record which window each report uses.
Platform data: TikTok Shop, Spark Ads and partnership ads
Platform reports are the only place you can see in-app purchases, and they follow the platform’s rules, not yours. On TikTok Shop Ads, a click credits orders placed within 7 days and a view within 24 hours; the most recent interaction wins, clicks take precedence over views, and one order cannot be credited to several ads1. Attribution is tied to your Shop ID rather than a product ID, so an ad built on one creator’s video can be credited with orders for other products in your shop1. Seller Center reports sales on the day they happen, while Ads Manager reports sales within the 7-day window, so the two never match line by line1.
When you run an affiliate creator’s video as a Shop ad, Seller Center shows GMV attribution between your ads and affiliate creators, and those ad-attributed orders may also earn the creator commission2. Budget for that commission on both layers. For Spark Ads, engagement gained from boosting (views, likes, comments, shares, follows) is credited to the original organic post3, so a creator’s “views” on a boosted post include paid views; separate them before you calculate cost per view. Outside TikTok Shop, TikTok’s web attribution windows can be set from 1 to 28 days for clicks and off, 1 or 7 days for views, with 7-day click and 1-day view as the default4.
This layer is bigger than many brands assume. In Kolect client programs, 80.4% of single-post creator collaborations included paid-usage, whitelisting or ad-code rights16, which means most creator posts can also run as Spark Ads or partnership ads, and their sales get split between “creator” and “ads” reports. The mechanics of both formats are in Spark Ads vs partnership ads.
Post-purchase surveys
A post-purchase survey is one question on the order confirmation page, such as “Where did you first hear about us?”, with a named-creator or “TikTok creator” option. It recovers orders that carry no code or link, and it is the cheapest way to see the size of the untracked share. It is self-reported, so people misremember and skip it. Report survey-attributed orders as their own line, never merged into tracked orders.
Holdout tests for incrementality
A holdout test compares a group that was exposed to the activity with a matched group that was not, over the same period, and reads the difference in total sales. There are two kinds. Platform conversion lift studies randomize users into test and control groups: Meta’s Conversion Lift is limited and requires a Meta representative6, and TikTok runs Conversion Lift Studies through account representatives, with over 1,400 advertisers having run one as of October 20245. Geo holdouts split regions instead, and you can run them yourself.
Holdouts are where the halo shows up. In a geo lift study TikTok published with Liquid I.V., 62% of TikTok’s incremental impact occurred on Amazon, and last-touch attribution had credited TikTok with 12 conversions11. Meta’s own white paper, as reported by PPC Land, found a 45% lower median cost per incremental purchase across 20 EMEA conversion lift studies when advertisers made at least 30% of their assets partnership ads; the analysis is Meta’s internal data at 92% confidence7.
Marketing mix models
A marketing mix model (MMM) is a statistical model that estimates how much each channel contributed to total sales from weekly spend, sales and control variables. Google’s open-source Meridian asks for at least two years of weekly data for geo-level models and three years for national models12; Meta’s open-source Robyn calibrates its models against geo tests and lift studies13. MMM is the only method that splits credit between creators, paid social, search and TV in one frame, but it cannot tell you which creator to rebook.
What every method misses: halo, delay and cross-channel sales
Tracked attribution misses three kinds of return, and all three are larger for creator content than for search ads.
- Halo. Sales in channels the post never linked to: Amazon, retail, the brand’s site via search. Haus’s averages above (34% more sales on Amazon, 57% in retail) are the scale to expect from TikTok exposure10.
- Delay. Creator posts keep working after the campaign. Haus found an additional 68% lift in the post-treatment window after experiments ended10; a 7-day click window cuts that off.
- Cross-channel effects. Creator content raises the response to other channels: people who saw the post click the brand’s search ad, and the search ad gets the credit. The same happens in AI answers when someone asks an assistant what to buy and the brand is named because creators reviewed it.
The return also includes outcomes that are not orders at all. Haus found 62% of TikTok-driven lift came from new customers for DTC brands10, and new customers buy again later. Branded search, AI-answer mentions and share of voice move before revenue does. In a Kolect client program for Jackery, the brand’s mention rate in AI answers rose from 10% to 30% over an 8-week program, tracked against a fixed set of questions14.
How to run a geo holdout for creator campaigns
A geo holdout tests what you can switch on and off by region. Organic creator posts cannot be geo-fenced, because a creator’s audience is wherever it is, so the geo test measures the paid layer (Spark Ads, partnership ads, TikTok Shop ads targeted by region) plus anything regional, such as local creators in one market. Measure the organic layer with a pulse: alternate weeks or months with and without posting and compare.
- Pick the layer. Choose what you can target by region: Spark Ads or partnership ads on creator videos, or creator activity limited to one market.
- Match the regions. Split states, metro areas or countries into test and control groups with similar sales over the past 8 to 12 weeks, across every channel you sell in.
- Set the length. Run long enough for the slow build: Haus’s TikTok experiments averaged 21 days plus a 9-day post-treatment window10. Keep everything else (prices, promotions, other media) the same in both groups.
- Measure everything. Compare total sales in TikTok Shop, your site, Amazon and retail, not just the tracked orders.
- Compute the factor. Incremental revenue divided by the revenue the platform reported for the same activity is your incrementality factor.
