My GenAI Edge Summit Reflections (Part 1)
What is the e-commerce shopping experience going to look like in 2030 - 2035?
What is the e-commerce shopping experience going to look like in 2030 - 2035?
What are the consequential use-cases that GenAI could deliver for e-commerce?
How can e-commerce organisations prepare for this coming future?
About the Author:
Co-founder of Datacop, agency that fulfils marketing operation roles in large eCommerce companies such as OluKai, Melin, Roark, Visual Comfort and Company, Dedoles and others.
In early September me and Lukas had the opportunity to attend the Edge Summit 2024 in London, where we had the opportunity to be confronted with the above questions of the future of our industry. The event, organised by NVidia, Google and Bloomreach, brings together 350+ industy leaders in e-commerce, technology and marketing - to share experiences and discuss major industry opportunities and challenges. At this summit the focus has been on what is the potential of GenAI for the e-commerce industry, consider the future ahead and what can companies do today to be prepared to capitalize on GenAI.
From all the key note speeches, open table discussions, conversations during the networking breaks I came away with 4 big takeaways.
GenAI will take another 10 years to truly disrupt the e-commerce industry
There are 4 GenAI use-cases disrupting the e-commerce industry (potentially)
Bloomreach Tech is well prepapred to power future GenAI applications in e-commerce
What can E-Commerce companies do today to prepare?
In part 1 of the reflections - I will dive deeper into the the first 2 takeaways:
GenAI will take 10 years to disrupt e-commerce
GenAI is no doubt a disruptive technology. Anyone who has worked with chatGPT will recognise that the potential is clearly there. In November 2022 it took chatGPT, 5 days to cross 1 million unique users and by August 2024, chatGPT is reported to have over 180M monthly active users and has recently crossed 1M paid users of chatGPT. (reference)
GenAI is already improving the productivity of software developers, helping with new breakthroughs in scientific and pharmaceutical research, enabled real-time translation between languages. GenAI’s freely accessible video and image generating tools is both causing an uproar in the intellectual property and creative industries; while also spawning a new genre of AI-generated entertainment. Many ask if this technology will be as foundational as perhaps the internet itself.
Given this backdrop of excitement how and when will this technology disrupt e-commerce? What tricky problems can GenAI solve and which opportunities will it open for the industry?
That is yet, hard to say. Despite the hype, many stakeholders at the Edge Summit cautioned that it will take at least 10 maybe 15 years to truly see the transformative potential of GenAI to come to e-commerce. Why is that?
It is important to appreciate that not even 2 years have passed since chatGPT went public. This technology and its wider application are very much in the earlier stages of the adoption curve of a new disruptive technology. To see why it may take some time, even years, consider the four “barriers for implementation” GenAI as a technology will need to overcome in e-commerce companies for the big results to come through.
Computational costs per user is still relatively high for B2C e-commerce applications compared to other B2C marketing technologies like Email / SMS / Push Notifications. It is clear the cost of GenAI is falling quickly, but it isn’t there yet.
Working and interacting with GenAI chatbots is a new skill. Both professionals and consumers are still learning how to best prompt and converse with chatbots. While 180M users worldwide of chatGPT is impressive, it is still less than 10% of the 2.1 billion people who have bought an item online. Users won’t get value from these technologies if they don’t know how what to expect from them, what not to expect from them and what sort of prompts they should start the conversation with.
The data infrastructure at many e-commerce companies is often not suitable for AI or GenAI use-cases. In many e-commerce companies data architecture & collection & maintenance is still a big challenge. Without good data - forget about good AI applications.
GenAI applications in e-commerce will take time, effort and cash to materialise. It is not yet clear what sort of applications will have a positive ROI on their investment. It will take a few years and perhaps even some failures to figure out what they are.
Don’t be swept away by the hype. The future is coming to e-commerce, but it won’t happen overnight. With this and the barrier to implementation in mind - there were 3 memorable quotes from key note speakers at the Summit I thought would be valuable to share.
The first quote was by Raj De Datta, CEO of Bloomreach → “focus on the consequential 10% of information about GenAI the rest is noise”. Because there is a lot of hype out there, thanks to undeniable breath-taking applications in different industries, it is easy to see every problem as something that can be solved by AI. Remember, we are still at the beginning of the adoption of GenAI in e-commerce.
The second quote was by Garima Singh, Chief Digital Architect of IKEA → “…good scoping of needs and opportunities for each AI investment is important. What is the problem we are solving? Is AI really necessary? Do we have the data available?.” Because this terrain is still relatively unexplored it can be easy to make costly investments in projects that don’t make sense. Ask critical questions like Garima to tease out whether a GenAI use-case is really going to be consequential.
The last memorable quote to be mentioned here was by Dan Finlay, CEO of Debenhams → “When investing into AI and Data project it is important to adopt an iterative approach and a Experiment / Evaluate / Learn mindset” Define your business outcomes and connect these with measurable KPIs. Did the use-case save XX number of man-hours because it automated certain manual tasks? Did the use-case drive incremental Y%+ increase in traffic, engagement or revenue?
4x (Potentially) Consequential GenAI Use-cases for e-commerce
As of right now; there only a few truly GenAI use-cases actually being implemented in e-commerce. Use-cases that are worth mentioning are:
Automated summarization and sentiment analysis of reviews and organic content
Automated translation for multillingual content / markets
SEO optimization of the front-end of websites
Most agree, that the above - while useful, they are not the sort of thing to “move the needle” in a significant way.
So what are some of the consequential experimental use-cases that were showcased, discussed or DEMO’d at the summit? Many of these are still in an experimental stage, but they provide an insight into what the future of our work and the future of e-commerce customer experiences look like.
UC1 - Documentation GenAI-powered Chat
A proven consequential use-case for GenAI in many organisations and industry has been in the field of Documentation. Retail Banks, Audit Firms, Law Firms and the like are all experimenting and implementing GenAI chatbots to engage with their mountains of documentation, creating “Digital Librarians” In short, companies like Bloomreach that have a rich documentation portal feed this documentation into a LLM and enable its users to ask natural language questions of the documentation.
The documentation-assistant has proven already to be a superior option to searching for the “right page” that contains the information I need, usually involving the classic “keyword search” method. In addition, the documentation-assistant by Bloomreach can answer harder questions that either required lots of time of research / or a consultation with a more experienced professional. Consider the question in the screenshot below.
Without the assistant the user would have to first find the adequate documentation page and then find the part of the documentation page that answers their question.
This feature is easy to access at documentation.bloomreach.com when you click on the search button on the page and would recommend it to both novice and experienced users of Bloomreach to use it when they have technical questions about the tool.
Unfortunately, I do not have access to how the success of this feature is measured. I would imagine this use-cases (probably) reduces the number of “easy” questions for the Support Team at Bloomreach, while improving the overall ability of users to discover solutions and answers to their questions on their own.
At Datacop, we have added the documentation assistant among our tools of choice and can attest that information retrieval is faster thanks to the assistant and there is a category of “easier to medium” difficulty questions anyone can quickly answer on their own without having to reach out to other members of the team or support.
UC2 - AI Co-Pilot “Loomi”
An experimental use-case that has been released is the Co-Pilot for users directly within the interface of the Bloomreach application, named Loomi.
Loomi’s primary function is to assist users of Bloomreach Engagement with building Reports / Segmentations / Expressions / Aggregates using only the user’s natural language prompt. Loomi is also being developed for Scenarios as well, with the same principle - building Scenario logic only using the user’s natural language prompt.
The promise of this use-case is two fold - unburden the experienced with “manual work” and enable the inexperienced to learn faster and be able to use more advanced features of the tool without the need of an experienced hand. This is what has already happened in the general software development industry - developers in languages like SQL, Javascript, Python, R, etc. are reporting GenAI co-pilots helping them standing up initial drafts of scrips quickly, de-bugging and particularly helpful in new situations the developer has not come across. Another common use is asking chatGPT to “describe” what a code is doing when you are unfamiliar with a script.
There are challenges this use-case faces in adoption and delivering consequential change. For experienced users, they are finding the interaction similar to working with a junior that just started working with the tool. The task needs to be well defined, well prompted and thoroughly QA’d. It often will take less time to simply do it yourself than to bother with the above. On the other hand, novice users are not able to check the results from Loomi themselves - so it has been hard to find users who are actually using this
I believe however, that given enough time, testing and development this feature will disrupt the way marketers and analysts work day to day. I imagine that experienced users in the future will first have Loomi build 70-80% of what they needs and then the final edits will be either done via an iterative conversation with Loomi, or they will simply make the final edits themselves. On the other end, novice users will be able to onboard a lot quicker into tools like Bloomreach, because they will be able to interact and ask questions with an experienced AI at any point during their learning journey.
From my personal experience testing this feature - I had mixed results were 50% of the time Loomi did what I wanted correctly and 50% of the time it missed the mark. I would recommend users to be using this feature to learn how to use it best to prepare for the future - but I have not seen this feature be used in earnest yet. For the moment it seems Loomi is unable to “describe” a Report or Scenario - or even “explain” why a Report or Segmentation is returning the value it is returning.
UC3 - Gen-Content / Image Email
Great potential is expected of GenAI’s ability to generate text, images and videos. It is the most commonly known use of GenAI since it was released to the public and has been subject to fierce debate and hype.
In e-commerce, the use of generating text has been mostly used to help Copywriters and Marketers with ideas. In e-commerce companies where there are multiple languages involved, text and copy that has been approved by a human can now be translated by GenAI instead of professionals. A big challenge with the use of GenAI images in e-commerce is the lack of consistency of a single object across images. This makes the generated images effectively unusable for Products listings, where there are strict needs and requirements of how the image looks like so as to not raise wrong expectations with the shopper.
