Skip to content

2023

  • 3 min read

Understanding AdaBoost and Gradient Boosting Machine

In the realm of machine learning, two of the most potent and widely-used algorithms are AdaBoost and Gradient Boosting Machine (GBM). Both of these techniques are used for boosting, a method that sequentially applies weak learners to improve model accuracy. Let's delve deeper into each of these algorithms, their workings, and differences.

AdaBoost: The Adaptive Boosting Pioneer

AdaBoost, short for Adaptive Boosting, was introduced in the late 1990s. This algorithm has a unique approach to improving model accuracy by focusing on the mistakes of previous iterations.

How AdaBoost Works

  1. Initial Equal Weighting: AdaBoost starts by assigning equal weights to all data points in the training set.
  2. Sequential Learning: It then applies a weak learner (like a decision tree) to classify the data.
  3. Emphasis on Errors: After each round, AdaBoost increases the weights of incorrectly classified instances. This makes the algorithm focus more on the difficult cases in subsequent iterations.
  4. Combining Learners: The final model is a weighted sum of the weak learners, with more accurate learners given higher weights.

AdaBoost's Key Features

  • Simplicity and Flexibility: It can be used with any learning algorithm and is easy to implement.
  • Sensitivity to Noisy Data: AdaBoost can be sensitive to outliers since it focuses on correcting mistakes.

Gradient Boosting Machine: The Evolution

Gradient Boosting Machine (GBM) is a more general approach and can be seen as an extension of AdaBoost. It was developed to address some of AdaBoost's limitations, particularly in handling a broader range of loss functions.

How GBM Works

  1. Sequential Learning with Gradient Descent: GBM uses gradient descent to minimize errors. It builds one tree at a time, where each new tree helps to correct errors made by the previous ones.
  2. Handling Various Loss Functions: Unlike AdaBoost, which focuses on classification errors, GBM can optimize any differentiable loss function, making it more versatile.
  3. Control Over Fitting: GBM includes parameters like the number of trees, tree depth, and learning rate, providing better control over fitting.

GBM's Key Features

  • Flexibility: It can be used for both regression and classification tasks.
  • Better Performance: Often provides better predictive accuracy than AdaBoost.
  • Complexity and Speed: More complex and typically slower to train than AdaBoost, especially with large datasets.

AdaBoost vs Gradient Boosting Machine: A Comparison

While both algorithms are based on the idea of boosting, they differ significantly in their approach and capabilities:

  • Focus: AdaBoost focuses on classification errors, while GBM focuses on minimizing a loss function.
  • Flexibility: GBM is more flexible than AdaBoost in terms of handling different types of data and loss functions.
  • Performance: GBM generally provides better performance, especially on more complex datasets.
  • Ease of Use: AdaBoost is simpler and faster to train, making it a good starting point for beginners.

Conclusion

Both AdaBoost and Gradient Boosting Machine have their unique strengths and are powerful tools in the machine learning toolbox. The choice between them depends on the specific requirements of the task, the nature of the data, and the desired balance between accuracy and computational efficiency. As machine learning continues to evolve, these algorithms will undoubtedly remain fundamental, continuing to empower new and innovative applications.

  • in zh
  • 1 min read

理解AdaBoost和梯度提升機器

在機器學習領域中，兩種最有力且被廣泛使用的算法是AdaBoost和梯度提升機器(GBM)。這兩種技術都被用於提升，一種逐步應用弱學習器以提高模型準確性的方法。讓我們深入了解每種算法的工作原理，以及它們的區別。

AdaBoost: 自我調整增強的先驅

AdaBoost，全名為自適應增強，於20世紀90年代末被介紹。這個算法通過專注於前一個迭代的錯誤來改進模型的準確性有一種獨特的方法。

AdaBoost的工作原理

  1. 初始等權重：AdaBoost首先給訓練集中的所有數據點分配相同的權重。
  2. 序列學習：然後，它應用一個弱學習器（如決策樹）對數據進行分類。
  3. 對錯誤的強調：每一輪過後，AdaBoost會增加分類不正確的實例的權重。這使得算法在後續的迭代中更加專注於困難的案例。
  4. 組合學習器：最終的模型是弱學習器的加權和，其中更準確的學習器給予更高的權重。

