Artificial intelligence is defined by Wikipedia as intelligence exhibited by machines, particularly computer systems. We look at the implications of their use for marketers, particularly in relation to the Amazon ecosystem.
The two popular extremes of opinion on the subject of Artificial Intelligence (AI) are:
At MinsterFB we’ve been looking at the ways that AI might impact success on Amazon. We would suggest that neither of the above views are accurate.
Artificial intelligence can do lots of the jobs that would ordinarily have featured on a marketer’s ‘To do’ list, but just not quite as well as an experienced professional. Most of us can spot AI images and Linked In posts that have used the ‘Rewrite with AI function’. AI solutions here will improve over time, but for now, these are skills where humans deliver the best results.
AI systems are great at some analytical tasks that are so big you just wouldn’t have been able to do them without AI assistance. Some of these tasks have been packaged up by big data providers and are very accessible. In other cases you’ll need to know how to write those data prompts. There’s a whole skill set to be learned here. Different AIs are best suited for different tasks and the science of writing the perfect prompt for each would be a separate blog in itself. This means you can be more effective, but doesn’t necessarily reduce workload.
It’s also important to understand how artificial intelligence is at work out there in the ecosystem. The Google algorithm makes use of at least two AI platforms to help return the best results for a user’s query. We’ve all needed to learn what inputs are important in order to try to ‘game’ that system. This will increasingly be true for ad placements, social media success and other search ranking algorithms.
We’ve developed this thinking in the Amazon ecosystem and have 6 key focus areas for you where AI has a part to play.
AI is now selling to AI
Amazon are now making use of two AI-based systems to generate the best search results for optimal conversion. Cosmo uses a large-language model to translate shopper searches into ‘intent,’ while Amazon Rekognition ‘reads’ product images to assess their search relevance.
At the same time, artificial intelligence systems can create or suggest improvements to product listings, improving your ability to sell to these AI-driven systems.
Cosmo knows what you really really want
Amazon has developed COSMO (Common Sense Knowledge Generation & Serving System), a semantic SEO system. This system uses product attributes to build a knowledge graph incorporating relationships like product functions, usage context, materials etc to get to the intent behind the words the shopper typed into the search bar. This reflects our own agency experience that product attributes are becoming far more important in product listings and Amazon search optimisation.
The example given by Leo Sgovio at the Ambizion conference in Talinn was that of a winter coat. The shopper types ‘winter coat’ but the search results returned may not include the keyword ‘winter’ or even necessarily ‘coat’. This is because COSMO has determined the intent behind the search to be that the shopper wants to be warm. The search results are products that deliver high level warmth whatever their keyword content.
If you want to use AI to sell back to COSMO in the right way, you should evaluate your product listing for semantic relevance for your key search terms. This will give you guidance for optimising your product listings to single-mindedly meet the needs of the shopper intent you are targeting. Use prompts like ‘Acting as an NLP expert evaluate this copy for semantic relevance to the keyphrase XXX.’
Rekognition Reads Images
The Rekognition system developed by Amazon is a cloud-based image and video analysis service that they are using to ‘read’ product images. This helps to prioritise position within Amazon search. What does this mean? Firstly it will literally read any text on the images and treat this as though if were part of your product description or bullet points. Some practitioners, including John Derkits presenting at Seller Sessions, have suggested that the additional keyword indexing bump you get is worth the investment to create pack images that include keywords that you want to prioritise. Even if this isn’t the pack you actually sell in.
Secondly, the system tries to decode what’s in the image, using colour, shape, outline etc. This is where interesting errors can occur. Similar shapes can confuse Rekognition eg a baby’s cot with a room radiator. We’ve found that sometimes multipack product images are rejected because the outline is confusing. Amazon’s systems prefer the clean outline of, for example, a single jar, as it knows what it is, even though a multipack more clearly demonstrates the quantity the shopper will actually receive.
As a result, it’s worth removing elements from your images that may confuse. If you’re selling a photocopier, but in a lifestyle image you can also see a fire extinguisher, your product may start to rank (unfavourably) for fire extinguishers rather than photocopiers. Likewise, a visible serial number on your product packaging may confuse the system so you should remove these if possible.
