Perspective is a leading developer of e-commerce solutions on the Magento platform and is actively implementing innovations in its projects. This year, the company took part in iForum, the major IT event of Ukraine and Eastern Europe, where CEO Artur Potulny presented the use of ChatGPT artificial intelligence in e-commerce. It’s worth noting that the artificial intelligence landscape is currently undergoing a remarkable surge, characterized by the constant emergence of new products, versions, and innovative solutions almost every week. The information presented in this article pertains to the state of affairs as of August 2023. However, all the case studies discussed here remain highly relevant in today’s dynamic AI-driven environment.
Artificial intelligence market
At present, many diverse AI products span various domains of human activity. Some AI systems were designed to assist individuals in their professional endeavors, while others can completely supplant specific vocations, such as content management or call center operations.
Artificial intelligence has reached a point where it can proficiently draft written content, generate visual imagery, discern and synthesize spoken language, and even compose musical pieces. Notably, there exist AI systems geared towards hacking or fortifying websites.
A noteworthy development is the emergence of artificial intelligence collaborating with other AI systems. Some AI platforms dissect intricate tasks into more manageable components, while others adeptly resolve them.
While a plethora of AI products inundate the market, this article will delve into the capabilities of ChatGPT, developed by OpenAI, within the realm of e-commerce.
OpenAI product overview
Introduction of GPT, Dall-E, and Whisper Models
GPT (Generative Pre-trained Transformer) is a neural network model rooted in transformer technology. Having undergone extensive pre-training on vast arrays of text data, GPT can craft text that closely mimics human writing and excel in natural language processing tasks.
Dall-E, on the other hand, is a neural network designed for image generation. Its remarkable capacity enables the creation of lifelike illustrations, drawings, or photographs based on text input, showcasing an exceptional level of creative imagination. Some of the images generated by Dall-E are indistinguishable from authentic photographs.
Meanwhile, Whisper is an AI-driven system specialized in speech recognition. Whisper facilitates the seamless conversion of audio recordings into text transcripts, a technology that has found application in various use cases such as for legal transcription companies.
The OpenAI models showcase remarkable capabilities and hold immense potential for addressing a wide range of business challenges, as demonstrated through the forthcoming integration examples.
How GPT works
GPT operates as follows:
The model processes the input text word by word, converting each word into a numerical vector using an embedding layer.
These vectors are sequentially passed through a multi-layer transformer architecture. Each layer examines word relationships and identifies semantic features within sentences.
Based on this analysis, the model predicts the probable following words and generates output text closely resembling human language.
Throughout training, GPT fine-tunes its internal parameters to minimize the error in predicting the subsequent word in a sentence.
Consequently, GPT is a robust language model capable of analyzing and generating text with high semantic and linguistic precision. Its language comprehension approaches human-level proficiency.
Advantages and disadvantages of different GPT versions – 3.5 Turbo, 4, and 4 Turbo
Several iterations of GPT are available, including GPT-3.5 Turbo, GPT-4, and the latest GPT-4 Turbo. Each subsequent version exhibits increasingly “human” behavior and a deeper comprehension of language and context. However, the newer models are slower due to their enhanced complexity.
Various versions of the GPT model offer distinct advantages and disadvantages in practical applications:
GPT-3.5 Turbo is well-suited for handling smaller amounts of text, such as brief responses or descriptions. Its streamlined architecture enables quick text generation. However, its responses may sometimes need more depth, as it has limitations in grasping context.
GPT-4 excels when dealing with larger volumes of text, showcasing a more coherent connection between sentences. Nevertheless, its overall performance is similar to that of GPT-3.5 Turbo.
GPT-4 Turbo boasts the most profound understanding of language and logic, allowing for the generation of high-quality text. However, due to its intricate architecture, the GPT-4 Turbo operates slower than its predecessors.
Hence, the optimal choice of the GPT version depends on your specific objectives. For swift processing of concise text, GPT-3.5 Turbo proves effective, while GPT-4 Turbo is better suited for in-depth analysis of extensive datasets. If text quality holds utmost importance, GPT-4 takes the lead.