AdaBoost的主要特點

  • 簡單和靈活：它可以與任何學習算法一起使用，並且易於實現。
  • 對噪聲數據的敏感性：AdaBoost可能對異常值敏感，因為它專注於糾正錯誤。

梯度增強機：演進

梯度提升機（GBM）是一種更一般的方法，可以被視為AdaBoost的擴充。它被開發出來解決AdaBoost的一些限制，尤其是在處理更廣泛的損失函數方面。

GBM的工作原理

  1. 用梯度下降進行序列學習：GBM使用梯度下降來最小化錯誤。它一次構建一棵樹，每棵新樹都有助於糾正前一棵樹的錯誤。
  2. 處理各種損失函數：與AdaBoost不同，調用對分類誤差，GBM可以優化任何可微分的損失函數，使其更具通用性。
  3. 對擬合的控制：GBM包含樹的數量，樹的深度和學習率等參數，提供了更好的對擬合的控制。

GBM的主要特點

  • 靈活性：它可以用於回歸和分類任務。
  • 更好的性能：通常比AdaBoost提供更好的預測準確性。
  • 複雜性和速度：比AdaBoost更複雜，尤其是對於大數據集來說，訓練通常較慢。

AdaBoost vs 梯度提升機：比較

雖然這兩種算法都基於增強的想法，但在其方法和能力方面有顯著的區別：

  • 焦點：AdaBoost關注分類錯誤，而GBM關注最小化損失函數。
  • 靈活性：在處理不同類型的數據和損失函數方面，GBM比AdaBoost更靈活。
  • 性能：GBM通常提供更好的性能，尤其是對於更複雜的數據集。
  • 使用的簡便性：AdaBoost更簡單，更快地訓練，因此它是初學者的一個好的起點。

結論

Adaboost和梯度提升機都有自己獨特的優點，並且是機器學習工具箱中的強大工具。在它們之間的選擇取決於任務的具體要求，數據的性質，以及在準確度和計算效率之間的平衡。隨著機器學習的不斷發展，這些算法無疑將繼續存在，並繼續賦予新的和創新的應用。

  • 3 min read

Understanding Bootstrap Aggregation and Random Forest

In the world of machine learning, there are numerous techniques and algorithms that empower predictive modeling and data analysis. Two such powerful methods are Bootstrap Aggregation, commonly known as Bagging, and Random Forest. These techniques are widely used for their robustness and ability to improve the accuracy and stability of machine learning models.

What is Bootstrap Aggregation (Bagging)?

Bootstrap Aggregation, or Bagging, is an ensemble learning technique used to improve the stability and accuracy of machine learning algorithms. It reduces variance and helps to avoid overfitting. The concept of Bagging was introduced by Leo Breiman in 1994 and has since become a cornerstone in the field of machine learning.

How Does Bagging Work?

Bagging involves creating multiple versions of a predictor and using these to get an aggregated predictor. The main steps are:

  1. Random Sampling with Replacement: The original dataset is sampled randomly with replacement, creating multiple bootstrapped datasets.
  2. Model Training: A model is trained separately on each bootstrapped dataset.
  3. Aggregation of Predictions: The predictions from each model are combined (usually by averaging for regression problems or voting for classification problems) to form a final prediction.

The beauty of Bagging lies in its simplicity and effectiveness, especially for decision tree algorithms, where it significantly reduces variance without increasing bias.

Random Forest: An Extension of Bagging

Random Forest is a popular ensemble learning technique that builds upon the concept of Bagging. Developed also by Leo Breiman, it involves constructing a multitude of decision trees at training time and outputting the class that is the mode of the classes (classification) or mean prediction (regression) of the individual trees.