If you want to check your images to see what Amazon’s Rekognition system is ‘reading’ then it’s available by API. Follow the link above.
Artificial Intelligence Proposes Listing Improvements
Even without using the complex systems outlined above, you can use AI to suggest improvements for your listings. You can upload your own product copy and images, together with one or two top-selling competitors, and ask eg Chat GPT to recommend changes. As with any of these suggestions, you may have to try different prompts. You might even need to try different AI platforms to get the results that are most helpful. You should tell the AI what role it is playing, ie evaluating the competitors and using the insight to improve your listing. Amy Wees gave a masterclass on this at the Ambizion conference, and gained some truly insightful additional selling points for her products.
Amazon’s AI-delivered review summaries can also help here. These sit at the top of the reviews and summarise the key themes without you having to read them all. They give you pointers for how to improve accuracy in your listing. This is a key element of securing favourable reviews, which in turn feeds the search algorithm.
Bullet Point and Image Generation
At a more prosaic level, a helpful role for AI is to create text and images for your product listing. AI can certainly help here, especially if you have a large number of products to go at.
Bullet Points
We have trialled AI for bullet points on product listings. When fed with some basic information about the brand and the product, together with the keywords that are important, you can get to bullet points that work. When you are working at scale or first listing a set of products they can play a part. For new accounts there’s so much else to get right, from choice of selling model to price to logistics etc. And that’s before you get into all the intricacies of the new line form itself.
If this approach to listing text makes it quicker and easier to get up and running then fair enough. However, AI generated copy is never quite as good as copy supplied by the client marketing team, who have had the opportunity to look at market research, trial different ways to express selling points and establish the most effective phrasing. If you use AI in the first instance, it’s worth coming back to these copy elements later.
Listing Images
Images are trickier. Amazon has its own AI Image Generator system within the Ad Console for creating assets. This can help if you don’t have what you need and don’t have a photography budget. First you choose your product. It then creates lifestyle images in which you can specify some of the content and also the theme (eg botanical, gifts, industrial etc). The lighting is reasonably well handled and the propping is appropriate (see example below). There are sometimes issues with scale and often there’s that sense that this is an AI image as opposed to something created in real life by a photographer.
However, beyond Amazon there’s very little in terms of AI image generation that will use an existing image. We’ve used AI to create all the images in this post, and they look great. However, when you try to put something that already exists into a context, it’s difficult to find a tool that works. If you can’t access a photography budget then give Amazon Image Generator a try.
Artificial intelligence in Amazon Advertising
What’s your favoured approach to Amazon advertising? Black box model where the system does the whole thing? Or fully manual so that you can see every moving part?
Amazon ads are increasingly complex with many different optimisations to consider. Most people have an instinctive leaning either toward or away from automated optimisation. When we onboard advertising accounts we can usually tell which way it’s been run historically. Every so often we trial a full black box AI system, although our team have so far always beaten them. Importantly they are also able to explain what approaches they have used to our clients which helps with trust.
However, one place where AI excels is in recommending advertising bid adjustments by daypart using Amazon Marketing Stream data. Processing hour-by-hour data for individual keywords and bids amongst thousands of campaigns into actionable recommendations is exactly the kind of task that AI was built for. At MinsterFB we use the Skai AI tool to process hourly ad data. We apply this to ad campaigns to optimise them by time of day. Processing this volume of data manually simply wouldn’t be possible.
There are other halfway house advertising tools. These take away some of the analysis and manual intervention without the loss of insight about what’s working. We’ll continue to evaluate different ad tools and adopt the ones that work for MinsterFB clients.
Conclusions
Artificial intelligence continues to develop quickly. Owners of retail, advertising, search and social platforms are incorporating it into many different elements of what they do. It’s important to stay up to date with how these developments are influencing the way that consumers engage with brands. AI can surface or react to consumer trends, but the core Marketing disciplines of anticipating and meeting consumer needs are unlikely to be replaced. Instead they will be re-shaped, and some of the skill sets required will evolve. Simply keeping up will require engagement, but you can always ask Bing’s Copilot to summarise the latest developments for you…