Click Consult’s Head of Paid Search has become a bit of a fan favourite with Benchmark crowds over the last few years, and his talk on machine learning did not disappoint
Karellen’s talk, which – as ever – managed to walk the line between humorous and informative, began with a brief look at the outliers in the argument as things stand; these, he states the two extremes fall in to two broad categories: those that believe in the impending AI apocalypse, and those that attest that machine learning is the only way forward.
As with anything, however, the answer is never really as black and white as we’d like to assume. Instead the answer falls upon a gradient that depends a lot on how things progress in a host of different areas.
From this brief discussion, Karellen moved on to explain a little how machine learning works before lamenting the black box nature of Google’s machine learning offerings for paid search.
The talk then moves on to detail the four main problems with black box algorithms and the difficulties that are inherent in a marketplace where every band is using the same algorithm – leading inevitably to a situation where it is impossible for any brand to win, with the main battle then being on maximum bid.
Only by using a mix of methods – including building your own bid optimisation spreadsheets – can we hope to develop any kind of competitive advantage that doesn’t just rely on bid.
Having wowed the audience, and securing his position among the most tweetable and enjoyable Benchmark speakers, Karellen wrapped up the talk with a summary of the four problems and his suggested solutions.
Problem one: – Very limited inputs available for users.
Solution: – Build your own tools instead (mostly not worth it).
Problem two: – Process is too ‘black box’.
Solution: – Test to ensure it matches your expected outputs. If possible make your own transparent tools.
Problem three: – No opportunity for competitive gain.
Solution: – Automate menial tasks but use the human touch to make competitive gains and add creativity.
Problem four: – Measurability lags behind.
Solution: – Accept machine learning in general, but overlay your own measurement from external analysis.
You can enjoy the full video of the talk here:
You’ll be able to see reviews of all of the Benchmark talks on our blog in the coming weeks, but to ensure you’re up to date with all things search, sign up to our blog. Alternatively, see how we can help to build your brand online by contacting us today.