Let's write your success story together
At SEM, we proudly share the success stories of our clients. Every global and local client who embarks on this journey with us becomes a part of these success stories. How about writing the success story of your brand with us?
LC Waikiki & SEM - BTS Case Study
*Results obtained during the campaign period from August to September 2024.
Performance Marketing Academy: The New Era of AI Targeting
883%
Increase in ROI
(Google App Engagement – iOS)
382%
Increase in ROI
(Google App Engagement – Android)
366%
Increase in CR
(Google Search)
400%
Increase in ROAS
(Meta retargeting group)
700%
Increase in ROAS
(Meta new users)
Back to school has never been so data-stylish! While taking children on a fashion adventure full of style from kindergarten to campus, we support them with the most comfortable and stylish products. With our data-based technologies, we know each of our users one-on-one and respond to their needs with the right campaign strategies. For both children and parents, back to school is not just a beginning, but an adventure full of LCW-chic!
By making sense of our users’ interest in age, gender and product groups with artificial intelligence and machine learning, we aimed to identify more valuable users who will reach our revenue target with the insights we have obtained from past campaign data.
While we use the gender, period, product and age insights we have obtained from our own first-party data in the most effective way in target audience weighting, such as the weight of our school bags, just as we share washing the nutrition covers we use together at lunch, Decoupling in campaign budgeting, we aimed to create a fashion story that grows from nursery to campus.
We collected the data through digital platforms and multiple channels in the cloud environment. By integrating Google Merchant Center product data with the LCW database, we have combined all products and modeled the user’s interest based on gender, Decadence and product groups even before purchasing with the signals we collected via GA4. This step, combined with users’ interactions not only with purchasing actions, but also with product listing, product details and adding to the cart, kept their interest alive. Based on the data at each step, we applied techniques blended with RFM & pareto algorithm, percentage rankings, hyper-scoring and ML models to find upper/lower segment users.
Based on our own data, we created 3 different models based on past behavior and compared the audiences both internally, with similar campaigns, and with adgroups in the same campaign.
In the first audience, we analyzed users’ interactions with products by product listing, detail review, adding to the cart, purchasing metrics, and created segments based on the product categories and prices they are interested in. We integrated gender, age and churn data into the model and determined the degrees of interest with coefficients. In this analysis, when evaluating the degree of interest of a user in gender and age groups, it was observed that product viewing behavior determines the user’s interest more by taking a higher coefficient than purchasing activity. The user’s proximity to the last purchase date determined the probability of future purchases with a high positive coefficient in the model. Optimal scores were calculated for each segment using the percentage rank algorithm, and these groups were divided into high- and medium-value user groups using ML models.
We also added the seasonality factor to the second audience group and included periodic product groups in the model in addition to gender. We have added a new dimension to the model as “target” and “non-target” by determining the products that children and young people who are preparing for school, kindergarten or campus can buy in the autumn/ winter season. The actions of users in these categories were rated according to the total user average as a result of ML with different coefficients in the model, taking into account the time difference between the most recent action dates and the present day. For example, when a user made a purchase from the non-target category, he received a positive coefficient of 0.12, while a purchase from the target category received a negative coefficient of -0.22. Recently, shopping in the target category was evaluated with a negative coefficient of -0.38, and in the non-target category with a positive coefficient of 0.28. With this seasonal analysis, we used ML algorithms to determine the users who will exchange BTS/BTK/BTC and create 12 optimized clusters on a platform-period basis.
We modeled our final audience based only on purchase, revenue and timeliness metrics. Our goal was to determine the weight of users in 12 clusters based on platform and period by reaching a coefficient of 1 in total. According to the figures obtained as a result of the model, while using 0.77 coefficient for purchase and 0.88 coefficient for income, we determined -0.18 coefficient for the day difference between the first and last purchase dates in Decennial product groups and -0.47 coefficient for the period from the last purchase date to the present. These values were used to create the final score for each user, and we calculated optimal scores using ML algorithms to divide users into high- and medium-value segments. We weighted and automated each group according to the approximate quartile, standard deviation and median values.
In order for these three audience groups to remain dynamic and living audiences, we have ensured that they are constantly and renewed transferred to our digital channels with the systems and automations we have developed. Thus, we were able to accurately target high- and medium-value user segments without touching them, and we dynamized these segments in our campaigns.
We have further strengthened our presence in all digital with this campaign setup and we have established a strategy that we call “brandformance” by conducting both branding and performance publications under one roof for the first time. We wanted to be everywhere users looked by taking advantage of all the inventory offered by the media and created Search, PMax, YouTube, Demand Gen and App Engagement publications for performance on Google, Efficient Reach, Video View and Demand Gen publications for branding. In addition to our standard measurement, we have strengthened the measurement of our branding publications with BLS and analyzed the effectiveness of our campaign in detail. Oct.
With the Broad targeting strategy for performance in Meta, App Install, we have moved forward with the lower funnel targeted Catalog Sales & Conversion campaigns with the insight that it performs the best in our targets. We have also included Reach and Video Views campaigns for Branding in our strategy. In the settlements, we used all platforms such as Instagram & Facebook Feed, Story and Reels.
We have positioned our campaigns appropriately at all steps of the famous e-commerce funnel known to everyone in order not only to be visible everywhere, but also to increase awareness and promote the campaign. We have created a whole by combining our high and mid audiences that we have created with the events they work on in the appropriate campaign structures, the relevant phase of the funnel, audiences, signals and our products.
In accordance with our audiences and strategy, we have customized our audiences in separate ad groups according to gender, period, environment and product groups in the videos and visuals we have specially created. By identifying the best performing creatives with the data we obtained from previous campaign trainings, we used these creatives in the most appropriate campaign structure in all media.