- Apply it until the next test. Multiply future platform-reported revenue by the factor in your weekly reporting, and retest each quarter or after a big change in creative or spend.
Worked example: one month, five readings of the same campaign
The numbers in this example are illustrative, made up to show the method; they are not client results. Only the creator fee is real data: it is the median agreed fee for 10K–100K-follower TikTok creators in Kolect client programs15. A brand sells a portable blender at $40 on TikTok Shop US, its own Shopify site and Amazon, with a 55% gross margin before creator costs.
The month’s spend: 40 TikTok micro creators at a $300 flat fee each15, $2,000 for samples and shipping, $6,000 of Spark Ads on the eight best videos, and 15% affiliate commission on TikTok Shop GMV from creator videos. That is $20,000 fixed plus $3,000 commission: $23,000. Break-even ROAS at a 55% margin is 1 ÷ 0.55 = 1.82.
Step 1: what each report claims.
| Report | Orders claimed | Revenue |
|---|---|---|
| Seller Center: affiliate creator videos (organic), counted on the day of sale1 | 300 | $12,000 |
| Ads Manager: Spark Ads on creator videos, 7-day click and 1-day view window1 | 200 | $8,000 |
| Shopify: creator discount codes | 90 | $3,600 |
| GA4: creator UTM links (40 of them also used a code) | 100 | $4,000 |
| Post-purchase survey: “a TikTok creator”, no code or link | 180 | $7,200 |
| Naive sum | 870 | $34,800 |
Illustrative figures; the naive sum double counts.
Step 2: deduplicate. TikTok Shop orders are counted once (500, $20,000), since TikTok credits one order to one ad at most1 and Seller Center splits GMV between ads and affiliate creators2. On the site, the 40 orders with both a code and a link are counted once, leaving 150 orders ($6,000). Survey orders stay on their own line.
Step 3: run the geo test. For four weeks, the Spark Ads ran only in test states; control states got none; sales were read for two more weeks, close to the 21 days plus 9-day post-treatment window Haus’s experiments average10. Across TikTok Shop, the site and Amazon, the test states sold $14,000 more than the control states would predict, against the $8,000 Ads Manager had reported: an incrementality factor of 1.75 for the paid layer, with $6,000 of the difference on Amazon.
Step 4: the five readings.
| What is counted | Revenue | ROAS | Contribution after creator costs |
|---|---|---|---|
| A. Tracked, deduplicated (costs include 40 fees at the $300 median15) | $26,000 | 1.13 | −$8,700 |
| B. A, with Spark Ads replaced by the geo-test result | $32,000 | 1.39 | −$5,400 |
| C. B plus survey-attributed site orders | $39,200 | 1.70 | −$1,440 |
| D. C plus 90-day repeat orders from new customers (120 orders) | $44,000 | 1.91 | +$1,200 |
| E. D plus branded search and AI-answer mentions | not in dollars | – | tracked as leading indicators |
Illustrative figures except the creator fee15. Contribution = revenue × 55% − $23,000.
The same month is a clear loss on row A and a small profit on row D. Three lessons follow. Decide before launch which row is the official one, as the ROI guide’s worked example recommends. Row D cannot be read until 90 days later, so report row B weekly and row D quarterly. And the survey row may partly overlap the site share of the geo-test result, so a strict finance team stops at row B; the honest report shows both.
A practical attribution stack by budget
Start with what is cheap and per creator, then add the methods that correct it as spend grows. The thresholds below are rules of thumb, not standards.
| Monthly creator budget | Stack | Decision it supports |
|---|---|---|
| Testing (a few thousand dollars) | One code and one UTM link per creator; TikTok Shop affiliate reports; one post-purchase survey question; a sheet with a precedence rule (platform order ID, then code, then link, then survey) | Which creators and hooks to rebook |
| Scaling (tens of thousands) | Above, plus Spark Ads or partnership ads in separate ad groups per creator; 90-day new-customer cohorts; weekly branded search; a geo holdout on the paid layer each quarter | How much to amplify, and the incrementality factor |
| Multi-market or multi-channel | Above, plus platform conversion lift studies via your rep, a pulse test for organic posting, an MMM once you have two to three years of weekly data, and AI-answer and share-of-voice tracking | How to split budget across creators, paid, search and markets |
Whatever the tier, write down three things before launch: the precedence rule, the attribution windows each report uses, and which row of the worked example is the official ROI. For setting up the campaign itself, see how to run an influencer campaign; for TikTok Shop affiliate mechanics, see how to find creators for TikTok Shop.
Counting the full return in one system
The hard part of influencer marketing measurement is not any single method; it is keeping codes, links, platform reports, surveys, tests and repeat purchases in one model, with one set of rules for every agency and tool. That is what the Kolect growth system does: it connects a brand’s creator, paid and repeat-purchase data and counts the full return of every budget, from orders and new-customer repeat purchases to branded search, AI answers and share of voice. It is run by Kolect’s team, the brand’s own team or its agencies, and the results Kolect clients have published show what it measures in practice.
If you are starting from scratch, set up codes, links, a survey question and a precedence rule this week, and plan your first geo holdout for the month after. To see how the growth system would fit one market and one category, start at /platform#start.