Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Ecommerce Show - Imogen Collins - Leveraging AI...

Ecommerce Show - Imogen Collins - Leveraging AI in Digital Marketing - With Notes

An updated version with all my notes on the slides for ease of use

Imogen Collins

September 21, 2023
Tweet

More Decks by Imogen Collins

Other Decks in Marketing & SEO

Transcript

  1. Leveraging AI in Digital Marketing for Ecommerce Imogen Collins Genie

    Goals Speakerdeck.com/imogencollins linkedin.com/in/imogenpickles
  2. about us An award winning PPC, Paid Social & Affiliate

    marketing agency 30 employees based out of Cambridge In Google’s top 1% of EMEA Agencies One of AWIN’s top 20 agency partners 45th Best Place to Work in the UK 2023 2
  3. • What is AI? • What changed? • How it

    is being used • The downsides • Tips and tricks
  4. What is AI? • Alan Turning described the Turing Test

    (a test to prove whether an AI could exhibit behaviour indistinguishable from a human) in 1950 • The first AI program was created in 1955 • A huge amount of progress has been made in the past 70 years
  5. 6 Artificial intelligence can be defined as a broad field

    of computer science focused on creating machines or systems that can perform tasks that typically require human intelligence.
  6. 7 Machine Learning is a subset of AI focused on

    training computers to learn from data. Like teaching a computer to recognize spam emails by showing it thousands of examples.
  7. 8 Deep learning goes further than machine learning and allows

    a system to learn using vast quantities of raw, unlabeled data, using neural networks inspired by the human brain.
  8. 9 Computer vision is a deep learning process which uses

    neural networks to decipher images within a photo or video, the way humans do with their eyes.
  9. 10 Natural Language Processing is all about enabling computers to

    understand, interpret, and generate human language.
  10. Notice anything strange about the images? These were generated in

    Midjourney using very simple prompts (“artificial intelligence” etc) and show how the data/images used to train AI can create a bias in the type of results they generate. AI is not an objective, unbiased arbiter – it reflects the bias of its training.
  11. 12 This Google Trends graph for “artificial intelligence” shows we’ve

    seen a huge shift in how much AI is being talked about. Understandably this has led to a lot of marketers wondering “how should I be using AI? How is everyone else using it? Can I trust AI to do a good job?”
  12. What changed? • AI has gotten fun and accessible for

    more people and through a route which allows people to be creative and see value quickly and cheaply • The real ways that AI impact us are not through tools like ChatGPT and Midjourney • The big stuff has been woven into the fabric big tech companies for years
  13. Data Privacy • Data privacy has been driving change since

    the Cambridge Analytica scandal in 2018 • As tech giants have removed features which were not aligned with data privacy laws, AI has been there to fill the performance gap. • We live in a world where we are expected to provide highly personalised experiences to our customers, with increasingly less access to data about our customers. • As we move towards a cookieless world, first party data and contextual targeting are going to be a part of the story • AI has allowed us to continue to granularly target based on audience signals without compromising on the data privacy standards that are becoming increasingly important
  14. How are we already using AI and machine learning in

    e-com? Examples: • Recommendation engines • Supply chain and inventory management • Data Security and Fraud Detection • Dynamic pricing • Customer service (messaging bots)
  15. 16 AI is used in: • Understanding search queries and

    delivering better results • Voice assistants • Optimising targeting, creative and bidding • Improving routing, traffic prediction and recommendations on maps • Recommending videos, content, groups, friends, products based on user behaviour • Detecting spam, fraud or harmful content • Email and content categorisation • Organising and tagging photos and creating videos and animations from existing assets • Translating text • Curated newsfeeds on social media • Messenger bots • Facial tracking and gesture recognition in VR and AR
  16. Pmax & Advantage+ In PPC and Paid Social, we have

    been impacted through smart bidding technology which evolved into automated campaign types, Google’s Pmax (November 2021) and Meta’s Advantage+ (August 2022 to select brands). AI analyses user behaviour in real-time, adjusting ad placements, targeting, and bids to maximise conversions.
  17. 19 These tools are a black box. Google and Meta

    struggle to give adequate information about what is going on in the box, what is driving the good or bad performance, answers to questions about odd events (weird uplifts in clicks or impressions etc).
  18. 2 These tools lack visibility of: • brand vs. generic

    • Keywords • Auction insights segmented by product type etc • channels/where your ads are appearing • What is doing well or badly in general
  19. 2 These campaign types lack control – in Advantage+ there

    is no audience targeting in the traditional sense. No age, gender, interest or lookalike targeting. No control over placements and platforms, or CPA goals. Everything is completely open and broad. And your only targeting option is country.
  20. 2 The algorithms are only as good as the instructions

    they are given and the data they are trained on. Poor quality data or the wrong instructions can lead to very poor decisions being made.
  21. 2 Optimising automated campaigns means starting with a period of

    data generation - i.e. paying for clicks without hope or expectation. The learning phase costs money and that money goes straight into the pocket of Google or Meta.
  22. 2 AI isn’t creative. Google can automatically find the best

    performing ad creative, by testing every possible combination of the assets that you give it. Because it’s a really smart tool, it is going to be able to make predictions about the types of assets that are going to drive the best results but it won’t be able to know if things sound weird or off brand or look ugly or clash - it’s up to us to make sure we’re putting the right constraints in place from the beginning to stop those things from happening.
  23. 2 AI can’t fix a bad strategy - if conversion

    rate is bad for every single combination, it won’t tell you what the problem is, it will just keep trying out combinations and throttle your traffic down to nothing. These automated campaign types are made to drive conversions, but most digital strategies are more complex than simply “drive conversions”.
  24. 2 A full funnel strategy has differentiated messaging, audiences and

    KPIs for different stages of the funnel. This can’t be replaced with a single campaign.
  25. Top Tips • Game the algorithm • Manage the learning

    phase • Inject creativity • Have a strategy • Regain control
  26. Top Tips Game the algorithm • As much data as

    possible • GAds tag & enhanced conversions in Google, consent mode in GA4, cAPI in Meta • Customer email lists uploaded in both • New vs. Existing users: in Google set a value for new users based on lifetime value to encourage the algorithm to spend less on existing customers, in Meta use the % cap to prevent overspending on existing customers in Advantage+
  27. Top Tips Manage the learning phase • When launching a

    new pMax campaign, Start with a ROAS target much lower than what you want to achieve (100% is a good place to start) • Once you’re achieving this, gradually increase this in small (less than 20%) increments up to where you want it to be
  28. Top Tips Inject creativity • Use as many assets as

    possible • All possible types and sizes of assets to prevent the system auto-creating its own assets • Keep text in images to a minimum, and the focal point to the centre of the image • Use pinning in pMax to control the messaging
  29. Top Tips Have a strategy • These campaign types are

    revenue/conversion driving campaigns • As we all know, we can’t just keep putting more money into bottom of funnel campaigns and expect more revenue at the same ROAS • These campaigns need to exist as a part of a full funnel strategy with different campaign types for awareness and consideration to be successful
  30. Top Tips Regain control • Pmax allows more control than

    Advantage+ • You’re able to mimic different business objectives with campaign structure, utilising several pMax campaigns for example to set a different budget or target for different product groups (based on margin, value, stock count, business priorities etc) • Separating brand and generic traffic is possible and can be very important for certain retailers
  31. In Summary… • AI is not new • AI is

    not a strategy • AI is a tool