Prescient AI (@PrescientAI) header
Prescient AI (@PrescientAI) profile picture

The world’s smartest marketing decision engine, powered by the most dynamic MMM. Forecast, test, optimize, and scale top-of-funnel spend with confidence.

Miami, FL
Joined September 2021
Photo by Prescient AI (@PrescientAI)Photo by Prescient AI (@PrescientAI)Photo by Prescient AI (@PrescientAI)Photo by Prescient AI (@PrescientAI)Photo by Prescient AI (@PrescientAI)Photo by Prescient AI (@PrescientAI)Photo by Prescient AI (@PrescientAI)Photo by Prescient AI (@PrescientAI)Photo by Prescient AI (@PrescientAI)Photo by Prescient AI (@PrescientAI)
Amazon ad-attributed sales and incremental sales are two different numbers. Amazon credits a sale to an ad when a shopper interacted with it within a set window before buying.
1
9
When demand is already high, more of the sales Amazon credits would have happened anyway, and the attribution report doesn’t adjust for that.
1
2
Heading into Black Friday and Cyber Monday, that can make Amazon Ads look like the channel doing the work, while the campaigns that sent the shoppers there get none of the credit. Budget decisions tend to follow the credit.
2
Multi-touch attribution (MTA) can only credit what it can track, and audio leaves almost nothing to track.
1
12
Earlier this week, we shared that 89% of the modeled revenue podcast and audio ads drove arrived through a path the ads never touched. This piece explains why measurement setups built on clicks and tracked touchpoints miss it.
1
Privacy changes leveled the playing field on data access. Almost every measurement vendor now works without cookies or user tracking. That means the differentiator has shifted from what data a vendor can collect to how they build and validate the model on top of it.
1
7
Precision claimed across every scenario is a less reliable sign than precision claimed where the evidence actually supports it.
1
Treat privacy compliance as the entry requirement. Then compare on model construction, validation method, and honesty about the limits of what the data can support.
Ask about one model on its own. “Show me attribution for my Shopify model only” now returns just that model, in Prescient, ChatGPT or Claude, instead of every model added together.
1
13
Modeling the Market, issue nine: Podcast and audio ads. In the week of October 19 to 25, 2026, 89% of the modeled revenue podcast and audio ads drove arrived through a path the ads never touched. Someone hears an ad, then searches the brand or buys on Amazon.
1
8
That’s halo effects: revenue a channel drives indirectly, through a storefront or search path the ad itself never touched. Nearly all of this channel’s modeled revenue landed on two storefronts, 55% on the brand’s own site and 44% on Amazon.
1
5
A budget review that counts only the revenue podcast and audio drives directly would be working from about a tenth of it. The breakdown is in the slides, and the last slide explains how we measured.
3
Promo codes, affiliate links, and last-click tracking capture only the buyers who act on creator content right away. Many of the people a creator actually moved search the brand later, or buy through Amazon or a retailer.
1
25
→ Stop-and-start spend is hard to read. With sparse data, a traditional model can’t cleanly separate the channel’s effect from background trend.
1
1
A single post from a small creator won’t leave a signal any model can read on its own. You can measure a single creator when they run as their own campaign, so the most useful questions are about campaigns and programs, not individual posts.
1