I have written a little bit about artifical intelligence in recent months. Enough to improve my superficial knowledge of its application in marketing.

What seems obvious is that much like most martech, AI will not endanger many jobs in the medium term. It merely creates more sophistication and more jobs, perhaps.

AI isn't about to write a convincing poem or a convincing marketing email. Okay, deep learning has got recommendations in its sights (e.g. Netflix), but keeping on top of 'quality' of content, cultural considerations, languages etc. is still incredibly difficult.

However, leaving more sophisticated implementations of AI aside, machine learning's chief benefit to the marketer is its relentless feedback, a.k.a. the pursuit of incremental improvement.

Algorithms can determine optimum product pricing, or optimise cost per acquisition in advertising (and target a user likely to click), or forecast sales, or aid segmentation.

AI a natural fit for conversion optimisation?

One area where machine learning would seem a good fit is conversion rate optimisation (CRO).

The assumptions of marketers are often confounded by A/B or multivariate testing, so allowing your website to quietly tweak itself based on machine learning would seem to make sense.

And, who'd have thought it, there are companies offering this very product - a SaaS-based CRO solution that uses AI. Sentient Technologies launched a product (called Sentient Ascend) in September 2016. 

Using a slightly specious analogy, Sentient's website explains: 'Our patented AI solution mimics biological evolution, enabling it to quickly learn, adapt and react to determine the best performing design from the building blocks you provide.'

The case studies are compelling though. Underwear brand Cosabella has used the software to test 15 different changes to the homepage header, category page, product page and cart layout.

Colours, headlines, button placement, and new layouts were being tested. You can see the changes being primed in the screenshot below.

cosabella sentient ascend

If one were to use strict multivariate testing for this number of changes, 160 tests would be required, and a whole lot more traffic.

In around a month, Cosabella had hit on design changes which generated 35% more conversions than the control.

Changes include button colours that match brand identity and highlight the checkout (presumably the shocking pink you can see used for buttons on the website at time of writing).

Branded messaging above the header was also changed and found to impact conversions.

What place for the conversion expert?

There's no doubt that, ultimately, software such as this will eventually eliminate a lot of the headache of conversion optimisation.

The consolidation of marketing tech and increasing sophistication of ecommerce platforms will mean that more and more marketers will be using machine learning in SaaS solutions, rather than utilising the precious services of a data scientist.

However, even as this tech allows testing at scale that's quick to set up and run, deciding what to do with it is still the job of someone with experience in analytics and web design.

Machine learning can only help if marketers have a hunch about where gains can be achieved. This comes from web analytics acumen and experience.

And for all the donkey work that can be eliminated, the marketer must still have the vision to suggest design changes that can be tested. If the change can't be envisioned, the software cannot test it.

Some of the changes made with AI powered CRO software

cro testing 

It's also worth noting that as personalisation of web communications and websites becomes possible, the scope for conversion optimisation increases, too.

Many might argue true website personalisation is not necessary in ecommerce, but when you split website traffic even fairly basically by nationality, landing page, search terms etc., all these characteristics impact on conversion.

Planning how to segment website visitors requires more testing and more expertise in analytics and conversion.

In summary...

The conversion agency/consultant may, if anything, become more relevant than ever as powerful CRO technology becomes easier to implement and more widely available.

Conversion consultants, machine learning experts, tell us what you think...

Ben Davis

Published 2 November, 2016 by Ben Davis @ Econsultancy

Ben Davis is Deputy Editor at Econsultancy. He lives in Manchester, England. You can contact him at ben.davis@econsultancy.com, follow at @herrhuld or connect via LinkedIn.

1109 more posts from this author

You might be interested in

Comments (9)

Comment
No-profile-pic
Save or Cancel
Pete Austin

Pete Austin, CINO at Fresh Relevance

Oh good grief. Changing button colours again. @Cosabella google "Novelty effect". Did the improvements persist and did you re-test? Serious question. Sorry to be a COC.
https://en.wikipedia.org/wiki/Novelty_effect

To be more positive, I really like how Cosabella republish PR articles as images for re-sharing. (At the bottom of some product pages.)