How Does Random Forest Differ from Basic Bagging?

  1. Use of Decision Trees: Random Forest specifically uses decision trees as its base learners.
  2. Feature Randomness: When building each tree, a random subset of features is chosen. This ensures that the trees are de-correlated and makes the model more robust to noise.
  3. Multiple Trees: A Random Forest typically involves a larger number of trees, providing a more accurate and stable prediction.

Advantages of Random Forest

  • High Accuracy: Random Forests often produce highly accurate models, especially for complex datasets.
  • Robust to Overfitting: Due to the averaging of multiple trees, the risk of overfitting is lower compared to individual decision trees.
  • Handles Large Datasets Efficiently: They are capable of handling large datasets with higher dimensionality.

Applications and Considerations

Both Bagging and Random Forest find applications in various fields, including finance for credit scoring, biology for gene classification, and many areas of research and development. However, while using these techniques, one must be mindful of the following:

  • Computational Complexity: Both methods can be computationally intensive, especially Random Forest with a large number of trees.
  • Interpretability: Decision trees are inherently interpretable, but when combined into a Random Forest, the interpretability decreases.
  • Parameter Tuning: Tuning parameters like the number of trees, depth of trees, and number of features considered at each split is crucial for optimal performance.

Conclusion

Bootstrap Aggregation and Random Forest are powerful techniques in the arsenal of a data scientist. By understanding and correctly applying these methods, one can significantly improve the performance of machine learning models, tackling both bias and variance, and thereby making robust and accurate predictions. As with any tool, their effectiveness depends largely on the skill and understanding of the practitioner in applying them to the right kind of problems.

  • in zh
  • 1 min read

理解Bootstrap Aggregation與隨機森林

在機器學習的世界中，有許多技術和算法可以強化預測模型和數據分析。其中兩種強大的方法就是Bootstrap Aggregation，通常被稱為Bagging，以及隨機森林。這兩種技術因其穩健性以及能夠提高機器學習模型的精確性和穩定性而被廣泛使用。

什麼是Bootstrap Aggregation (Bagging)？

Bootstrap Aggregation，即Bagging，是一種集成學習技術，用於提高機器學習算法的穩定性和準確性。它能減少方差並有助於避免過度擬合。Bagging的概念由Leo Breiman於1994年提出，並已成為機器學習領域的基石。

Bagging如何運作？

Bagging包括創建預測器的多個版本並使用它們來得到一個聚合的預測器。主要步驟包括：

  1. 隨機抽樣並替換：原始資料集經過隨機抽樣並替換，創造出多個自助的資料集。
  2. 模型訓練：每個自助的資料集都單獨訓練一個模型。
  3. 預測結果匯總：所有模型的預測結果合併（通常對於迴歸問題進行平均或對於分類問題進行投票）形成最終的預測。

Bagging的美在於其簡單有效，特別是對於決策樹算法，它顯著地降低了方差而沒有增加偏差。

隨機森林：Bagging的擴展

隨機森林是一種流行的集成學習技術，建立在Bagging的概念之上。由Leo Breiman同樣發展出來，它包括在訓練時構建多個決策樹，並輸出各決策樹的類別模式（分類）或平均預測（迴歸）。

隨機森林與基礎Bagging的區別？

  1. 使用決策樹：隨機森林具體使用決策樹作為其基礎學習器。
  2. 特徵隨機選擇：構建每棵樹時，會選擇一組隨機的特徵子集。這確保了樹的相關性降低，並使模型對噪音更具韌性。
  3. 多棵樹：隨機森林通常包括更多的樹，提供更準確和穩定的預測。

隨機森林的優點

  • 高精確度：對於複雜的數據集，隨機森林常能產生高精確度的模型。
  • 對於過度擬合的韌性：由於多個樹的平均，相較於單一的決策樹，隨機森林對於過度擬合的風險降低。
  • 有效處理大數據集：它們能夠有效地處理具有較高維度的大數據集。