Prerequisites and Client Approach
Market Challenges: Relocalization and Stagnation
In the contemporary market landscape, Ukrainian companies are grappling with two significant challenges, compelling them to seek innovative solutions like integrating ChatGPT.
The first challenge revolves around relocation, wherein companies venture into new markets. Ukrainian and Western enterprises are actively localizing their projects to regions such as Poland, the Baltic States, and beyond. Consequently, companies are confronted with an urgent demand for substantial content generation.
A pivotal factor driving this demand is the need to craft localized website versions in diverse languages while entering new markets. This entails the creation of bespoke content tailored to each region’s specific audience and cultural nuances.
Another critical objective entails translating and adapting a company’s existing content to align with the unique characteristics of new markets. This encompasses the translation of websites, technical documentation, user manuals, and more. Furthermore, continual content refreshment across all company websites is imperative to maintain audience engagement and relevance. This entails regularly creating fresh articles, news updates, and reviews.
The second problem is the stagnation of traditional markets and falling sales. Companies must look for new ways to attract customers, including personalization, to boost sales. After all, customers today expect an individualized approach and consistency.
According to recent research, 80% of consumers want to see personalized content, recommendations, and marketing messages from brands. However, only 20% of companies can provide this level of personalization.
To provide relevant customer experiences, you need to analyze their data, behavior, interests, etc., and adapt content and offers accordingly. It is impossible to set up such personalization for each user manually.
Thus, relocalization and stagnation create an urgent need for new approaches to content and personalization, where technologies like ChatGPT can be beneficial.
We’ve also made intriguing observations regarding the responses of Ukrainian companies when we proposed integrating ChatGPT into their projects.
Firstly, it’s noteworthy that nearly all our clients were already familiar with this technology and had deliberated its potential applications within their teams. This underscores that Ukrainian businesses are staying abreast of global trends in artificial intelligence.
Secondly, most of these companies had preconceived notions about how they wished to employ ChatGPT in their projects. Some even brought forth innovative ideas that Perspective had not considered before. This underscores the high interest among businesses in the practical utilization of AI.
Consequently, the market is sufficiently prepared and receptive to adopting cutting-edge technologies. Ukrainian companies aren’t just aware of ChatGPT; they’re actively exploring ways to integrate it to solve business challenges. This stands as a highly positive indicator of the technology’s prospects.
Practical cases: Content
Case #1 – Writing product descriptions
One of the most prevalent applications of ChatGPT in the realm of e-commerce is the automated generation of SEO-optimized product descriptions.
Within the scope of Perspective cases, instances arose wherein suppliers furnished solely a collection of product specifications devoid of comprehensive descriptions. This scenario mainly applied to straightforward household items such as electric kettles.
Usually, copywriters had to painstakingly craft descriptions manually, drawing from these austere specifications, which proved inefficient. Consequently, the decision was made to leverage GPT-3 for the automated generation of these descriptions.
The approach involved submitting a package of technical specifications for the kettle, along with its accompanying photograph and name, to ChatGPT, requesting a concise product description.
The outcomes were quite promising: ChatGPT generated a reasonably compelling commercial description of the product within mere seconds based on the provided data. While subsequent refinement by a copywriter remains necessary, AI significantly streamlines this process, and it comes at a meager cost, amounting to mere fractions of a cent per request.
Case #2 – Reviews Translation
One of the most prevalent tasks when a company expands into new markets is the translation of substantial volumes of content into different languages. This enables the localization of websites and products for specific target audiences.
Perspective faced such a situation while working with clients. One of the stores received numerous reviews in Russian and Ukrainian languages. They needed translations from Ukrainian to Russian and vice versa to feature these reviews on their website.
Traditionally, content managers handled this task manually, which proved slow and expensive. Consequently, the decision was made to incorporate ChatGPT.
