2022/12/08

⏱ 5mins 

What is important for the growth of the e-commerce and fashion industries?

In the past, when PC-based commerce was converted to mobile, product content such as product pages and detailed descriptions within the app were very important. After this period, as mobile commerce transactions became more active and numerous shopping apps that could satisfy consumers' tastes emerged, upgrading the personalized recommendation technology of the app itself became a necessity rather than an option .

IT IS BECAUSE THE AI RECOMMENDATION SYSTEM CAN HAVE A GREAT EFFECT ON INCREASING SALES BY INCREASING THE PURCHASE CONVERSION RATE AS WELL AS LOYAL CUSTOMER PICTORIAL BY DISPLAYING PRODUCTS THAT ARE PERFECT FOR CUSTOMERS' TASTES.

Brandi, a vertical commerce company, introduced Upstage’s recommended AI Pack and achieved a 60% increase in purchase conversion per total impression, and an average increase of up to 150% compared to the previous one based on the Upstage AI model alone . We met Choi Won-jo, head of Brandi's Data Optimization Office, to hear more about the role of Upstage's recommended AI pack.

The driving force behind the growth of Brandi, a vertical commerce company

1. Hello, please introduce yourself.

Hello, I am Choi Won -jo, who is in charge of the Brandi Data Optimization Office. I am mainly in charge of the AI optimization business that matches users who are likely to purchase and products in real time to increase the purchase rate.

AI is used to match these users and products . Real-time personalized product recommendation based on user's propensity, product category, and other attributes. For additional purposes, we are trying to maximize the interactive platform experience by increasing advertising efficiency by preventing meaningless advertising expenditure through personalized product recommendation, and by exposing only meaningful advertising to buyers.

2. What kind of company is Brandy?

Brandi is an e-commerce company that operates a customer-specific vertical commerce platform and global business. As a B2C service, we operate Brandi, a fashion shopping app, Hiver, a men's shopping app, and Seoul Store, a brand fashion platform .

In particular, Brandi is a fashion shopping app for women in their 10s and 20s. It introduced one-day delivery, the first fast delivery service in the industry , to meet the needs of customers who want to receive fashion products as soon as possible. It is also a service that is loved by customers enough to achieve a 500% growth in daily delivery cumulative transaction volume over the past three years since its launch.

3. In the face of intensifying competition in the fashion application market, what is the driving force behind the growth of fashion shopping app Brandy ?

The reason why the one-day delivery service has been able to achieve such great growth is mainly due to Brandi's logistics infrastructure and technological prowess that it has accumulated so far .

As Brandi developed and operated its own FMS (Fulfillment Management System), an integrated fulfillment management system, it was able to have its own system that could quickly and accurately handle everything from order collection to delivery.

From the second half of 2021, we are strengthening our capabilities as a data platform, starting with the AI optimization business, for a new driving force. The AI optimization business also has the effect of increasing the app's purchase conversion rate by improving users' convenience, so it is also giving positive synergy from the seller's point of view. In particular, Brandi's advertising service is highly efficient and enables direct communication with the development team, so we are getting a lot of positive feedback from sellers. In the future, in addition to Brandi, we are trying to expand it to Hiver and Seoul Store.

AI RECOMMENDATION SYSTEM THAT COMMERCE AND FASHION MARKETS SHOULD PAY ATTENTION TO

4. What kind of work are you currently doing with Upstage?

Our AI optimization business is a key partner with Upstage.

In addition to personalized recommendations, we are also involved in recommending personalized advertisements, other platforms, and various publications . In particular, the PoC started and the results were very good right away from last August, so we started to apply Upstage's technology to the second platform, Hiver's personalized recommendation . We are discussing expanding the application of personalized recommendations to other areas and various brandi functions in the future.

5. How influential is recommendation AI and data in Brandy ? Why is data important in fashion applications ?

As of 2022, recommendation AI and conversion rates are the top focus at Brandi. This is because today's startups are in a market where efficiency is important. Looking back at the past, until last year, e-commerce platforms were an industry that aimed for transaction volume, that is, market share. However, now is the time for the platform to focus on more intrinsic strength due to the high interest rate market.

In particular, rather than a direct method of raising the total amount of variable costs that can be invested in order to increase the transaction amount, the key to the current meta is how much more transaction amount and profit can be generated even with the same variable cost .

All of this comes from your conversion rate . If you buy something you won't buy once, buy it once or buy it twice, that's a good platform. Various marketing, CRM, coupon play, efficient events, and copywriting written by God are important, but the biggest part of the pie is whether the product you want to buy appears on the basic screen visible when you turn on the app . Data exists to analyze whether we are doing well in these areas. We plan to increase the purchase conversion rate by collaborating more with Upstage and launching various high-performance AIs through various analyses.

The secret to increasing purchase conversion amount, Upstage's recommendation AI pack

6. What were your expectations for Brandy before introducing Upstage's recommended AI Pack? I wonder what made you decide to introduce Upstage's Recommended AI Pack.

When solving a problem with AI, the first thing that usually comes to mind is to develop it in-house through modelers and AI experts . Of course, there is no dispute that the self-developed model has the most customizable fit.

However, since the resources and research time required to build an AI development team are considerable, this leads to excessive expenditure . Therefore, the place where such self-development can proceed is either a company that focuses on AI products like Upstage, or a large company that has entered the so-called AI war.

