While many are asking what sort of threat ChatGPT and DALL·E pose for society, marketers and entrepreneurs are looking for ways to make productive use of the powerful AI tools. The opportunities appear to be greatest in areas where resources are low. Frank Puscher, who led a talk on the subject of AI at this year’s Shopware Community Day, provides an overview.
Table of contents:
“When a chatbot writes your homework,” read the title of a Tagesschau news segment. “Images can no longer be trusted,” wrote the Catholic newspaper Tagespost, when faked photos of Donald Trump being taken away by police, long before his trial began, were seen around the globe.
Yes, there’s no denying it – generative AI is hugely powerful. “The technology has become so good that we are now beginning to test it for commercial use,” Amy Gershkoff Bolles asserts, explaining that customers are no longer able to distinguish real images from fake ones. But Gershkoff Bolles is on the other side of the AI debate – very much pro the opportunities and possibilities that generative AI offers. Amy Gershkoff Bolles is the chief innovator at Levi Strauss. And she sees the fact that computer-generated images of people in jeans can’t be distinguished from pictures from a photo shoot in Barbados as a good thing.
The big AI debate
Opposing views of AI as a good or bad thing generally come down to the impact of the technology. Where it is used to induce people to act in a way they did not originally intend, in other words where they are manipulated using false information, it has a negative impact and needs to be regulated and controlled.
When it simulates or reproduces something that tallies with the facts – or when it is applied in areas such as art, entertainment, or fiction, where facts are not a priority – its use can be considered legitimate. In fact, you can safely say that generative AI can genuinely do good here, particularly when the party employing it doesn’t have the resources to do what they want to or ought.
Take the example of inclusivity. Levi Strauss has a long history of working with a diverse spectrum of models. “Using models that reflect different body types, ages, sizes, and skin tones is in our DNA,” asserts Gershkoff Bolles. She recently announced a new partnership with Lalaland, a company from Amsterdam which creates hyper-realistic avatars that can be “dressed” in Levi products. The avatars can be made to reflect any body type, age, size, and skin tone, completing gaps in the jeans manufacturer’s current spectrum of models. Generative AI can improve inclusivity.
Of course, inclusivity here is not an end in itself. “We know our customers want to shop with models who look like them,” Gershkoff Bolles explains. In other words: The more the product depiction and description resonate with the individual customer, the higher the conversion rate. This tautology has plagued the world of marketers and online merchants for as long as mail-order shopping has existed. Of course, every form of personalization holds the potential to sell more products. But up until now, businesses have not been able to afford to create an individual product image for every customer. Up until now.
Now, all of a sudden, we have a handful of systems that can generate presentable images from text prompts. The current leaders of the pack are DALL·E, Stable Diffusion, and MidJourney. According to a survey by the agency Syzygy, around a quarter of all Germans have already experimented with these tools. And if tools like this can generate images from text, they can do the same from database entries – which are, after all, text too.
Online merchants will certainly be sitting up and taking notice at this point, as that solves two problems at once. Firstly, you can never have too many visuals – see Levis. With other forms of product presentation, you can always adapt your customer communications – through personalization, for example, or references to events such as Easter or the summer vacation.
And secondly, merchants like to stand out from the crowd. All merchants receive the same images from the manufacturers, but they can use AI tools to cater more effectively to different target groups. Unique content is a key asset when it comes to the customer – and to Google too, of course.
Generative AI in ecommerce
For a clear overview of what generative AI tools can do for retail and ecommerce, it’s good to separate out the following areas of use:
Synthetic product descriptions
Personalization
Reporting
SEO
Chatbots
Synthetic product descriptions
Generative AI can generate new content in whichever form of media you choose, whether text, images, videos, podcasts, or music. The systems can be used to speed up the production of each of these forms of media, lower costs, make processes more efficient, or even introduce automation.
It’s a well-known fact that multimedia content has the potential to sell more. But many businesses are put off by the expense and are afraid to experiment. Generative AI makes it much easier. Photo: Vaimo.
