Çiçeksepeti Instant BQML

[Arda Karatas]

Selamlar,

Instant BQML kurulumunu bizim tarafta gerçekleştirdik. Pipeline'lar için gerekli parametreleri girerek örnek olarak add_to_card event'i için json dosyalarını oluşturduk.

Oluşan training ve prediction json dosyalarını CRMint App üzerine import ettik. Öncelikle Training pipeline'ını çalıştırdık ve ilgili model oluştu. Daha sonra, Prediction pipeline'ını çalıştırdık. Belirttiğimiz proje ve datasette tüm tablolar oluşturuldu. 

addcart_prob_performance_insights tablosu haricindeki tüm tablolarda veriler oluştu. Google Analytics üzerinde de Kitleler klasörü oluştu ama içerisindeki bilgiler boş gelmektedir. Tüm sürecin ekran görüntülerini ekte paylaşıyorum. GA üzerinde verileri boş görmemizin sebebi ne olabilir, süreç içerisinde yapmamız gereken farklı bir adım var mıdır?

İyi çalışmalar,

--

Arda KARATAŞ Data & BI Analytic Engineer Technology E : arda.karatas@ciceksepeti.com                    A : Istanbloom, Esentepe Mah. Zincirlikuyu, Kore Şehitleri Cad. No: 16/1 34394, İstanbul

[Arda Karatas]

Hello All,

I am also sharing the English version of my previous email.

We have successfully completed the Instant BQML setup on our end. By entering the required parameters for the pipelines, we created sample JSON files for the add_to_cart event.

The generated training and prediction JSON files were imported into the CRMint App. We first ran the training pipeline, which successfully created the model. Then, we executed the prediction pipeline. All related tables were created in the specified project and dataset.

Data was generated in all tables except for the addcart_prob_performance_insights table. Additionally, an “Audiences” folder was created in Google Analytics, but it appears to be empty. I am attaching screenshots of the entire process for your reference.

Could you please advise on why the data might not be appearing in GA? Is there any additional step we might have missed during the process?

Best Regards,

Arda KARATAŞ

Arda Karatas <arda.k...@ciceksepeti.com>, 3 Tem 2025 Per, 17:08 tarihinde şunu yazdı:

[Arda Karatas]

Hi Team,

We have recently trained a model named addcart_prob_model_web using Instant BQML, aimed at predicting the probability of a product being added to cart on our web platform. The entire pipeline is running successfully through CRMint.

As we analyze the output, we have several questions regarding the model evaluation metrics and prediction-related tables that were automatically generated by the Instant BQML process.

• Mean Absolute Error: 0.1222
• Mean Squared Error: 0.0488
• Mean Squared Log Error: 0.0246
• Median Absolute Error: 0.0536
• R² Score: 0.2212

Whether these metrics are within acceptable ranges for binary classification problems in e-commerce contexts (e.g., add-to-cart prediction)?

How should we interpret a relatively low R² score (0.22) – is this expected for behavioral models trained via Instant BQML?

Do you have any internal benchmarking data or best practices for interpreting these specific metrics when using Instant BQML?

After running the pipeline, the following tables were created:

• addcart_prob_predictions_web
• addcart_prob_scored_users_log_web
• addcart_prob_uservaluemap_web
• addcart_prob_performance_insights
• addcart_prob_audience_boundaries_web
• addcart_prob_calculated_fields
• addcart_prob_calculated_session_id_timestamps
• addcart_prob_calculated_visitors
• addcart_prob_conversionvalues_web
• addcart_prob_measurement_protocol_formatted_session_attribution_web
• addcart_prob_measurement_protocol_formatted_web

A brief explanation of the purpose of each table and how they should be used in practice (e.g., final scores, diagnostic, attribution, input references).

Whether other auxiliary or optional tables could also be created depending on configuration.

And most importantly: is there an official Instant BQML documentation or guide that explains these table structures and their intended usage?

Best Regards,

Arda Karatas <arda.k...@ciceksepeti.com>, 3 Tem 2025 Per, 17:23 tarihinde şunu yazdı:

[Arda Karatas]

Hi Team,

I wanted to follow up on the requests and questions I shared in my previous email. I’d appreciate it if you could let me know if there have been any updates on these matters.

Looking forward to your response.

Best regards,

Arda KARATAŞ Data & BI Analytic Engineer Technology E : arda.karatas@ciceksepeti.com                    A : Istanbloom, Esentepe Mah. Zincirlikuyu, Kore Şehitleri Cad. No: 16/1 34394, İstanbul

7 Tem 2025 Pzt, saat 14:19 tarihinde Arda Karatas <arda.k...@ciceksepeti.com> şunu yazdı: