Using AI in ecommerce is possibly the most exciting thing to hit since ecommerce was invented in the mid 90’s
This article has not been written by a Robot and you can knwo this by the typohs I have left in on purpose !
I like to refer to it as automated intelligence as all the data it uses is already there and its making many assumptions unlike real intelligence does. However I believe we are at the very early stages of this. But not unlike rocket research, all you need is 1 or 2 pivotal discoveries and you are catapulted into a high speed development. In some areas I think its advanced and in others its a distraction.
Ecommerce content AI.
In ecommerce we have been writing content for products and told by search engines that unique content is key and the better the quality the better. After almost 30 years of product content writings I am guessing there are little or no products that have no content associated with them. The exception being large inventories of spare parts etc.
If we take for example whiskey – this is a product that is collected and curated by many people around the world. Its a product that matures with age and naturally gets rarer and rare as the years pass due to it being consumed and there being less bottles of limited released available. New entrants looking to fill in content on sites where they get a feed of data from the supplier and then they have to merge this into a uniform format to match the rest of their site, while also maintaining uniqueness and quality is a challenge. Many will look fr the lazy way out first and then realize google is a lot more cleave than we thought. In the past it took an expert in the field to write this content and curate it. It was a game winner in search engine rankings where the first with the content was basically the winner as they instantly ranked no 1 by being the first.
So AI generated content maybe a quick way to generate content for products on your site. But what if AI is using your site to create this content on other sites. Can you stop it ? After all is taking your content and using this as an influence to create other content. AI will possibly cite your content but no matter who borrows / steals or uses your content as an influence – your content is the origin and should remain on being cited as the source for the generated content on other sites, thus providing backlinks. All of this makes the assumption that googles way of ranking sites will remain the same and as we know the only constant in the SEO world is change.
The lines are now becoming blurry now as to who is the actual expert and who is the AI writer generating content.
There are news sites now generating 100% of their news from AI and making AI generated videos with AI narration by sourcing other information to make these reports. The veracity and source of the contributing content is the key to the accuracy.
The same applies to product content and information.
We have done some experiments on generating product content and we have seen where it could actually lead to a contagion of wrong information about a product proliferating the marketplace.
For the examples I used copy.ai
Example :
We seeded the term “Middleton very rare whiskey”
Note : whiskey is spelt with the Irish spelling and not the US spelling. Middleton very rare is a very specific whiskey released in limited batches every year. A very collectable bottle.
“this single barrel release of very rare is the perfect way to get acquainted with this world famous single malt. The color of honey and the taste of ripe fruit make this whiskey a true gem.”
The tasting note is too vague “what fruit ? ” a lemon or a banana ? very different taste !
version 2
“Distilled in 1923, Highland Park is rare whiskies that mainly obtained by collectors. Bottled at 46% ABV / 94 Proof, it has the color of honey and the taste of ripe fruit.”
The date is not correct. The location is not correct. Looks like its coming from the same source as version 1 ?
version 3
A rare and exquisite whisky of high quality aged for up to 30 years in oak casks, Highland Park has a delicate honey color, soft texture and subtle flavor with notes of dried fruit
Completely mixing itself up now. Highland Park is another whiskey and not middelton whiskey at all. Using vague words and has moved from “ripe fruit” to “dried fruit” a little more specific but the fundamentals are now completely wrong.
Test on a specific Honda Spare Part for a Motor Bike
Input is : “Honda ct110 1980 Pin A, Dowel, 10X14 (94301-10140)” Taken from https://j10.ie/94301-10140
Its a specific part for an engine block on a Honda c110 from 1980. Very specific input here.
Output
Pin A, Dowel, 10X14 (94301-10140)
or
A corona ring that is used to hold the clutch and pressure plate together.
“The first is correct – the 2 description is not correct at all. “
Getting the engine to provide more detail on the first suggestion
This Pin A, Dowel, 10X14 is used when bolt holes and length of material do not match up.
“Again this description is getting generic for a very specific part.”
Testing on “BEKO Freestanding A 9kg 1400rpm Washing Machine | WTL94151W2
option 1
The WTL94151W is an A+ energy rated 9Kg washing machine from BEKO. It features a 1400rpm spin speed and has a 1 year manufacturer’s guarantee.
“its accurate and short because its specification is well publicised online.”
option 2
Washing machines are supposed to make life easier but they are noisy, bulky and expensive. Not anymore! The Beko WTL94151W is a compact washing machine with nine kilograms capacity and 1400 rpm spin speed. You can take it with you when moving thanks to its carry handle on top. This model will fit nicely in any kitchen as it is fitted with sleek lines and modern design that highlights its functionality as well.
“Carry handle on the top ?? “
My feeling on this at the moment ( dec 2022 ) is that we are at a curousity stage but for real world accuracy and veracity and quality we are way off. The time it would take to correct the content might be better spent writing it from scratch. The data being suggested is at worst not correct and at best vague with the correct grammar and tone of voice.
No matter the result it gives you – I would always recommend checking the final content in a plagurism checker and also sanity checking it, else you could have some one returning a washing machine because there was no carry handle on the top !
Product Listing Display AI
Based on the customers first selection on the site then large categories can be sorted down to suit the customer more accurately. In B2B sales this pretty easy to do by listing all the previously purchased products in order of recency and QTY. But in retail it is a more difficult thing to do and this is where AI comes into play. Up until now this was the preserve and the secret sauce of Market Places where they had literally millions of shoppers and were deriving intelligence from them. ( albeit badly in my opinion ). So now using AI based on the shoppers activity in the current session is now going to make the listing of the product more clever.
Think of a shopper heading to a car parts site. They visit the windscreen wiper category. They then search for de-icer – now we know they are in a cold place, and wipers that are freeze resistant are more apt. Also there are other winter products that are more apt now. Snow covers for tyres, shovels, snow blowers etc etc.
Imagine now linking up the local weather of the shopper to the AI inputs. We will know what the weather was , is and will be. So we know if its wet or dry spell and if this is the case then we will list products in the category that are more suited.
Ok thats all very fine for accessories for cars but what about service parts. Almost the same can be applied. Purchasing an oil filter – chances are you are doing a service and you are going to need a whole lot of other parts like, spark plugs, rags, maybe some gaskets and new oil too ? A service is some thing where the parts are required quickly.
Shipping Method AI
So based on the product in the basket we can now upsell enhanced shipping services and charge extra for these based on the knowledge of what is in the shoppers basket. Example ordering spare parts, chances are you are out of action until these parts arrive, so speed of shipping is essential. Again the location of the shipper, the warehouse and the shipping destination can be used by the store owner to identify the most economic shipper to use and to set a fair expectation to the shopper of when the product is going to arrive.
When selling larger items where there is a longer wait time and sometimes the product is not made until there is an order for it may allow for different speeds of production for the customer to avail of. The customer may be willing to pay more to jump the queue and the store owner will have a good idea of this based on a profile of the customer built during their shopping session.
Cross Selling and AI
Many SAAS services have popped up creating AI based cross sells. These are using a matrix of product attributes with customer behavour to make relevant suggestions to the client of other products that they should consider when purchasing this product. It reminds me of the old school retail training I got when I worked in retail in the 80’s in the travel goods business. We would show the customer a suit case and ask the size they wanted. We would then show them 3 suit cases and depending on the first one they picked up to look at we would eliminate one and introduce another. This was helping the customer to narrow down and come to a decision quickly based on the product I was suggesting and steering them into . There was a delicate balance as to push the customer would risk destroying the sale whereas not pushing enough would allow them to cool off and indecision to come into play.
So AI has to collect the purchase patterns of lots of shoppers to do this correctly you would think. Well actually it does not, it can work on almost no data and just operate on a rules based set of suggestions based on the product attributes and price level. Example if we are buying aquariums then the selection is based on the area we have to put the aquarium into and the budget of the shopper and then the look of the aquarium. No point in showing an aquarium that’s cheap if its the wrong size. So once the rules and the hierarchy of the product attributes are setup then the corss selling becomes more relevant to the product.
Payments and AI
Once the shopper is on the site using AI the store can get a good indication of the demographic of the shopper and what payment methods they are likely to use. This is information that payment providers will have massive amounts of data on. It is really important as abandoned shopping carts due to payment issues are possibly the most painful of all. The shopper was almost in there transacting and your site was not as friendly and easy as it could have been for the shopper. Gone are the days of offering 1 payment method on a site . You need to offer as many as possible so as your shopper no matter where they are in the world are going to find a method that works for them.
Chat Bots using AI
Possibly the first successful use of AI in ecommerce has been in the service sector using chat bots on busy sites to triage customer enquiries before passing them onto real people if the issue is unique and not solvable automatically.
Then there is the backend / admin where AI in ecommerce can be very powerful.
However some AI chatbots on sites are so thin they can do more harm than good and only lead to frustration of the customer. Training chatbots to the point where the user is not aware its a bot takes along time and huge effort.
The time and effort required to set this up versus a real person long term is what makes it viable.
Pricing and AI
It can be used to forecast sales based on gathering information on other competitors in the market, the demand for the product and the scarcity of the product.
Combined with price watching services it can be used to maximize price while remaining the most competitive. This is possibly the biggest single instant result a store owner can do to raise their prices without hurting their competitiveness.
There are many SAAS services in the marketplace for watching and tracking prices and stock levels of defined competitors.
Using AI in marketing is going to help you spend your advertising dollar in a more effective way. It is too easy to waste alot of budget testing the market with advertising and not fully understanding the results. It may well be that you would never fully understand the results and may have to rely on AI to guide you on the best path.
I would see the biggest opportunity for growth being the development of tools that automatically place and bid on ads that will provide the best return on investment for the spend.
The true goal of a digital marketeer is 1 click 1 sale.
In conclusion :
AI is here to stay – its going to be an essential part of ecommerce platforms features. Those using it will be getting results that will make them pivot their business and sales offerings as a result of the intelligence they will get from their applications.
For content created by a bot I would think there needs to be a standard where content is marked as generated by a bot.
For SEO is a bad idea – as most search engines will recognize bot generated content.
However it will not solve every issue and there are other technologies that will fill the gaps. The content generation for products is still some way off.
about the author :
Sean Owens has been writing and building ecommerce sites since 1994 and comes from a tech and retail background. He is an expert in large scale complex ecommerce project delivery and promotion. He speaks a conferences , podcasts, publishes papers and contributes to ecommerce opinion pieces .