On May 15, 2024, the "AI Genesis Era - 2024 Jiazi Gravity X Technology Summit," hosted by Beijing Jiazi Light-Year Technology Services Co., Ltd. and co-organized by Zhongguancun Dongsheng Science City, was held at the Wanda Realm Hotel in Zhongguancun Dongsheng Science and Technology Park, Beijing. Dozens of professionals from the technology industry gathered to focus on cutting-edge topics in the current technology field and delve into the development trends and broad prospects of the technology industry in the AI genesis era.
During the afternoon session on the 15th, Zhang Audi, Vice General Manager of the Technology Center at EasyPoint, delivered a speech titled "Harnessing AI for Production Capacity, Leveraging Data for Decision-Making, and Building a New Ecosystem for Overseas Marketing."
Zhang Audi emphasized that in the process of overseas marketing, relying solely on AI content production is insufficient to achieve a significant improvement in overall effectiveness. To achieve high conversion rates in subsequent advertising placements, an intelligent decision-making system becomes crucial.
Below is an excerpt from Zhang Audi's speech at EasyPoint Technology Center, edited and compiled by "Jiazi Light-Year":
"Thank you very much to Jiazi Light-Year for this event. It is an honor to represent EasyPoint at this event. Today, I would like to briefly share with you the development and exploration of EasyPoint in the fields of overseas marketing and AI.
Firstly, EasyPoint is a marketing technology company that has pioneered the AIGC strategy in the field of overseas marketing. Through performance advertising services, brand advertising services, and account management services for top-tier media, we tailor comprehensive solutions for our corporate clients to help them achieve growth in their overseas ventures, whether it be user growth or monetization growth. In this process, we have already expanded globally, processing up to 160 billion ad requests daily and accumulating data from over 7 billion independent global devices. This vast amount of data provides us with robust support, enabling us to conduct more precise and effective performance marketing. We have established deep cooperative relationships with numerous well-known enterprises, including Alibaba, Lazada, as well as collaborating closely with overseas projects of 3C brands such as Transsion and Kudy Coffee.
Reflecting on the past year of service experience, we have identified some new trends. Traditional hotspots for overseas expansion such as North America and Europe continue to show strength, but with policy openness and improved infrastructure, Latin America and the Middle East are gradually emerging as new overseas markets. At the same time, some domestic industries going overseas, such as e-commerce, gaming, film and television, and short videos, have gained widespread promotion and recognition abroad, with increasing attention and acceptance.
Throughout the entire process of overseas marketing, we have noticed that clients commonly encounter some fundamental challenges. Firstly, there is the issue of localization, as mentioned by our guest just now. In addition to creating localized content, clients also need to consider how to effectively promote it locally. Secondly, data growth is a major challenge for clients in the overseas expansion process. The third challenge pertains to IT infrastructure. Additionally, compliance and payment channels are also important considerations in the overseas expansion process.
Specifically, in the initial stages of overseas advertising placements, to create a large amount of material for advertising, clients need to invest heavily in localized shooting, including using foreign models of different skin tones for studio photography and producing materials in multiple languages. However, the involvement of language, culture, and creative content in this process makes it extremely costly. After completing the material production, the next step is multi-channel advertising to reach users. If the business involves advertising monetization or similar activities, multiple monetization strategies across different channels need to be implemented. In this process, precise attribution analysis becomes particularly important. We need to know which ads perform well, which monetization channels are more effective, and how data performs in different countries and channels. To achieve this, we need to build an entire data growth model to better guide our marketing strategies.
Once our data model construction is refined, if the business is thriving, the next challenge we face is how to effectively reduce cloud resource costs. In the overseas expansion process, we typically adopt a multi-cloud strategy to meet the cloud resource needs of different regions and reduce costs. However, a multi-cloud environment brings new challenges, including multi-cloud scheduling, cost optimization, stability assurance, and scalability management.
Facing these challenges, we have taken into account the rapid development of AI large models in recent years, as well as the long-term accumulation of data middle platforms, industry experience, and strategic models. These resources have led us to consider whether we can enhance our production capacity and efficiency through AI. Specifically, we hope to make intelligent decision-making chains more precise and efficient by combining industry models, data models, and AI technology. To achieve this, we have proposed a clear direction: harnessing AI for production capacity and leveraging BI for decision-making. We plan to build a multi-cloud-based foundational layer, integrate various cloud resources, and construct a powerful computing platform.