During the review moderation process, the website administrator promptly employs AI to translate the reviews into the desired language. This has significantly expedited the publication process and reduced costs.
Another advantage was ChatGPT’s ability to analyze the review content. When presented with hundreds of recent reviews, the AI swiftly identified the 5-7 most positive and negative points highlighted by customers within seconds.
This expedited identifying pressing issues, such as those related to delivery or call center operations.
Hence, the integration of ChatGPT simplified and substantially lowered the costs associated with managing customer feedback, encompassing both translation and content analysis aspects.
Case Study #3 – SEO Content Rewriting and Landing Page Optimization
Another common challenge when expanding into new markets is the need to rewrite existing content to avoid duplicate content.
For instance, when a brand extends its presence from one English-speaking market to another, substantial content rewriting is required to cater to the local audience, including optimizing search queries.
In the Aquapolis project, Perspective harnessed the capabilities of ChatGPT to automate the process of content rewriting with a unique twist—it works seamlessly with HTML markup.
We instruct ChatGPT with all the landing pages: “Keep the markup intact but rework the text.” It excels in rewriting text while preserving the HTML structure.
For effective SEO, it’s crucial to specify specific key phrases that should remain untouched by the AI during the rewriting process.
We’ve integrated two buttons into the admin panel to streamline this workflow: “Rewrite” and “Translate.” This empowers administrators to initiate content rewriting or translation through ChatGPT swiftly.
This approach has significantly expedited the process and lowered the costs of adapting existing English-language materials for new markets, all while maintaining text uniqueness. Following automatic rewriting, the content receives a final review by an expert.
Case Study #4 – Streamlining Image Handling with AI
In the realm of e-commerce, visual content carries substantial weight, making the automation of image processing a pertinent challenge.
One of Perspective’s customers encountered a quandary involving product photos being out of context. In the case of Rowenta, a limitation arose with the Dall-E system—it could add elements to an input image but not alter it. We devised a workaround by removing the background through a separate AI algorithm. Then, the manager could instruct Dall-E on what to do with the image using a query.
Here’s another example: a manufacturer supplied us with images of shoes against a plain white background without any models. To realistically convey the product’s appearance, we needed to “dress” it in a lifelike setting—the traditional approach involved manual labor, which was both time-consuming and costly. Hence, we ventured into resolving this quandary with the assistance of AI.
For instance, phrases like “Black men’s boots on the feet of a mannequin adorned in jeans against a white backdrop” or “Converse sneakers showcased on a girl wearing sweatpants.” Dall-E was tasked with generating the corresponding images.
This method entails a degree of trial and error since Dall-E’s initial attempts might yield perplexing results. Thus, one must cycle through the “Next image” button until a satisfactory outcome emerges, often requiring 5 to 10 iterations.
In essence, this approach enabled us to substantially enhance the efficiency of designers while expediting the creation of authentic product photos for the website. After auto-generation, the images underwent slight refinements by human touch.
Practical cases: Personalization
Traditional personalization methods in e-commerce are often superficial, as they are based on limited data about the user. Segmentation by socio-demographic characteristics needs to provide a deeper understanding of their context and needs. Instead, artificial intelligence can analyze much more data on human behavior in the dynamics to understand their current interests, intentions, and emotional state. This allows for personalization that is truly unique and relevant to a particular user and situation.
Case Study #5 – Enhanced Product Recommendations
In e-commerce, personalized product recommendations are a potent tool for boosting conversion rates and fostering customer loyalty. ChatGPT can play a pivotal role in simplifying the creation of such tailored recommendations.
Take, for instance, the GoodWine online wine store case, where Perspective implemented personalized recommendations by seamlessly integrating AI with visitor data gleaned from site interactions. Initially, ChatGPT was granted access to the entire wine inventory, replete with detailed descriptions.
Upon a user’s arrival at the website, ChatGPT taps into CRM and various analytical platforms to gather insights regarding the visitor’s past interactions, demographics, and preferences, such as their inclination toward pricey or budget-friendly wines or interest in wines from specific countries.