A startup on the market - Ecomtent, is trying to solve just that. Its AI product image generator allows e-commerce companies to create high-quality, realistic product photos quickly without the need for traditional photoshoots. The tool uses AI to generate custom lifestyle and product images, tailored to various scenarios. It helps brands save time and money by enabling them to create visually engaging images of their products in seconds. The tool is especially popular with Amazon sellers looking to enhance their listings with content - quickly. You can find more about the GenAI tool here: (https://www.ecomtent.ai/product-image)
As this technology develops further, not only should we expect a lot more deepfake content online; but also I imagine a lot of companies will be able to save costs on creating great images for their products. Going one step further, I imagine companies will be able to A/B test a lot more freely the images used for their products on their product pages and collection pages; as it will be easier to generate various alternatives for lots of products without the cost of both time and cash to generate alternative product images to test. Things that could be tested could include - what is the most optimal background for my image? Is a lifestyle image better than a white background?
UC4 - Commerce Conversational AI (Banner / Email / SMS)
The potentially consequential use-case I am personally most excited for is the potential of “Conversational AI” for E-commerce. Imagine combining all the relevant shopping e-commerce data:
customer’s engagement on site
customer’s engagement in-app
customer’s engagement in-store
customer’s engagement with marketing channels
customer’s transactional / refund history
shop’s product information
shop’s inventory data
shop’s in-depth product descriptions
products’ relationships with each other
logistical updates about order delivery
… combining all of the above with the power of the LLM models to create a chatbot that is aware of the customer’s current situation as well as deeply aware of all the product information on the site. In the below two screenshots you can see an example conversation between the chatbot (blue) and the customer (white). These screenshots are taken from this Conversational AI demo video.
It could be a big shift for consumer’s ability to interact with e-shops in deeper ways. It would allow customers to engage with the e-shops product catalogue via natural language prompts and queries. Essentially, shoppers could have a conversation with the e-shop in a similar way we can today have a conversation with chatGPT about how to best plan my itinerary for my 2 week trip. That could be disruptive for the industry. There are 3 most common current ways shoppers engage with products in e-commerce today - via browsing, filtering and searching using keywords
Browsing can be a fun activity, but in practice only a very small portion of traffic looks at more than 10 products in a session. If you have more than a 1000+ products in a catalogue, how likely is it the customer saw the products that best fit their needs from the e-shop’s catalogue during their session? Filtering and searching for products using keywords can be fast and effective, however it is most suitable when a visitor already knows what they are looking for. This is why at Amazon, 80-90% of purchases start via the search bar - meaning to some extent the customer already knows what they are looking for. This isn’t to say that these 3 methods are going away - I believe they are here to stay → instead I think this will open a 4th method.
Conversational Commerce opens the possibility for new type of e-commerce shopping experience that is currently impossible. It would allow user’s who don’t quite know what they are looking for, just have a general idea - to much more quickly locate what they need. It would allow user’s to ask questions similarly to how one could ask a shop assistant in a store. Consider a shopper that is looking for a book online. Most book purchases online aren’t of the “browse & discover” kind. Those are much more common in physical bookshops.
Imagine then, that the customer tells the chatbot who are the last 3 authors they really enjoyed reading and ask if they could recommend a few titles. Or it could ask the chatbot if it could generate the best selling title for “their segment of customers” - as opposed to just the overall best sellers list. Maybe it could ask the chatbot for a list of recommendations based on a specific product they have purchased last month. Or they could describe the reading preferences of their friend for whom they would like to buy a gift. The user would not have to know about the product before hand to find it, nor would they have to spend time on the site looking for it.
Today, it is estimated approx. 20% of all retail sales are taking place online and industry observers are predicting a slower pace of growth of e-commerce in the next few years, only reaching 22% by 2027. Innovations like a Conversational E-Shop AI could potentially claw another few percent away from offline stores to online shopping.
There are still many questions left to answer. At what stage of the shopping journey should a shopper be approached by the GenAI? What channel should be used for this purpose? SMS? A Banner? Could users ask to show them the most visited products on the site? Will users that engage in such a conversation on the website be ultimately be more likely to convert?
Bloomreach claims to be developing this feature and that there are already e-shop’s BETA testing and experimenting with this feature. It is still being developed. Unfortunately, there wasn’t an updated demo of this feature at the Summit - we will look forward to future announcements and events from Bloomreach in regards to this feature.
If you found this post valuable…
We hope you found this article valuable. If so, please consider subscribing (for free!) to receive updates on our latest publications.
Additionally, if you'd like to discuss any aspect of marketing operations in more detail, feel free to book a meeting with us using the link below or check out our website at datacop.services .