> For the complete documentation index, see [llms.txt](https://docs.roivenue.com/roivenue-resources/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.roivenue.com/roivenue-resources/roivenue-training/core-concepts/last-touch-vs.-ai-data-driven-attribution.md).

# Last Touch vs. AI Data-Driven Attribution

If you've ever pulled the same report under two different attribution models and gotten two very different answers, you're not alone. Comparing Last Touch and AI Data-Driven (AIDD) is one of the most common sources of confusion in Roivenue. Neither number is "wrong." They're answering two different questions.

This page explains why the two models disagree, how to read them side by side, and why AIDD numbers can shift slightly over time even when nothing in your account has changed.

### Two different questions, not two different answers to the same question

**Last Touch** asks: *which single touchpoint happened right before the conversion?* It then gives that one touchpoint 100% of the credit. It's simple and easy to explain, but it ignores everything that happened earlier in the journey.

**AI Data-Driven** asks: *how much did each touchpoint in the entire journey actually contribute to the conversion happening?* It spreads credit across every touchpoint based on patterns learned from your historical data. See [AI Data Driven Attribution](https://docs.roivenue.com/roivenue-resources/roivenue-training/core-concepts/ai-data-driven-attribution) for how that model is trained.

Because they're answering different questions, they will only ever agree by coincidence – for example, on a very short journey with a single touchpoint.

### Why the same channel shows different revenue in each model

Take a channel that's frequently the *first* touchpoint in a journey – a display or social prospecting campaign, for instance. Under Last Touch, that channel gets almost no credit, because it's rarely the final interaction before conversion. Under AIDD, the same channel can show meaningful revenue, because the model recognizes it played a real role in starting journeys that eventually converted.

The reverse is also true: a channel that tends to close journeys (branded search, retargeting) usually looks stronger under Last Touch than it does under AIDD, because Last Touch hands it 100% of the credit for journeys it didn't start.

### Why Direct often looks so different between the two models

Direct is the channel most people notice first when comparing models, and it deserves its own explanation – see [Direct & Organic Traffic](https://docs.roivenue.com/roivenue-resources/roivenue-training/core-concepts/direct-and-organic-traffic) for the full picture, including the ZeroOutDirect setting that redistributes Direct credit into other touchpoints. In short: Last Touch tends to overstate Direct's share, because "the customer typed the URL directly" is a very common last step even when paid channels did the actual work earlier in the journey. AIDD, especially with ZeroOutDirect enabled, corrects for this.

### Reading both models together

The [Model Comparison](https://docs.roivenue.com/roivenue-resources/roivenue-training/roivenue-app/attribution/model-comparison) page is where you look at Last Touch and AIDD side by side for the same period and channels. When you compare them, focus on the *shape* of the difference rather than trying to reconcile the totals exactly:

* Channels that score much higher under AIDD than Last Touch are typically upper-funnel channels doing real work that Last Touch can't see.
* Channels that score much higher under Last Touch than AIDD are typically closing channels that are getting more credit than their actual influence on the journey.
* A channel scoring similarly under both models is a good sign that its role in the journey is straightforward.

### Why AIDD numbers can change after the fact

The AI Data-Driven model is retrained every 14 days on the latest customer journey data. That means if you pull an AIDD report today and pull the exact same report again in a month, small differences in historical numbers are expected – the model has learned from more data since then and refined how it distributes credit. This is normal behavior, not a data error. Last Touch numbers, by contrast, never change retroactively, since the rule behind them ("100% to the last touchpoint") never changes.

#### FAQ

<details>

<summary>Which model should I trust?</summary>

Neither model is universally "correct" – they answer different questions. AI Data-Driven is generally the more accurate reflection of each channel's actual contribution, because it accounts for the full journey rather than just the last step. Last Touch remains useful as a simple, stable reference point, especially when comparing against how ad platforms report their own performance (most platforms default to a Last Touch-like logic internally).

</details>

<details>

<summary>Why did my AI Data-Driven revenue for last month change since I last checked it?</summary>

The model retrains every 14 days on the latest data, which can shift how credit is distributed across historical touchpoints. Small movements are expected. If the change is large or a channel's numbers look implausible, reach out to your CSM or the [Help Center](https://scanmar.atlassian.net/servicedesk/customer/portal/87).

</details>

<details>

<summary>Can I compare Last Touch and AI Data-Driven for the same date range and channels at the same time?</summary>

Yes, that's exactly what the [Model Comparison](https://docs.roivenue.com/roivenue-resources/roivenue-training/roivenue-app/attribution/model-comparison) page is for.

</details>