Leveraging this analysis, the AI engine handpicks the top 5 wines from the vast catalog, custom-tailored for the individual user. These personalized recommendations are prominently showcased on the homepage, yielding significantly higher conversion rates for the suggested products.
ChatGPT automates the personalization process, seamlessly incorporating each user’s unique context through AI-driven data analysis—and remarkably cost-effectively, to boot.
Case Study #6 – Chat with AI
Chatbots are becoming an increasingly popular channel of communication with customers in e-commerce. However, creating a chatbot that maintains an exciting conversation takes much work. This is where ChatGPT comes in handy.
One of Perspective’s clients, an online clothing store, wanted to improve its chatbot for customer support. Previously, it was a typical boring FAQ chatbot with limited capabilities.
To revitalize it, they integrated ChatGPT and trained it on the assortment and descriptions of the store’s products. Now, customers can communicate with the AI in a chat: ask questions, ask for clothing recommendations, and clarify details. It is worth noting that since the communication takes place in real time, where the data array cannot be downloaded in advance, the request’s processing time is quite long – up to 15 seconds, but the cost remains low.
ChatGPT analyzes users’ requests, context, and the history of previous messages and provides answers in natural language. The bot has become a more intelligent and exciting interlocutor for store visitors.
Thus, the integration of AI allows the chatbot to rise to a new level of customer service and meet the audience’s demand for personalized communication.
Case Study #7 – Search Optimization
Improving the relevance of search results is another way to increase conversions with AI. ChatGPT can “understand” the context of user queries and adapt the results output appropriately. In this case, artificial intelligence is not responsible for search, which is still performed by the corresponding search engine. The role of artificial intelligence is to sort the search results.
For example, one of Perspective’s customers, an online agricultural machinery store, wanted to optimize its search results. Previously, the results were generated only based on the relevance of the keywords to the query.
AI was integrated to improve their relevance. Now, when a user searches for a product, ChatGPT analyzes their previous search and purchase history on the site. For example, a user bought a lawn mower yesterday and is looking for a knife today. GPT understands that he probably needs a knife to match the lawn mower and adapts the search results accordingly.
This approach allows you to show the most suitable products to each user, increasing the likelihood of a purchase.
Case study #8 – Emails
Email marketing remains an effective promotional tool, so optimizing email campaigns is urgent. In particular, sending customers the most personalized emails is essential.
In Perspective practice, we came across the following case. The client, MammyClub, an online store for children, had a subscriber base and wanted to send personalized emails.
Previously, the marketer was preparing universal newsletters. However, after integrating ChatGPT, the process changed. Before each mailing, AI analyzes user data: child’s age, gender, and upbringing stage.
Based on this information, ChatGPT generates a personalized email for each subscriber, recommending exactly the relevant products to their child.
Such targeted emails have brought a significant increase in conversion and audience loyalty. AI allows automation of the process of creating highly personalized emails.
Practical cases: Communication
During long-term communication with many customers, people can experience emotional and psychological burnout. This leads to a decrease in the quality of service, an increase in the number of mistakes, and conflict situations. Instead, artificial intelligence has no emotional state, so that it can work steadily and efficiently without fatigue. It also makes fewer mistakes than humans because it can analyze more data. AI can quickly process routine requests at a low cost.
Case Study #9 – Enhancing Message Tone Automatically
In e-commerce customer communications, a need often arises to swiftly refine the tone of messages, rendering them friendlier or more professional. This process can be automated with the assistance of AI.
A prominent retailer collaborating with Perspective confronted the challenge of customer complaints regarding the inadequately friendly tone of support emails. For instance, formal denials for product returns frequently irked customers.
The operator’s admin panel introduced an “Enhance Tone” button to address this issue. Upon activation, it dispatches an email to ChatGPT with a request to rephrase the message more amiably.
Within seconds, AI crafts an alternative message that exudes a softer tone and refrains from eliciting negative emotions. The same approach can be applied to achieve an “Official Tone” or any other desired email style.
This swift, automated enhancement in communication tone has proven instrumental in bolstering customer loyalty and curbing the incidence of complaints. AI affords the flexibility to tailor the style of messages to your specific needs.
Case study #10 – Auto-responses in support
High-quality customer support is the key to success in e-commerce. However, it requires a lot of resources and effort. Part of the routine work in tickets can be optimized with ChatGPT.
One of Perspective’s clients, a prominent online electronics store, needed help with the workload of its support team. Many similar and simple questions came from users.
They transferred some of the tickets to AI for processing to optimize this process. When a standard user request comes in, an employee can forward the request to ChatGPT.
AI analyzes the question and available materials from the company’s knowledge base and generates a comprehensive answer for the user. This has significantly relieved employees and accelerated the processing of typical cases.
Thus, AI helps optimize customer support and improve CX by automating routine requests and support ticket analysis. The staff can focus on non-standard cases.
Conclusions
Technical conclusions
Analyzing the experience of using ChatGPT in practice, the following technical conclusions can be drawn:
- In most cases, it is optimal to use older GPT models, such as GPT-3,5, rather than the latest GPT-4. They work faster and are much cheaper, and their functionality is enough to solve problems in e-commerce.
- Formulating requests to ChatGPT in English is more profitable, as the cost of process requests in local languages is about 1.5 to 3 times higher.
- Usually, there is no need to train your AI models from scratch, as existing models are already smart enough for most tasks.
- It is worth limiting ChatGPT to stop phrases to avoid excessive talkativeness and save money, as it pays for the number of characters in the answer.
- Sometimes, combining several different AI models to solve complex tasks, such as text and image processing, is beneficial.
- The main technical problem is still the speed of AI, so it needs to be optimized.
To implement ChatGPT effectively, consider these technical nuances.
Business conclusions
From a business standpoint, the implementation of ChatGPT yields the following key conclusions:
Artificial intelligence already demonstrates a relatively deep understanding of subject areas and business processes. Its analytical abilities often impress and exceed expectations.
The cost of implementing AI solutions in e-commerce is quite affordable. They pay off by automating routine tasks and increasing conversion rates.
However, AI should only be viewed as a panacea that will solve some problems. It requires a flexible approach and an understanding of the technology’s limitations.
Human involvement and control over AI remain crucial. Fully autonomous operation often leads to mistakes.
AI must be combined with the company’s existing customer data and behavior data to get the most out of it.
Thus, artificial intelligence opens up new business opportunities, but it needs to be appropriately managed and combined with human intelligence.
Ukrainian Market Prospects
Despite a full-scale war, Ukraine has showcased its remarkable capacity for innovation and rapid adaptation to new technologies. Our IT companies have pioneered numerous digital solutions to support the military and society.
The time has come to harness this potential for broader purposes, particularly in artificial intelligence. Ukraine has a historic window of opportunity to position itself as a global leader in implementing and utilizing AI.
Our IT enterprises have already exhibited keen interest and readiness to experiment with ChatGPT and similar technologies. There is a genuine opportunity to catch up and surpass developed nations in the practical application of artificial intelligence across businesses and society.
The widespread adoption of AI has the potential to serve as a robust catalyst for Ukraine’s economic expansion and the advancement of high-tech industries. Therefore, we should boldly venture into experimentation, innovation implementation, and cultivating unique expertise in applying artificial intelligence within our nation. This is how Ukraine can ascend to a global leadership position in the AI industry.
About the authors
Article by Artur Potulnyy, CEO Perspective.
Has this article sparked ideas on how AI could optimize your e-commerce operations? Allow Perspective’s 14 years of experience in Magento development to work for you! Contact us to arrange a consultation on integrating innovations like ChatGPT into your platform. Drop us a message, and we’ll explore ways to elevate your business to the next level!
Stay informed about retail and e-commerce news and insights through our Facebook page and Telegram channel and by subscribing to our weekly email newsletter.
***