應用與考量

Bagging和隨機森林在許多領域都有應用，包括金融中的信用評分，生物學中的基因分類，以及各種研究和開發領域。然而，在使用這些技術時，必須謹記以下幾點：

  • 計算複雜性：這兩種方法可能會非常消耗計算資源，特別是隨機森林中樹的數量較多的情況。
  • 可解釋性：決策樹本質上是可以解釋的，但當它組合成隨機森林時，可解釋性會降低。
  • 參數調整：調整像樹的數量、樹的深度以及每個分割點考慮的特徵數量等參數對於獲得最佳性能非常關鍵。

結論

在數據科學家的工具箱中，Bootstrap Aggregation和隨機森林都是強大的技術。通過理解和正确應用這些方法，可以顯著提高機器學習模型的性能，同時處理偏差和方差，從而使預測更為穩健和準確。像任何工具一樣，他們的有效性大部分取決於應用他們來解決適當問題的實踐者的技能和理解。

  • 2 min read

Understanding Inertia and Silhouette Coefficient - Key Metrics in Clustering Analysis

Clustering is a fundamental technique in data science and machine learning, used for grouping similar data points together. Among the various metrics to evaluate the quality of clustering, Inertia and Silhouette Coefficient stand out for their insightful feedback on cluster quality. Let's dive into what these metrics are and how they help in analyzing clusters.

What is Inertia?

Inertia, also known as within-cluster sum-of-squares, measures the compactness of clusters. It calculates the total variance within the clusters. In simpler terms, it's the sum of the distances of each data point in a cluster to the centroid of that cluster, squared and summed up for all clusters.

Key Points:

  • A lower inertia value implies a better model, as it indicates tighter clustering.
  • However, the inertia metric has a drawback: it keeps decreasing with an increase in the number of clusters ( k ). This is where the "elbow method" is often used to find the optimal ( k ).
Understanding the Silhouette Coefficient

The Silhouette Coefficient is a measure of how similar an object is to its own cluster (cohesion) compared to other clusters (separation). The silhouette ranges from -1 to +1, where a high value indicates that the object is well matched to its own cluster and poorly matched to neighboring clusters.

Key Points:

  • A high silhouette score indicates well-clustered data.
  • Unlike inertia, the silhouette score provides more nuanced insight into the separation distance between the resulting clusters.
When to Use Each Metric

  1. Inertia:

  2. Good for assessing the compactness of clusters.

  3. Best when used with the elbow method to determine the optimal number of clusters.

  4. More sensitive to the scale of the data, so normalization or standardization might be necessary.

  5. Silhouette Coefficient:

  6. Ideal for validating the consistency within clusters of data.
  7. Useful when the number of clusters is not known.
  8. Offers a more balanced view, incorporating both cohesion and separation.
Conclusion

Inertia and Silhouette Coefficient are crucial metrics for evaluating the performance of clustering algorithms like K-Means. They provide different perspectives: inertia focuses on internal cluster compactness, while silhouette coefficient assesses how well-separated the clusters are. The choice of metric often depends on the specific requirements of the clustering problem at hand.

  • in zh
  • 1 min read

理解慣性和輪廓係數 - 分群分析中的關鍵指標

分群是資料科學和機器學習中的基本技術，用於將相似的資料點分組在一起。在評估分群質量的各種指標中，慣性和輪廓係數以其對分群質量深入的反饋而脫穎而出。讓我們深入了解這些指標是什麽，以及它們如何幫助分析分群。

什麽是慣性？

慣性也稱為群內平方和，用於衡量分群的緊密度。它計算分群內的總變異。簡單來說，就是每個資料點到該分群重心的距離的平方值的總和，並為所有分群加總。

關鍵點：

  • 較低的慣性值表示模型較好，因為它表示分群較為緊密。
  • 但是，慣性指標有一個缺點：隨著分群數量（k）的增加，它會持續下降。這就是常常使用"肘部方法"來找到最佳的（k）的地方。
理解輪廓係數