After integrating a large number of excellent large models, our goal is not just to pursue quantity but to ensure that these models are closely integrated with our business scenarios and customer needs through meticulous testing, thereby building more precise business models. In this process, we rely not only on AI large models but also combine historical industry models, especially strategic models. By basing our results on the combination of strategic models and large models, we ensure efficient conversions, thereby producing models for AI content production, intelligent advertising management, AI budget allocation models, and AI estimation models. Overall, through AI content production, operation, and management, we aim to build a complete AI marketing ecosystem." Through this integrated application, we have successfully developed multiple AI-driven content production and marketing models. For instance, our AI content production model can quickly generate high-converting marketing script based on simple keywords input by users, combining large language models and historical high-quality marketing copies. We also support over 1000 different industry-specific voice tones and more than 140 character templates, reducing the entire process from keyword input to generating digital voiceovers and marketing materials from 12 hours to just 5 minutes.
Comparing with traditional content production methods at the application level, we have significantly increased efficiency and effectiveness by leveraging AI. Previously, tasks like finding overseas models for photo shoots, live streaming, manual copywriting, voiceovers, and extensive designer involvement were necessary. Now, with AI intervention, users only need to input simple keywords to quickly generate high-converting marketing scripts using large language models and quality historical marketing copies. Additionally, we offer over 1000 industry-specific voice tones and 140+ character templates. For example, for Lazada, we provided a comprehensive marketing solution that significantly improved the efficiency and results of their marketing activities through AI content production and management. This not only demonstrates the immense potential of AI in content production but also provides strong practical support for our future AI marketing full-chain construction.
Currently, we can generate a voiceover video in just 5 minutes. Thanks to the realism and effectiveness of our digital characters and AI models, we have increased click-through rates by approximately 20%.
In early 2023, we first experimented with AI models. This was not a proactive exploration but a result of customer demand driving innovation. Among our overseas clients, there are many vertical e-commerce businesses, such as wig e-commerce. Despite being a niche industry, the demand is significant. A challenge was presented to us by a client: could we quickly generate product display images for wigs using AI technology? To meet this demand, we developed an AI model solution utilizing Lora tuning technology from the widely used SD model at that time. Compared to traditional overseas model photo shoots costing up to 320 yuan per photo, our costs were reduced by about 70% using AI models. Whether it's showcasing glasses or accessories, AI models can present them perfectly. Ultimately, the client adopted our AI advertising creative materials, resulting in a 35% increase in CTR and over 45% reduction in single cost.
Starting from 2023, short dramas have seen explosive growth again. Many of our previous clients who originally advertised in the European and American markets now see the maturity of the short drama market and want our assistance in quickly expanding into the Southeast Asian market.
To meet this demand, we developed a comprehensive AI solution. Based on the client's original European and American version short dramas, we used AI technology to quickly swap faces for Southeast Asian characters and provided dubbing with voice tones tailored to small languages like Filipino and Indonesian. This initiative allows clients to quickly produce short drama content tailored to the Southeast Asian market, helping them rapidly capture emerging markets.
During this process, we also discovered new demands from e-commerce clients. They wanted to use AI technology to guide customers to understand and purchase their products or visit their websites. To meet this demand, we combined our internal AI knowledge base and AI customer service technology, both with strong conversational capabilities. We also integrated our existing digital human technology to create a comprehensive e-commerce shopping guide solution for e-commerce clients.
Compared to traditional human shopping guides, intelligent shopping guides indeed have significant advantages. Firstly, intelligent shopping guides have the ability to work 24/7, meaning they can provide services anytime, anywhere, whenever needed. Secondly, intelligent shopping guides support online consultations in over 300 character images and more than 140 languages. This is due to our powerful AI knowledge base and AI question-answering capabilities, enabling intelligent shopping guides to provide diverse and personalized services to meet different customer needs. Furthermore, intelligent shopping guides can provide detailed explanations of individual products and introduce multiple product categories on the homepage. This flexibility allows intelligent shopping guides to comprehensively showcase products, help customers better understand products, and increase purchase intent. Lastly, to help clients quickly build an overall strategy for overseas e-commerce, we have integrated functions from eight major e-commerce platforms. This allows clients to quickly implement intelligent shopping guide enhancements on these mainstream e-commerce platforms, greatly improving the efficiency and convenience of e-commerce operations.
