The impact of AI on our day-to-day marketing doesn’t grow gradually – it moves in leaps.
Those moments come on the relatively rare occasions when we try something new, and see game-changing results.
That happened to me last week.
Quick background…
The latest few versions of Mike Rhodes’ excellent PMax script have included performance data by asset.
(Asset-level conversion data is now available in interface too – with a few caveats…

and the significant limitation of appearing without the click or cost data for needed for deeper evaluation)
The output of the PMax script on the other hand contains all of those CPA/ROAS/CvR KPIs needed to make meaningful comparisons.

Needless to say, this is extremely valuable (and thank you, Mike).
But last week I tried something that took this feast of data to another level.
I uploaded a screenshot of this output to Chat GPT, with a prompt along the lines of
“Looking at this table, what insights can you give me about which kinds of images are producing what kinds of results? Try to produce some insights that a human would be unlikely to spot”
(I’ve found that last sentence to a useful addition to various prompts).
The results were phenomenal.
Here’s a sample:


This was a real ‘Aha! moment’ for me.
The wider lesson I take from it is that AI analysis is REALLY good at merging quantitative data with less tangible attributes like patterns of similarity or difference between images.
Traditionally this kind of image analysis has relied on grouping images according to some predefined categories like ‘family’, ‘entertainment’, ‘gastronomy’… and aggregating results from each of these groups.
But getting AI analysis on the case allows us to be much more flexible in assessing what attributes actually make a difference to performance.
Not to mention the simple reminder that we can now generate quick analysis from the screenshot of any report.
The use-cases are extensive, to say the least…