12 months ago

Chris Gibbins

Chris Gibbins, Director of User Experience & Optimisation at BiglightSmall Business Multi-user

Thanks Ben for the discussing this extremely interesting and relevant topic.
Actually from our perspective I think the prospect of AI powered testing / experimentation is very exciting! One of the great things of having a large creative and UX team working together alongside the analytics team to come up with ideas for CRO (or CXO as we call it) is the high number of quality test ideas and possible test variations we generate. Whilst this is great in one way, in normal circumstances for our clients (depending on traffic) it's only possible to include 3 or 4 variations at a time for any single experiment.
So if AI works well and lives up to its promises then this will allow us to be even more creative in what we can test and in the quantity of variants included, which will lead to a higher success rate and bigger uplifts in a shorter amount of time. So we'll be able to help our clients grow their ecommerce businesses even more quickly! So yes very exciting.

12 months ago

Ben Davis

Ben Davis, Deputy Editor at EconsultancyStaff

Good to hear, Chris. Let us know as and when you start to experiment.

12 months ago

Ben Davis

Ben Davis, Deputy Editor at EconsultancyStaff

@Pete, yes would be good to know what happens with new visitors alone, which I presume would control for novelty?

12 months ago

Parry Malm

Parry Malm, CEO at Phrasee Ltd.

@Ben:

"AI isn't about to write a convincing poem or a convincing marketing email."

Just wondering what you're basing the second half of that sentence on... because that's why huge brands around the world use our A.I. tech to do...

Poems, on the other hand, I agree.

:)

12 months ago

Ben Davis

Ben Davis, Deputy Editor at EconsultancyStaff

I set you up for that plug. Am I right in thinking that Phrasee doesn't optimise all body copy? Just the punchy stuff?

12 months ago

Parry Malm

Parry Malm, CEO at Phrasee Ltd.

@ben hahaha cheers I think :)

It ultimately depends on what the customer outcome is. We focus on copy that sells - if it's, for example, a mellifluous review of a Nickleback album, then a human (probably an idiotic one with awful musical taste) is best fit for it.

A.I. ultimately relies on patterns, both current and past, regardless of the school of A.I. For certain types of copy, there's no pattern to follow. But for other types of copy - A.I. can recognise patterns invisible to the human eye... and process much larger volumes of data than humans can.

So yeah, A.I. ain't gonna write a Man Booker Prize winning book any time soon. But does it need to? Books don't make money. Effective email marketing does.

12 months ago

Parry Malm

Parry Malm, CEO at Phrasee Ltd.

If you're really bored, here's a research paper from Google Brain et al that explores the theoretical limit of recurrent neural networks for language generation (one interesting method that has extreme limitations)

https://arxiv.org/abs/1602.02410

Good NLG platforms will use techniques like this as *part* of the A.I., but not all of it. If anyone ever tells you they have an RNN that can generate perfect language, run fast and far, because they liars.

12 months ago

Avatar-blank-50x50

Sander Nagtegaal, CEO at Instant

You may want to check out Instant.cm - a service that lets your website adapt its content to different audiences. Connect your Adwords account and make content variations of your own pages, based on automated advise - all meant to optimize conversion and Quality Score by using symmetric messaging. Read more here: https://instant.cm (disclaimer: I am the CEO).

7 months ago

Comment
No-profile-pic
Save or Cancel
Daily_pulse_signup_wide

Enjoying this article?

Get more just like this, delivered to your inbox.

Keep up to date with the latest analysis, inspiration and learning from the Econsultancy blog with our free Digital Pulse newsletter. You will receive a hand-picked digest of the latest and greatest articles, as well as snippets of new market data, best practice guides and trends research.