First-touch, linear and last-touch ROAS
The same ad can have three very different ROAS numbers depending on how credit for a sale is split across the ads a customer saw. The trick is matching the model to the job each ad is doing.
The model decides how credit is split. The window decides how far back a touchpoint can be and still count.
What are attribution models?
Customers rarely buy after one ad. They might discover you through a reel, search for you on Google a week later, then come back through a retargeting ad and buy. Three ads touched that sale. ROAS needs a rule for who gets the revenue.
That rule is the attribution model. None of them is the truth. Each answers a different question, and each makes a different kind of ad look good.
How each model splits credit
Take one customer and one ₹3,000 order. On day 1 they see a Meta reel and click. On day 6 they search on Google and click a non-brand search ad. On day 9 they click a Meta retargeting ad and buy.
| Touchpoint | First touch | Linear | Last touch | Any touch |
|---|---|---|---|---|
| Day 1 · Meta reel | ₹3,000 | ₹1,000 | ₹0 | ₹3,000 |
| Day 6 · Google search | ₹0 | ₹1,000 | ₹0 | ₹3,000 |
| Day 9 · Meta retargeting | ₹0 | ₹1,000 | ₹3,000 | ₹3,000 |
| Total credited | ₹3,000 | ₹3,000 | ₹3,000 | ₹9,000 |
Windows: how far back a touch can count
Every model comes with a window. A 3-day last-touch window only credits an ad if it was the last click within 3 days of the order. A 30-day first-touch window lets an ad that found the customer up to 30 days earlier take the credit.
Indian D2C journeys are often long, with COD hesitation, price comparison and festive-sale waiting. A short window quietly drops the ads that started those journeys. That's why prospecting needs a longer window than retargeting.
- 1Track every touchRecord each ad click per visitor, stitched across browsers on the same device.
- 2Join to the orderWhen an order lands, pull every touch inside the window.
- 3Split by the modelHand out the order's revenue using the model's rule.
- 4Roll up per adSum credited revenue per ad and divide by its spend.
Worked examples
Example 1: Three campaigns, three models
An illustrative month. The same ₹4,90,000 of revenue from orders these three campaigns touched, split three ways.
| Campaign | Spend | First-touch revenue | First-touch ROAS | Linear ROAS | Last-touch ROAS |
|---|---|---|---|---|---|
| Meta prospecting reels | ₹1,00,000 | ₹3,20,000 | 3.20 | 2.10 | 0.90 |
| Google non-brand search | ₹80,000 | ₹1,20,000 | 1.50 | 2.00 | 2.20 |
| Meta retargeting | ₹60,000 | ₹50,000 | 0.83 | 2.00 | 3.73 |
Revenue credited per model: first touch ₹3,20,000 / ₹1,20,000 / ₹50,000, linear ₹2,10,000 / ₹1,60,000 / ₹1,20,000, last touch ₹90,000 / ₹1,76,000 / ₹2,24,000. Each column sums to ₹4,90,000. Highlighted cells are the model each campaign should be judged on.
- ProspectingLooks like a loss on last touch (0.90), yet it started journeys worth ₹3,20,000. Judge it on last touch and you'd cut the top of your funnel.
- RetargetingShines on last touch (3.73) because it closes sales others started. On first touch it barely registers, which is fine for its job.
- SearchSits in the middle on every model. Linear is the fairest read for an ad that helps along the way.
Example 2: Same model, different window
The prospecting reels campaign above, judged on first touch with two different windows.
Which model for which ad
Judge each campaign by its funnel stage, and hold every channel at that stage to the same model. This is the practice we recommend as a starting point.
Common mistakes
- One model for everythingLast touch everywhere starves prospecting. First touch everywhere overfunds it.
- Summing any-touch revenueEvery ad gets full credit, so the total is many times real revenue.
- Comparing across modelsA 3.2 on first touch and a 3.7 on last touch aren't a contest. They answer different questions.
- Windows too shortA 7-day window cuts off customers who take weeks to decide, which is common for considered purchases.
Frequently asked questions
Which attribution model is the most accurate?
None of them on its own. Each answers a different question. Accuracy comes from tracking every touch and order correctly, then using the model that matches the ad's job.
How is this different from Meta's 7-day click, 1-day view?
Meta's setting is a last-touch-style model that only sees Meta's own touches and counts views. These models see every channel's clicks on your site and give each order to one owner.
Why does linear use 14 days and first touch 30?
Middle-of-funnel ads work on people already considering you, so their influence is closer to the purchase. Discovery can happen weeks before. Tune both windows to your own purchase cycle.
Related metrics
Switch models without switching tools
BooleanMaths tracks every touch and order, so you can see first-touch, linear, last-touch and any-touch ROAS side by side for any window.