And the report card day after the long marathon, we have good news, all the courses are very good, let’s check it out together!
At the end of the BTS period, the shopping frequency, revenue and conversion rates of the audiences we modeled in all the data increased in direct proportion to the predicted scores, while all our audiences were many times higher than the average, as we expected when we compared 3 models, our most trusted audience became the best employee.
With 332M impressions on Google, all of our campaigns, where our videos were watched 24M times, provided a full-funnel increase, while in their details;
In our App Engagement campaigns, which are our main source of performance; iOS campaigns, which have become difficult to measure and target with the zero teacher in the entire account, performed higher for the first time even than Android, which is the class mother of the account, thanks to audience/campaign/creative alignment, and these campaigns provided 5 times higher ROAS and CR than category campaigns working in the same focus, and 2 times higher than sales campaigns.
In PMAX, where we have allocated the majority of our budget, where we have added our high-value users as a signal, there was a 220% increase in ROI compared to a 53% increase in CR.
On YouTube, which we use to increase our brand awareness with audio and visual stories, we increased ROAS by 10% with a 12% lower CPA cost. By the way, let Deca not hear, the rival campaign was also the predictive masses of GA4, we passed them with our tailor-made masses.
Our Demand Gen campaign, which we are conducting with our demand generation strategy, has the best conversion rate among its inventory, generating exactly 24 times the revenue of our spending, and is the 2nd best employee of the account. Dec.we turned it into a campaign.
Those who are researching school products have not forgotten you either, we have achieved a 4.5-times CR increase with the broad match strategy of the words we have adopted in our Search publications, while bringing a 2-times ROI increase with us.
We created Efficient Reach for Awareness phase, Video View for Consideration in our branding publications where we own the upper phase of the funnel, and then we created a Demand Gen campaign where we ran Product View because we wanted to bring a low-cost session in the conversion phase. With our Efficient Reach publications, we have achieved cost-effective reach with low CPM rates. In addition, our Efficient Reach campaign and our Video View campaign were fed and enabled us to get a View with low CPV costs. We also managed to attract quality traffic with the Demand gen campaign.
Finally, we achieved great results by achieving 3% absolute brand lift, 8% relative brand lift, 5% headroom brand lift in the BLS results where we proved the impact of our branding publications on performance.
In Meta, all of our full-funnel campaigns provided obvious increases in all campaign types, while in their details;
You couldn’t have missed our ads that we received 2.5 times as many impressions as the population of Turkey (205M), that we reached as many people as the entire number of children attending school in 2023 (25M), that our videos were watched 2 times as much as teachers in our country (6.7M), and that the average frequency of a person’s exposure to an ad was 8, couldn’t you?
We left the spending on all the placements in our fiction to Meta’s campaign learning; we experienced that this strategy maximizes performance.
Analyzing our own data, modeling and determining the correct customer segmentations and user behaviors, integrating these audiences into Meta and using them in a way appropriate to the strategy, made a significant contribution to the campaign success. The relevant audiences achieved an unprecedented success that we have not seen before with an average ROAS of 300-400% higher compared to retargeting audiences and 500-700% higher compared to new user audiences.
In Mid funnel, we found that our Data audiences achieved 39% lower session costs compared to the Interest audience in the ViewContent-oriented campaign strategy in Catalog Sales campaigns.
By separating the contents specially prepared for BTS from our general catalog publications, we preferred to evaluate them in a special conversion campaign this time. In our previous experiences, we did not really prefer this approach because we could not get the performance we expected from the campaigns we created in this structure, but this time the campaign performance of the audience we integrated into the modeled audiences and Meta strategies was completed with a stunning result that had never been achieved before, such as 40 ROAS. In interest and retargeting audiences, this ratio remained around 1-2. These results, which we previously did not prefer due to low results, but in this project we achieved 20 times higher ROAS thanks to the power of the masses specifically, showed us that the high audience is quite prone to shopping during this period, correctly modeled and quite consistent for this campaign and period.
In our Catalog Sales objective campaign, we found that our data audiences provided 5 times more ROAS compared to interest and dynamic remarketing audiences in our sales-oriented publications. Considering the target audience size between the new user audience and the fact that the spending was made more to the modeled audience clearly shows that Meta also prefers this audience in campaigns working with the CBO, this once again proved how critical our audiences play in our strategy.
We have created a special campaign setup to measure the performance of the 3 audiences we have created in different segments, as well as in themselves. We have observed that the category audiences that we trust more in this campaign have performed the best compared to the first and third audiences. We achieved the highest efficiency with a 6-times difference at the lowest cost in both session and ROAS metrics. In addition, it should be noted that all audiences work with equal budgets, which further highlights the efficiency of category audiences.
We have written a success story that shows how the fiction, data-oriented models we have realized during this period have been made functional with the right fictions and will almost be told as a lesson in schools. We not only collected data, but also modeled audiences with strategic insights, found the highest-value users, and built a multi-channel “brandformance” digital ecosystem. Thanks to this, we have achieved many times more successful results in both sales and brand awareness compared to previous years. We hope that the little ones and young people will have a successful period as well as us during this education and training period, let’s go to classes now!
We would like to thank SEM team, firstly Berkan Şişman who built the project, ilker uğurlu, Merve Güre Karakaya, Halil Serhat Yıldız, Taylan Gök, Alp Eren Gönüllü, Orhan Gümülcineli, Yağmur Karabaş, Turgut Kutlay Boya, Nurtay Nejat Ermiş and Buse Betoner Öründü, Kadir Kadıoğlu, Kemal Sögütlü, Omer Barbaros Yis, Irem YILMAZ, Büşra Sarıman, Sinem Akgül Yılmaz and Ebru Derindere Öner from the LC Waikiki team for their support 🎉
*Results obtained during the campaign period from August to September 2024.