輪廓係數是一種衡量物體與自己分群的相似度（凝聚力）與其他分群（分離度）之間的差異的度量。輪廓值範圍是-1到+1，其中高值表明物體與自己的分群匹配得很好，並且與相鄰分群的匹配度差。

關鍵點：

  • 高輪廓得分表示資料分群良好。
  • 與慣性不同，輪廓得分對分群間的分離距離提供了更細微的見解。
何時使用每一個指標

  1. 慣性：

  2. 良好的分群緊密度評估工具。

  3. 目測分群最佳數量時與肘部方法配合使用最佳。

  4. 對資料的尺度更敏感，因此可能需要正規化或標準化。

  5. 輪廓係數：

  6. 驗證分群資料內部一致性的理想工具。
  7. 在不知道分群數量的情況下很有用。
  8. 提供了更均衡的視角，包括凝聚力和分離度。
結論

慣性和輪廓係數是評估像K-Means這樣的分群演算法性能的關鍵指標。它們提供了不同的視角：慣性專注於內部分群的緊密度，而輪廓係數評估分群之間的分離性如何。選擇使用哪個指標通常取決於手頭分群問題的具體要求。

  • 3 min read

Understanding Regularization - Lasso, Ridge, and Elastic Net Regression

In the field of machine learning and statistical modeling, regularization is a crucial technique used to prevent overfitting and improve the generalization of models. This blog post will delve into three popular regularization methods: Lasso, Ridge, and Elastic Net Regression, elucidating how they function and when to use them.

What is Regularization?

Regularization is a technique used to reduce overfitting in machine learning models. Overfitting occurs when a model learns not only the underlying pattern in the training data but also the noise. This leads to poor performance on unseen data. Regularization addresses this issue by adding a penalty term to the loss function used to train the model. This penalty term constrains the model, making it simpler and less prone to overfitting.

Ridge Regression (L2 Regularization)

Ridge Regression, also known as L2 regularization, adds a penalty equal to the square of the magnitude of the coefficients. The regularization term is added to the loss function, and it includes a tuning parameter, λ (lambda), which determines the strength of the penalty. A higher value of λ shrinks the coefficients more, leading to a simpler model.

Key Features of Ridge Regression:

  • It tends to shrink the coefficients of the model uniformly.
  • Suitable for scenarios where many features have a small or moderate effect on the output variable.
  • Ridge regression does not perform variable selection - it includes all features in the final model.

Lasso Regression (L1 Regularization)

Lasso Regression, short for Least Absolute Shrinkage and Selection Operator, involves L1 regularization. It adds a penalty equal to the absolute value of the magnitude of coefficients. Like Ridge, it also has a tuning parameter, λ, which controls the strength of the penalty.

Key Features of Lasso Regression:

  • Lasso can shrink the coefficients of less important features to exactly zero, thus performing variable selection.
  • Useful when we have a large number of features, and we suspect that many of them might be irrelevant or redundant.
  • Can lead to sparse models where only a subset of the features contributes to the prediction.

Elastic Net Regression

Elastic Net Regression is a hybrid approach that combines both L1 and L2 regularization. It adds both penalties to the loss function. Elastic Net is particularly useful when there are multiple correlated features. It includes two parameters: λ (like in Lasso and Ridge) and α, which balances the weight given to L1 and L2 regularization.

Key Features of Elastic Net Regression:

  • Balances the properties of both Lasso and Ridge.
  • Works well when several features are correlated.
  • Elastic Net can be tuned to behave like Lasso or Ridge regression by adjusting the α parameter.

Choosing the Right Regularization Method

The choice between Lasso, Ridge, and Elastic Net depends on the data and the problem at hand:

  • Ridge is a good default when there is not much feature selection needed or if the features are expected to have roughly equal importance.
  • Lasso is preferred if feature selection is essential, and there is a need to identify the most significant variables.
  • Elastic Net is ideal when there are multiple correlated features, or a balance between feature selection and uniform coefficient reduction is required.