However, to significantly improve overall effectiveness in overseas marketing, relying solely on AI content production is not enough, as it may only account for 30% to 40% of the total. To achieve high conversion rates in subsequent advertising placements, an intelligent decision-making system is crucial.
Based on our existing data middle platform, although traditional BI systems can perform ETL data cleaning, data display, and basic self-service analysis, without industry models, they struggle to provide complete guidance for specific business scenarios. To overcome this limitation, we combined industry experience and a large amount of vertical data, along with large models, to develop four growth models. Firstly, we have a mixed monetization model; secondly, we provide self-service analysis capabilities; thirdly, we introduce an AI budget allocation model; and finally, we have an AI estimation model. In practical applications, these models play a crucial role from the start of advertising placements. For instance, when allocating budgets to different accounts and plans during ad placements, traditional methods often rely on manual calculations and thinking. However, our AI budget allocation model can autonomously learn and intelligently allocate budgets, significantly improving advertising efficiency and effectiveness. Last year, we developed an AI budget allocation system that has been adopted by our major e-commerce clients expanding overseas. By setting initial KPIs and budget inputs, this system can automatically suggest fund allocations for each account to ensure the achievement of set KPI goals. This process, which used to take 40 minutes initially, has been reduced to just 5 minutes, significantly enhancing work efficiency and leveling up all users' capabilities to a relatively average level of around 85 points. This means that whether the original capability was 90 points or 60 points, using this system can lead to more stable and efficient performance.
Upon the conclusion of campaigns, to increase monetization and user acquisition, we require a model to integrate data on user acquisition and monetization attribution behaviors, as well as four-party data. This is where our hybrid monetization data model comes into play. Through this model, you can clearly understand the amount invested today, the traffic gained, and the number of users brought in by this traffic. Furthermore, it can instantly generate all decay data for these users in the subsequent 0-90 days.
If anomalies are detected during the analysis, we can utilize the system for in-depth drilling down to track user behavior trajectories at the device level, which is our self-service analysis feature. Through this function, we can gain a more accurate understanding of user behaviors and needs. Furthermore, if we wish to forecast data performance in advance, such as when ROI will turn positive, our AI estimation model can be of assistance. This model, based on a vast amount of historical behavioral data for algorithm learning, can predict key indicators such as LTV, UP, and ROI within an estimated range of 1 to 90 days in the future. The accuracy of these estimated data is very high, with differences from actual data controlled within 5%.
Of course, we have also explored combining natural language with data platforms at the interaction level. Initially, one might think of using natural language to generate SQL, but in this process, we adopted a different approach. We constructed a large number of business logic vernacular and business gap description combinations, forming templates for cue words, and input these templates into our AI large model. Through extensive cue word engineering, we successfully integrated natural language tightly with data platforms, providing users with a more convenient and intelligent data query and analysis experience.
Ultimately, our system can transform detailed data requirements in complex business scenarios into tangible outputs. For example, it can calculate complex metrics such as ECPM values for users within specific registration time ranges based on different dimensions of active days. However, in facing the challenge of business complexity, we realized that mere cue word engineering may not always be the most effective solution. When business becomes complex and diverse, maintaining a large number of cue words can become cumbersome and inefficient.
Therefore, we began to contemplate a new approach. Since we already have four different data decision platform models, can we directly call these models through a large language model? Based on this idea, we recently launched an intelligent assistant. This assistant uses a large language model to construct API parameters, thereby understanding user intent and requirements, and calling the aforementioned four models. In this way, we can quickly and accurately respond to users' data analysis needs and generate visual data charts.
As our business gradually expands, in order to reduce costs in a multi-cloud environment, we have been researching intelligent multi-cloud scheduling and management strategies. Recently, we have released an intelligent multi-cloud management platform, combined with the intelligent assistant, providing customers with a full-chain overseas solution from basic cloud services to integrated marketing.
Through this full-chain overseas solution, we aim to help customers build an overall growth model at the lowest cost, whether starting from scratch or already having a certain scale of enterprise. We always adhere to the concept of technology driving marketing growth, committed to achieving more efficient marketing results through technology.
To promote the application and innovation of artificial intelligence in the marketing field, we have jointly established the Artificial Intelligence Innovation Application Joint Laboratory with partners such as the Western Branch of China Academy of Information and Communications Technology, and universities like Jiaotong University and Xiamen University. Through close collaboration with our partners, we hope to jointly build a high-quality, healthy marketing ecosystem, providing customers with better services and solutions.
Thank you all for listening!
