Kinase Ranks Google’s Already Existing AI Tools for Digital Marketers
As Google’s PaLM 2 ushers in a ‘next gen’ for AI, Kinase looked back at Google’s existing AI tools. We ranked each one for how useful it has been for Digital Marketers.
On May 10th at I/O, Google officially launched PaLM 2. Google’s take on a Large Language Model AI is a direct competitor to Chat GPT. In terms of hype, Google is playing catch up with their Microsoft-backed rival.
Who will prevail? Perhaps neither, with Google already concerned about the lack of moats around AI.
However Google does have the advantage of a whole ecosystem to integrate AI into. It plans to build PaLM 2 into no less than 25 products and features, from their BARD chatbot, to Workspace apps (Gmail, Docs, Slides & Sheets), and of course Google Ads, which we will assess in due course.
What sometimes gets lost in the hype however is that AI is not new at Google. They have been deploying it, and Kinase has been harnessing it for a long time now.
So today we take the chance to look back at previous deployments, asking how successful they were and wondering how this bodes for the next big wave of disruption to sweep across the internet.
Conversion Modelling - 7/10
Measuring performance is critical to marketers. Entities from Apple to Regulators are busy making that harder (we don’t want to invade anyone’s privacy, we just need to know what ROI we’re getting, but that’s another story). Google’s tools include:
- Consent Mode & Enhanced Conversions, enabling AI to match and model missing data
- DDA (Dynamic Data Attribution ) in GA
- SSD/SSI (Store Sales Direct / Improved) to model Store Sales
Providing much better on- and offline sales data makes this a valuable tool, scoring it 7 out of 10.
However, gaps remain, especially across non-Google products and offline interactions, so it remains essential to also run robust geo-experiments to accurately understand performance.
Google Cloud Tools - 8/10
A powerful selection of tools that we’ve been able to make great use of. For instance, using the Cloud Vision API to automatically categorise images in order to flesh out Shopping feed attributes or better adhere to SafeSearch adult content policies.
Broad Match + Smart Bidding - 8/10
Understanding user value potential based on 100’s of signals far beyond the keyword itself is a task only an AI can handle, and the results have been impressive, unlocking additional volume and cost efficiencies for our clients.
Read our case study - Testing AI to Measure Overall PPC Impact for Lovehoney
The main reason for marking down is that it sometimes treats two different Brands as synonymous when actually they are arch competitors within a market, so checking those search terms and negatives remains an essential task.
Workspace v1 - 5/10
It’s amusing when Gmail suggests a phrase to reply with but little more. However tools like ‘speech to text’ transcription in Google meetings have proved to be useful. A lowish score for the initial foray here does not imply low expectations for the next phase however, with LLM promising orders of magnitude improvement when Duet launches.
https://workspace.google.com/blog/product-announcements/duet-ai
PMax AI video creative - 0/10
Sorry Google, but this was a great example of AI’s limitations. Think ‘my first animated PowerPoint’ with lots of scrolling text, static images and a terrible selection of music. Running auto generated video assets on YouTube performed so badly we had to develop a script to delete them as fast as they get made.
Run YouTube Solo and See it Soar
Contrary to popular belief, Performance Max isn’t a replacement for all your creative placements such as YouTube or Display and shouldn’t be treated as such.
DeepMind - 10/10
Not technically used directly in our digital marketing, but the source of much of the innovation we benefit from once it gets productised.
Acquired by Google in 2014 it made high profile advances in games ranging from the brute-forceable Chess to far more nuanced examples like Diplomacy. Excelled within closed rules sets, and invented new meta-strategies for Starcraft but little direct impact on the wider outside world (unless your friends won’t game with you anymore after one double-cross too many). Then AlphaFold predicted the structure of 200M proteins, a biochemical breakthrough that definitely gains the Kinase seal of approval.
By Richard Brooks, Kinase Founder