Conclusion

Regularization is a powerful tool in machine learning, helping to enhance the performance and interpretability of models. Lasso, Ridge, and Elastic Net are versatile methods that can be applied to various regression problems. Understanding their differences and applications is key to building robust and accurate predictive models.

  • in zh
  • 1 min read

理解正則化 - Lasso、Ridge 和 Elastic Net 回歸

在機器學習和統計建模領域中，正則化是一種關鍵的技術，用於防止過度擬合並提高模型的泛化能力。此博客文章將深入探討三種熱門的正則化方法： Lasso、Ridge 和 Elastic Net 回歸，闡述它們的運作方式以及何時使用它們。

什麼是正則化?

正則化是一種用於減少機器學習模型過度擬合的技術。過度擬合，發生在模型不僅學習了訓練數據中的基本模式，還學習了噪聲。這會導致對未見過的數據表現不佳。正則化通過在用來訓練模型的損失函數中添加一個懲罰項來解決這個問題。這個懲罰項限制了模型，使其更簡單且不易過度擬合。

Ridge 回歸 (L2 正則化)

Ridge回歸，也稱為L2正則化，添加的懲罰等於係數幅度的平方。正則化項被添加到損失函數中，並包括一個調節參數 λ (lambda)，λ 決定了懲罰的強度。 λ 的值越大，係數縮減得越多，導致模型更簡單。

Ridge回歸的關鍵特徵：

  • 它傾向於均勻縮小模型的係數。
  • 適合許多特徵對輸出變量有小到中度影響的場景。
  • Ridge 回歸不進行變量選擇 - 它包含所有特徵在最終模型中。

Lasso 回歸（L1正則化）

Lasso回歸（最小絕對收縮和選擇算子）涉及到L1正則化。它添加的懲罰等於係數幅度的絕對值。同樣，它也有一個調節參數，λ ，用於控制懲罰的強度。

Lasso回歸的關鍵特徵：

  • Lasso可以將較不重要特徵的係數縮減到完全為零，從而進行變量選擇。
  • 當我們有大量的特徵，並且懷疑其中許多可能是無關的或冗餘的，Lasso特別有用。
  • 可以導致稀疏模型，其中只有一部分特徵對於預測有貢獻。

Elastic Net 回歸

Elastic Net 回歸是一種結合了L1和L2 正則化的混合方法。它在損失函數中添加兩種懲罰。當有多個相關特徵時，Elastic Net 特別有用。它包含兩個參數： η（如在Lasso和Ridge中）和α ，用於平衡給予L1和L2正則化的權重。

Elastic Net 回歸的關鍵特徵：

  • 平衡了Lasso和Ridge的特性。
  • 當數個特徵相關時，它的效果很好。
  • 通過調整 α 參數，Elastic Net 可以調節成像 Lasso 或 Ridge 回歸的行為。

選擇正確的正則化方法

選擇Lasso，Ridge和Elastic Net依賴於數據和手頭的問題：

  • Ridge 當不需要太多特徵選擇，或者預期特徵具有大致相等的重要性時，Ridge是一個好的默認選擇。
  • Lasso 如果特徵選擇至關重要，並且需要識別出最重要的變量，則首選Lasso。
  • Elastic Net 當有多個相關特徵，或者需要在特徵選擇和均勻係數減少之間進行平衡時，Elastic Net是理想選擇。

結論

正則化是機器學習中的一種強大工具，幫助提高模型的性能和可解釋性。Lasso，Ridge和Elastic Net是可以應用於各種回歸問題的多功能方法。理解它們的差異和應用是構建強大和準確預測模型的關鍵。

  • 7 min read

Go To Market (GTM) Strategy

In this blog post, I am writing to compare the Go-to-Market (GTM) strategies of two organizations in the banking industry: the Development Bank of Singapore (DBS) as an incumbent and Trust Bank as a disruptor. Over the last few years, advancements in technology and shifts in consumer behavior have led to significant changes. Established players like DBS face challenges from local digital banks that hold full banking licenses in Singapore, including Trust Bank, which was jointly launched by the FairPrice Group and Standard Chartered Bank.