However, in well-known and standardized areas such as personalized recommendation , the question itself about what you want to do rather than the performance of the model algorithm and modeling the problem in the form of data through the characteristics of available data determine most of the performance.

We are clear about what we want to do, and because indicators and data are ready, we can leave the actual AI model development to a professional company. This is why we introduced the Upstage Recommendation AI Pack.

We only have one AI that chooses an AI model based on traffic and context. Many AI recommendation companies as well as Upstage participated in the development of our AI model, but Upstage is currently the number one runner in terms of performance compared to other companies .

This will create the effect of promoting the company's technology and performance to Upstage and other participating companies, acquiring a portfolio, and creating good models in other areas by utilizing the experience of Brandi. I think it can create a win- win effect that can have an AI model. Anyone can sell an AI model to Brandi in this virtuous circle and challenge it.

7. Upstage's recommended AI Pack was specifically applied to Brandy, and what effect did it have?

Currently, Upstage is appearing in the form of personalized AI recommendations on the brandi home page . These days, I have occasional opportunities to meet brandi users, and I've heard that products I want to buy are frequently uploaded recently.

Here are the questions we initially posed to Upstage: “Maximize multi-touch attribution for conversion amount per impression” , which exactly means maximizing conversion rate per impression, i.e. conversion amount. Multi-touch attribution is a way to accurately distribute the contributions of recommendations made by other AIs or by our internal solutions. This allows us to aggregate and compare the performance of all other AI models.

After applying Upstage's recommended AI Pack, the amount of purchase conversion per total impression increased by nearly 60% compared to before starting the business. That's almost double the +32% I set for when I started my business earlier this year. Looking at the upstage AI model alone, the average increase was up to +150% compared to the previous one.

As you can see, the increase in upstage is very large, right? Currently, it is applied only to the home screen and some pages, but in the future, it will be expanded to search and category pages, so the total average purchase conversion amount per impression will be further increased. We are currently uploading upstage recommended AI packs to our other services as well.

Also, other recommendation AI companies are taking on Brandi's comparison challenge. If anyone comes and shows off performance and gives superior performance, we can distribute that amount of space and raise the amount of use. I think this test method could work like ILSVRC (ImageNet Large Scale Visual Recognition Challenge) in recommender systems. In fact, before ILSVRC started 10 years ago, AI surpassed human image discrimination ability. In the commercial part, such as recommendations, live data from real companies and user evaluations are essential for performance. In this way, I believe that the so-called “Brandi Kaggle” can provide a blueprint for modeling and performance indicators of correct AI recommendations.

8. Why do you feel that the performance of the Upstage AI Pack is higher in the field ?

Upstage did not start with integration work. From the time of PoC before integration, we conducted numerous EDA (Exploratory Data Analysis) and data verification as if we were a team.

So we also got a strong feeling that something was going on during the month-long EDA period, and we instinctively knew that the success rate was high when we approached this way. In fact, I think that how much real data you look at and focus on the domain when doing AI modeling determines your performance.

I also admired Upstage's approach. I thought that if we proceeded in this way, we could achieve very good results, and the upstage side also quickly improved our performance by understanding the indicators and data we were aiming for and studying important data and methods.

As a result, performance improved quickly from the beginning. Among many companies, it was Upstage that directly delved into domain knowledge and started with data verification.

From data analysis to recommended AI, together as One Team!

9. WERE THERE ANY INTERESTING EPISODES OR MEMORABLE MOMENTS IN THE PROCESS OF INTRODUCING UPSTAGE’S RECOMMENDATION AI?

In fact, for a long time before recommendation AI was introduced, Upstage conducted EDA (Exploratory Data Analysis), and we also received feedback on data verification and consistency . Rather, I was wondering if I could hand over the data and get a way to optimize it further.

In fact, I think it moved like a team with our ML team (Machine Learning). As you will know if you use AI solution SaaS, there are not many people who work in a way that increases performance by sticking to domain knowledge in this way.

Also impressive is the performance . Usually, it takes 2 weeks to 2 months to improve performance, so the PoC period is longer, and in the meantime, whether or not to introduce a solution is put on hold . From there, it started outperforming our reference model and outperformed all other models.

10. Do you have any future plans related to upgrading the recommendation AI system ?

Upstage's recommendation platform is also being introduced to other platforms within Brandi. We are currently in the early stages of testing. In addition to unlimited recommendations, we are also discussing solutions such as search engines and search engine personalization to improve the quality of other pages within Brandi. All of these will be distributed through Bandit, along with the models we have previously created and used, and we are considering applying the upstage model to recommendations from outsiders, not inside the screen.

Since this entire collaboration system is measured by all attribution, I think that if those who create and research other recommendation models also join us, we will be able to participate in a virtuous cycle.

11. How would you describe upstage to commerce platform Brandy in one word?

For Brandi, Upstage is the jockey for boosting our platform performance . At the same time, I think of it as a key player and colleague who inspires the motivation of other players and creates a virtuous cycle .

I actually met a lot of people at Upstage, and they actively tried to help us with our work. I think it's not just a solution company, but a good relationship to pursue common interests with us.

Upstage's recommendation AI function will not only compete with our existing model, but will also create a new virtuous cycle. Brandi is a fashion commerce platform, but we expect that it will be a good relationship to develop AI in the field of recommendation with Upstage .

💡 IF YOU'RE CURIOUS ABOUT THE AI RECOMMENDATION SYSTEM? 👀