Gaurav Rai

AI | MARKETING | WEB 3 | FINTECH.
For more:

[Gaurav Rai @gaur_ai]

@buildwithshyam Night Travel seems amazing.

[Gaurav Rai @gaur_ai]

@nikitabier Woah

[Gaurav Rai @gaur_ai]

@Pinterest The pattern underneath all of it:

When platforms can't raise prices, they automate performance instead.

Pinterest's 1% pricing growth is the reason the AI roadmap looks the way it does.

Source: Pinterest Q2 2026 earnings, August 4.

[Gaurav Rai @gaur_ai]

@Pinterest beat on revenue. Beat on earnings. Record 640 million users. Raised margin guidance.

Stock fell 8% after hours.

The number that explains it is buried in the ad metrics, and it tells you why their whole AI roadmap looks the way it does:

[Gaurav Rai @gaur_ai]

@Pinterest 1. No product catalog and you wrote off Pinterest? Test it again. The barrier is gone.

2. Get fluent in AI-default setups now, while manual campaigns still exist as a control group.

3. Treat platform-reported lift numbers as directional, not proof.

[Gaurav Rai @gaur_ai]

The manually built campaign is being retired as the starting point and repositioned as an advanced option.

Three things I'd do:

[Gaurav Rai @gaur_ai]

Now put it next to Google.

Dynamic Search Ads retire in September and auto-upgrade into AI Max. AI Max left beta in April with 500,000 advertisers on it.

Two platforms. Same quarter. Same move.

[Gaurav Rai @gaur_ai]

Sit with that phrasing.

AI setup is the default.

Manual control is the exception, for people who "need" it.

[Gaurav Rai @gaur_ai]

Then CEO Bill Ready said the part that actually matters, on the call:

The goal is for nearly every lower-funnel campaign to start in an AI-powered best-practice setup, with added controls for more complex buyers who need them.

[Gaurav Rai @gaur_ai]

How hard they're pushing it:

Performance+ carried ~30% of lower-funnel ad revenue as of Q1.

Adopters grew lower-funnel spend 2x+ faster than non-adopters.

Updated delivery models: 28% ROAS improvement in testing for SMBs using ROAS bidding.

[Gaurav Rai @gaur_ai]

Read past the 6%.

Pinterest just opened performance advertising to everyone who was locked out of it.

That's a market expansion play wearing a creative tool costume.

[Gaurav Rai @gaur_ai]

Smart Assembly removes that requirement.

Upload several images. Pinterest's AI assembles and serves whichever combination performs best per impression.

Early alpha: 6% average lift in click-through rate.

[Gaurav Rai @gaur_ai]

Until now, running Pinterest's best-performing ad format required a product catalog.

No catalog, no Performance+ campaign.

That excluded service businesses, B2B, publishers, anyone selling something that isn't a SKU.

[Gaurav Rai @gaur_ai]

Which explains why every product they shipped this quarter is an automation product.

The one worth your attention is Smart Assembly.

[Gaurav Rai @gaur_ai]

@solopribuilds Information Leverage, you can't ask useful questions with little knowledge in Head and AI won't work without it.

[Gaurav Rai @gaur_ai]

And the satisfaction finding is the actual lesson.

Three-quarters satisfied. 11% wasted.

Satisfaction with a measurement tool tells you how it feels to use. Not whether it's right.

Source: StackAdapt (July 30), Affinity Solutions (May 15).

[Gaurav Rai @gaur_ai]

Three-quarters of marketers say they're satisfied with their measurement tools.

The same survey found those tools waste 11% of media budget on bad optimization signals.

Both numbers, Affinity Solutions, May.

Last week their data landed inside a product built for that gap:

[Gaurav Rai @gaur_ai]

Where I'd land:

Use it. Purchase-adjacent signal beats engagement signal. In-flight beats post-mortem. For travel and location-based, it's a straightforward upgrade.

Don't call it proof in a board meeting. Panel-based attribution has error bars nobody publishes.

[Gaurav Rai @gaur_ai]

That's better than clicks.

It's still a model.

[Gaurav Rai @gaur_ai]

But the link between your ad and that purchase runs through modelled attribution on aggregated spending behaviour.

StackAdapt's own release says so.

So the chain is: real purchase data, plus a model that assigns credit for it.