A GTM strategy is a comprehensive plan that integrates the distinct functions of a bank to deliver a product and service to its target audience. The strategy provides a holistic view of how the various components of the bank and its ecosystem work together in an integrated, synchronized manner. In this comparison, I will analyze how effectively each bank’s GTM strategy supports its business objectives. I will also examine value creation, brand positioning, unique selling points, customer base, business model, products, pricing, sales channels, marketing levers, and more.

DBS – The Incumbent

As one of the largest banks in Singapore in terms of total assets, DBS has established trust and a strong reputation in the banking industry through a comprehensive GTM strategy. Regarding the "why" of their GTM approach, their primary objectives focus on maintaining their existing customer base on Digibank while expanding into new services such as the PayLah wallet. The bank creates value by providing a full range of services in consumer, Small and Medium-sized Enterprise (SME), and corporate banking. Its brand is positioned as the "World's Best Bank," recognized by Euromoney, and it has also been awarded the title of "Safest Bank in Asia." DBS offers a unique selling point by emphasizing its deep understanding of doing business in the Asia region, backed by insights and connectivity. Although the bank benefits from over 50 years of history, it is also constrained by its legacy systems and organizational bureaucracy, requiring multiple phases of digital transformation to remain competitive.

In terms of “who” on their target market segments, DBS's customer base is quite diverse, encompassing young expatriates working in Singapore as well as older individuals in the neighborhood who are not digitally savvy. The bank aims to fulfill various customer needs, such as providing a secure place to deposit monthly salaries and offering easy access to funds through ATMs and online banking for withdrawals and money transfers. Customers also seek a convenient way to pay utility bills and manage their finances, which includes budgeting, tracking expenses, and financial planning. Additionally, some customers require access to credit for purposes like securing a mortgage to buy a house or obtaining a loan to handle emergencies.

Regarding “what” on their messaging, DBS's value proposition is to "Make banking joyful" by leveraging digital technologies and seamlessly integrating themselves into their customers' lives. They aim to provide simple, fast, and contextual banking solutions along with differentiated customer experiences, achieved through the management of multiple customer journey maps. This has been enabled by their decade-long digital transformation journey, which involved investing in staff skills and re-architecting their infrastructure to be microservices-based, cloud-native, resilient, and scalable. Despite not offering the best interest rates on savings accounts, people are still willing to deposit money with DBS due to trust in their digital infrastructure. The bank's business model is primarily service-fee-based, with revenue streams coming from lending, investment services, and traditional banking functions. DBS offers a broad spectrum of financial products, including savings accounts, credit cards, mortgages, and other value-added services.

In terms of "where," their sales channels are diversified and well-established, including in-branch services, online banking, and mobile applications. Their marketing strategies rely on traditional channels such as billboards in SMRT stations, TV advertisements, and digital marketing. For premier customers, they maintain regular engagement through relationship managers and subject matter specialists. For the public, they offer active interaction and prompt follow-up to queries and feedback received via social media platforms like Facebook, LinkedIn, and Twitter, as well as through online communities on their corporate websites. Additionally, they regularly conduct customer satisfaction surveys and immerse themselves in customer experiences to continuously iterate on their journey roadmaps.

Regarding "how" they deliver experiences, DBS has adopted an ecosystem strategy that allows them to scale up their business in the large Asian market without incurring high customer acquisition costs. They have doubled their ecosystem-led customer finance lending through partnerships with companies such as ByteDance and Ctrip in China and Indonesia. They have gained good traction in onboarding large platform companies through their API (Application Programming Interfaces) suite, thereby digitally financing their extensive supply chains.

In addition, DBS is using machine learning models to generate hyper-personalized prompts that guide customers in making better investment decisions and in financial planning. They have also extended this digital transformation to other parts of the bank, such as private banking. This enables relationship managers to deliver hyper-personalized services to customers, resulting in increased customer transactions.