First-touch, linear and last-touch ROAS
The same ad can have three very different ROAS numbers depending on how credit for a sale is split across the ads a customer saw. The trick is matching the model to the job each ad is doing.
The model decides how credit is split. The window decides how far back a touchpoint can be and still count.
What are attribution models?
Customers rarely buy after one ad. They might discover you through a reel, search for you on Google a week later, then come back through a retargeting ad and buy. Three ads touched that sale. ROAS needs a rule for who gets the revenue.
That rule is the attribution model. None of them is the truth. Each answers a different question, and each makes a different kind of ad look good.
How each model splits credit
Take one customer and one ₹3,000 order. On day 1 they see a Meta reel and click. On day 6 they search on Google and click a non-brand search ad. On day 9 they click a Meta retargeting ad and buy.
| Touchpoint | First touch | Linear | Last touch | Any touch |
|---|---|---|---|---|
| Day 1 · Meta reel | ₹3,000 | ₹1,000 | ₹0 | ₹3,000 |
| Day 6 · Google search | ₹0 | ₹1,000 | ₹0 | ₹3,000 |
| Day 9 · Meta retargeting | ₹0 | ₹1,000 | ₹3,000 | ₹3,000 |
| Total credited | ₹3,000 | ₹3,000 | ₹3,000 | ₹9,000 |
Windows: how far back a touch can count
Every model comes with a window. A 3-day last-touch window only credits an ad if it was the last click within 3 days of the order. A 30-day first-touch window lets an ad that found the customer up to 30 days earlier take the credit.
Indian D2C journeys are often long, with COD hesitation, price comparison and festive-sale waiting. A short window quietly drops the ads that started those journeys. That's why prospecting needs a longer window than retargeting.
- 1Track every touchRecord each ad click per visitor, stitched across browsers on the same device.
- 2Join to the orderWhen an order lands, pull every touch inside the window.
- 3Split by the modelHand out the order's revenue using the model's rule.
- 4Roll up per adSum credited revenue per ad and divide by its spend.
Worked examples
Example 1: Three campaigns, three models
An illustrative month. The same ₹4,90,000 of revenue from orders these three campaigns touched, split three ways.
| Campaign | Spend | First-touch revenue | First-touch ROAS | Linear ROAS | Last-touch ROAS |
|---|---|---|---|---|---|
| Meta prospecting reels | ₹1,00,000 | ₹3,20,000 | 3.20 | 2.10 | 0.90 |
| Google non-brand search | ₹80,000 | ₹1,20,000 | 1.50 | 2.00 | 2.20 |
| Meta retargeting | ₹60,000 | ₹50,000 | 0.83 | 2.00 | 3.73 |
Revenue credited per model: first touch ₹3,20,000 / ₹1,20,000 / ₹50,000, linear ₹2,10,000 / ₹1,60,000 / ₹1,20,000, last touch ₹90,000 / ₹1,76,000 / ₹2,24,000. Each column sums to ₹4,90,000. Highlighted cells are the model each campaign should be judged on.
- ProspectingLooks like a loss on last touch (0.90), yet it started journeys worth ₹3,20,000. Judge it on last touch and you'd cut the top of your funnel.
- RetargetingShines on last touch (3.73) because it closes sales others started. On first touch it barely registers, which is fine for its job.
- SearchSits in the middle on every model. Linear is the fairest read for an ad that helps along the way.
Example 2: Same model, different window
The prospecting reels campaign above, judged on first touch with two different windows.
Which model for which ad
Judge each campaign by its funnel stage, and hold every channel at that stage to the same model. This is the practice we recommend as a starting point.
Common mistakes
- One model for everythingLast touch everywhere starves prospecting. First touch everywhere overfunds it.
- Summing any-touch revenueEvery ad gets full credit, so the total is many times real revenue.
- Comparing across modelsA 3.2 on first touch and a 3.7 on last touch aren't a contest. They answer different questions.
- Windows too shortA 7-day window cuts off customers who take weeks to decide, which is common for considered purchases.
Frequently asked questions
Which attribution model is the most accurate?
None of them on its own. Each answers a different question. Accuracy comes from tracking every touch and order correctly, then using the model that matches the ad's job.
How is this different from Meta's 7-day click, 1-day view?
Meta's setting is a last-touch-style model that only sees Meta's own touches and counts views. These models see every channel's clicks on your site and give each order to one owner.
Why does linear use 14 days and first touch 30?
Middle-of-funnel ads work on people already considering you, so their influence is closer to the purchase. Discovery can happen weeks before. Tune both windows to your own purchase cycle.
Related metrics
Switch models without switching tools
BooleanMaths tracks every touch and order, so you can see first-touch, linear, last-touch and any-touch ROAS side by side for any window.