Algorithms vs Intuition: A Conversation with Altcraft’s CEO on ML – The Best Send Time Model

The world is changing rapidly, and those who can adapt to modern technologies stay one step ahead of their competitors. Today, we’ll discuss how machine learning (ML) is transforming the field of marketing.

Machine learning (ML) is a technology that enables systems to analyze data, identify patterns, and use them to make predictions or decisions without requiring manual programming for every step.

The relevance of ML in marketing cannot be overstated. It helps not only with segmenting target audiences but also with optimizing advertising campaigns, forecasting sales, and even personalizing offers. For instance, predictive analytics algorithms can process vast amounts of data on user behavior on a website and recommend products that align with their interests. This not only boosts conversion rates but also enhances the user experience.

There is a wide range of machine learning algorithms available, each designed to solve different types of tasks. Choosing the right mechanism can significantly influence the final analysis results. Let’s take a closer look at one of the most popular models developed using ML methods — the Best Send Time model.

To learn more about other applications of machine learning algorithms, check out our article "Forecasting Marketing Trends with Data Analysis and Machine Learning."

Best Send Time model

The Best Send Time model helps determine the optimal time to send a message to a customer. Why is this so important?

Imagine sending a promotional message at 3 a.m. What are the chances that the customer will see and open it? Or, consider sending it on an inconvenient day when your target segment is preoccupied with other tasks—the message might simply go unnoticed. Poor timing can lower open rates, diminish customer engagement, and ultimately lead to a drop in revenue.

One of the major international companies approached us to integrate the Best Send Time model into their marketing strategy. The client faced challenges in accurately segmenting their database for message timing. Due to the sheer volume of data, they couldn’t handle the task on their own and entrusted it to our team. How did the process of "creating" and "bringing to life" the Best Send Time model within CDP Altcraft unfold? Denis Chumachenko, the company’s CEO, shares the story.

— Denis, in your opinion, how does machine learning impact marketing?

— In today’s world, machine learning significantly accelerates and optimizes processes by automating routine tasks. Marketing professionals use it to predict user behavior, allowing them to adjust their strategies. For marketers, it’s crucial to predict the best time to send communications, as this increases the likelihood of messages being opened and products being purchased. One of the key goals of marketing is generating sales, which means striving to boost them and engage customers more effectively.

— Can you explain the essence of this model and its unique features?

— There’s a common belief that the process simply involves creating multiple time intervals and, based on segmentation, concluding that if a person clicked at a certain time before, they will click again at the same time. However, this is an incorrect approach. What if the calculations show their active time is 2 p.m., but it’s already 3 p.m.? Should you wait until the next day? That’s illogical. In reality, people have several active periods throughout the day. Our model determines when the client is most ready to engage while also considering their nearest active time window. This ensures you never miss the best moment to send a message to your audience. That’s undoubtedly a key feature of the Best Send Time model in Altcraft.

— How was the model created, and what challenges did you face?

— Initially, we chose the Gradient Boosting method to solve the task and used the XGBoost library. However, this approach turned out to be unsuccessful, as it didn’t provide the required performance on large datasets. To speed up predictions, we retrained the model using the CatBoost library and achieved a performance increase of more than 30 times compared to the initial version. We also faced the challenge that no ready-made library for working with CatBoost existed for our main programming language, GoLang. So, we had to develop one from scratch. We named it CatBoostGo and began actively using it to solve machine learning tasks within our team. The final result of our work was the previously mentioned Best Send Time model, which predicts the best time to send a message with mathematically confirmed accuracy of 82%.

— But why not 100% or something close to it?

— The model in question analyzes clients’ historical data. It processes information about each user and their activity. At the right moment, when a marketer needs to initiate communication—like within a specific scenario—the model predicts the optimal time for this in real time. It’s worth mentioning that data processing and predictions occur very quickly, even on large volumes, up to 100 million profiles. Regarding accuracy: if an ML model shows 100% accuracy, it likely means there’s an error in the training process, or the answer to the question is contained within the question itself. For predicting the best time to send messages, accuracy between 80% and 90% is considered excellent. The 82% accuracy figure was achieved using very large datasets provided by the client, containing data on individuals from various countries worldwide. However, when we narrowed the dataset to focus on a single country, the model demonstrated 93% accuracy.

— What enabled you to achieve such results?

— I believe it’s simply because our team consists of highly skilled professionals who approach familiar challenges with fresh perspectives. We also owe a great deal of gratitude to our client, who provided the data for training; without it, we wouldn’t have achieved such results. The Altcraft team is always open to new proposals and ready to implement any innovative ideas.

— What advantages set Altcraft Platform apart from similar solutions?

— Modern software often comes with an overwhelming amount of functionality, making it difficult to navigate. Users expect AI to automatically handle all tasks for them, but this approach comes with a limitation: an ML module taking full control prevents users from making adjustments or customizing settings. At Altcraft Platform, we strive to maintain a balance between automation powered by ML models and functions that users can manage themselves.

For more detailed information about how the Best Send Time model works within the CDP Altcraft platform, please refer to the documentation.

Conclusion

It can be confidently stated that machine learning is becoming a key tool in the arsenal of the modern marketer. The ability to analyze vast amounts of data, uncover hidden patterns, and predict consumer behavior opens up new horizons for optimizing strategies and improving campaign efficiency. Integrating machine learning into marketing processes not only helps businesses better understand their audience but also provides a competitive edge in a dynamically changing market. In a world where data plays a pivotal role, leveraging machine learning is not just a trend but a necessity for achieving success!