Levi Strauss is not only planning to showcase its products on different virtual models but also in different virtual scenarios. If, for example, the brand wants to show the jeans model in front of Cologne Cathedral or at the Port of Hamburg, that’s a classic job for Photoshop. The model is cropped out of the original picture and placed on a new background, then the light and shadows are altered to make the image appear “real”. These are precisely the steps that AI will now take care of. And if you spend less time producing images, you can either create a bigger range of images, enhance the quality of the individual images, or use the time for something else.
On the subject of Photoshop: Adobe has just introduced a toolset known as Firefly, which will soon allow this kind of automation directly in Photoshop, Illustrator, or Premiere (video).
Still in beta: Adobe Firefly
What goes for images goes for text too. From just a handful of product details, ChatGPT can generate an impressive product description. With tools like Shopify’s new Autowrite feature, you can even choose whether you want the text to be playful or more sophisticated. This too is something that will delight search engine experts: Custom product descriptions are good for rankings.
With Shopify’s text generator, you can set the tone of voice for your product descriptions.
Coca-Cola recently announced a cooperation with ChatGPT creator OpenAI. The brand is looking to produce “hyperlocal” commercials that specifically reference a place, the language spoken there, local specifics, and so on – an incredible scaling of advertising production that is only possible as a result of automation.
But it doesn’t stop there. Tools such as Synthesia and Maverick can generate whole videos from scripts. With Maverick, you can also integrate data variables so that the videos that are generated automatically include the first name of the email recipient, for example.
With Curie Vision, you can turn 2D images of products into 3D assets in a fully automated process. These 3D assets can then be adapted and used, for example, for your virtual try-on. Or – and this was Curie’s plan from the start – the products can be integrated into 3D environments such as Fortnite or augmented reality applications.
This is where all the hype of recent years suddenly comes together. Consultants across the country have been recommending online merchants to set up systems like this for some time, because research has clearly demonstrated the added value of interactive product presentation. But what was always missing from the calculation was the expense of creating systems like this and, as such, the ROI (return on investment) factor. Generative AI significantly reduces that expense.
Personalization
From here it is a small step to the subject of personalization. AI systems have long been used for recommendations. Now the next stage has been launched. Generative AI can do more than calculate probabilities. It can also communicate everything in attractive visuals and text. According to a McKinsey report, 71 percent of consumers expect companies to deliver personalized interactions.
Brooklyn Bikes has set up a fun intro quiz on its website. It keeps visitors on the site for longer and fills the CRM database with plenty of solid data for effective personalization. Photo: Screenshot.
Brooklyn Bikes, a cycle manufacturer, came up with the idea of welcoming customers to their website with a quiz to find out their preferences where bikes are concerned. This will be familiar to anyone who’s ever installed Spotify. These data are fed into a CRM system and are used to control all communication with the customer. Those who have expressed an interest, for example, will automatically be sent an invitation whenever there is a local cycle show or doors open day so that they can get a hands-on feel for the bikes.
Imteaz Ahamed, Director, Performance Marketing at Reckitt, the company that manufactures products such as Calgon, Clearasil, and Sagrotan, puts it succinctly: “We’ve become very focused on managing the exceptions rather than simply managing the rules all the time.”
Reporting
“We’re just starting out on our journey and are still gathering insight,” states Amy Gershkoff Bolles of Levi Strauss. And perhaps this is the biggest added value that generative AI can contribute, at least where text is concerned. Large language models such as GPT understand language well and can “speak” well. They are also ridiculously fast. They can trawl through masses of text and summarize it in a way that is easy to understand.
Patrick Bunk is doing just that at Unicepta. The company analyzes everything that is written in the public domain and published online in order to provide intelligence for global corporations. It trawls through three million websites every day, for one customer alone, looking for new information. And uncovering 200,000 new items of text on a daily basis. With the aid of intelligent AI, Bunk then generates a 15-sentence report so that the CEO or Head of Marketing knows what to focus on.
This can work in a similar way but on a much smaller scale for online merchants or brands. They can analyze chat histories, for example, or customer emails, and try to find out what is going well and what isn’t. This information can then be fed back to where it can be acted on, whether that’s in support or even in product development.
The result, if done properly, is unified commerce. Cycle retailer tokyobike has even incorporated its in-store POS, giving the marketer a full overview of customer engagement.
SEO
Everything that I’ve said about synthetic product descriptions applies to SEO too. Except that the approach here is much broader. Businesses can use tools such as ChatGPT or Bard to generate much more content in a short time. And this means that they can finally get around to tasks that should have been done a long time ago, such as setting up a FAQ section (which, incidentally, can be compiled by running an automated analysis of customer communications). But you can also use these tools to react quickly and cost-effectively to new developments or to risk the odd experiment.
Does a website work better if you offer individual user groups extremely local content? At Reckitt they believe so. "How do I get to every single cluster that I want to go after? If AI can do foreign languages, it can do dialects too,” explains Imteaz Ahamed. Advertising in dialect could be worth a try.
Needless to say, using generative AI can boost the number of blog posts, explanatory videos, or help tutorials you are able to produce. And that’s good for your Google visibility. Although SEOs recommend that you always slightly tweak the generated text before using it.
Chatbots
The last item on this list, but by no means the last area of use for generative AI, is the chatbot. Thanks to GPT etc., chatbots finally understand what people are saying to them. And are able to answer in clear, legible terms. That’s not only helpful for regular users but can also help those with a disability such as impaired vision. Well-designed chatbots can make websites more accessible.
Accessibility can also be about locating content that is deeply hidden in the website. Blogs in the USA are already warning about the risk of navigation fatigue. People don’t want to click through search results anymore. So the chatbot is taking on the role of universal navigation aid.
The new chatbots understand customers better and communicate content more clearly. In an ideal scenario, they can help directly with solutions. Photo: Screenshot.
But for Stefan Trockel, that’s not enough. The founder of Mercury.ai wants chatbots to become a control hub for business functions rather than just information systems. “Why does the bot send me a link to the page with the returns label instead of just letting me fill it out in the chat window?”
And what about tracking? Mail-order business Erwin Müller receives 100,000 inquiries a year about the status of orders. “The team has virtually nothing to do there,” comments Christian Wirth, head of the Communication Center. Thanks to automation.
Juniper Research has projected that USD 161 billion will be spent on chatbots in 2024. As opposed to just under 3 billion in 2019. The explanation: Chatbots are a business’s 24/7 service center and, as such, have come to be seen as a company’s business card. 90 percent of standard inquiries are handled using automation, and only the most complex 10 percent are forwarded to a human support team.
The company Forethought has developed a system known as Triage. This scans customers’ incoming messages, prioritizes the support tickets, and assigns them to the right contact. The business was inspired by its own customer Lime. For the e-scooter and e-bike sharing business, it matters whether a customer is getting in touch because they have just had an accident or because they want to know what prices you charge. Lime now deals with a quarter of all inquiries via the chatbot. According to Lime’s own figures, this, in conjunction with prioritization, has reduced response times by a whopping 77 percent.
At Lime, automation is now used to handle a quarter of all customer inquiries. Photo: Lime.
AI is everywhere
There is no denying that generative AI can dramatically speed up and improve existing processes. It is clear that it will also lead to the creation of new processes, business models, and job titles. For example, it has become even easier for businesses to build an international digital presence, even if they are effectively a two-person outfit.
One new job title is prompt engineer. “Prompt engineering is the process of communicating effectively with an AI to achieve desired results,” explains Goda Juskeviciute. The Latvian is currently learning the profession and helping others at learnprompting.org to distinguish good tutorials from bad. “A lot of tutorials and manuals are written by ChatGPT rather than by humans. And the authors haven’t even tested them.”
This too is part of the brave new world of generative AI.
Even more exciting tools: Top 10 generative AI tools for ecommerce enhancing customer experience with AI-generated videos and more
Watch the AI expert talk from Shopware Community Day 2023 in full
Want to learn more about AI? Watch the SCD expert talk with Frank Puscher, Manon Dave (Mindvalley), Brian Lange (Future Commerce), Stefan Hamann (Shopware), and Mark Stanley (Shopware). You’ll find all the recorded talks and presentations from Shopware Community Day 2023 on our video platform.