Trust bank – The disrupter

Trust Bank's GTM strategy focuses on user experience and offers digital-only banking services using the latest technology. The bank provides value by offering services with no hidden fees, competitive multi-currency exchange rates, and the security features to easily lock and unlock cards. Its brand is centered on being trustworthy, user-centric, and rewarding. The bank offers unique selling points through its partnership with the FairPrice Group, the nation's leading grocery retailer. This partnership allows customers to earn NTUC Linkpoints and other voucher rewards while spending with the bank. Additionally, Trust Bank offers budgeting tools, no-fee international spending, and real-time spending notifications.

In terms of “who” is their target market segment, Trust Bank primarily focuses on digital-savvy millennials and a growing segment of residents who frequently shop at FairPrice supermarkets.

In terms of “what” is their business model, Trust Bank operates on a no-fee model for its savings products, generating revenue through lending products such as credit cards, instant loans, and insurance. While it offers a more limited range of products compared to DBS, it compensates with a lower-fee pricing model.

Regarding “where” on the touchpoints, the bank's primary online sales channel is its mobile application, supported by a robust non-digital marketing campaign in NTUC supermarkets. This strategy raises awareness among a lot of grocery shoppers, who may later consider using Trust Bank because of its unique rewards points system. Those who choose to open an account could become loyal customers and advocates for Trust Bank, further promoting the bank through referrals. To increase customer engagement, the bank's marketing efforts also include lucky draws for Tesla cars.

Compare and Contrast

Both DBS and Trust Bank offer value creation through their banking services. DBS focuses on variety and maintaining its market dominance, while Trust Bank emphasizes innovation and user experience. In terms of brand positioning, DBS has an established reputation and uses the tagline "Live more, bank less," promising to deliver simple, seamless, and invisible banking. On the other hand, Trust Bank's tagline, "The digital bank for everyday use," is less well-known.

DBS's unique selling points are based on its deep understanding of business in the Asian region and the wide variety of services it offers under one roof. Meanwhile, Trust Bank, with only one year of history, focuses on providing digital-first services.

In terms of target audiences, DBS caters to a more diversified customer base, ranging from young working expatriates to the elderly in the neighborhood, whereas Trust Bank is better suited for digitally savvy millennials.

The business model of DBS is more traditional and relies on charging service fees, while Trust Bank adopts a completely different strategy, emphasizing transparency and no fees.

The distribution channels for DBS include automated teller machines, call centers, and physical branches, as well as cash handling, all of which contribute to higher operating costs. In contrast, Trust Bank benefits from an app-based model, providing a streamlined user experience that is more cost-effective to operate. Additionally, DBS incurs higher marketing costs due to traditional methods like SMRT billboards, whereas Trust Bank leverages its partnership with NTUC to gain word-of-mouth marketing through supermarket reward points.

Critique

DBS's GTM strategy is more robust than Trust Bank, but it is under threat due to its pace of digital transformation. Trust Bank has the advantage of being more agile, as it is not encumbered by legacy code that could inhibit innovation. Its GTM strategy aims to provide a superior customer experience at an exceptionally low cost. However, Trust Bank could face challenges related to scalability and regulatory issues as it expands in the Asian region.

Overall, DBS's success stems from a redesign of their operational approach. To ensure that customers are at the heart of the banking experience, they have trained their staff in customer journey thinking, which includes elements like customer discovery. They have reorganized to function as a horizontal organization, with cross-functional teams across the bank collaborating to deliver value and enhanced customer experiences. These efforts have resulted in improved customer satisfaction.

Meanwhile, Trust Bank has also adopted a customer-centric approach in their business, with a greater ability than DBS to act on customer feedback for the development of features and products. Their go-to-market speed is faster. When Trust Bank was launched, it was able to deliver an exceptional onboarding experience. The account opening process was much quicker than that of DBS; their seamless digital approval process enabled Trust Bank to complete the onboarding for savings accounts in under three minutes. This is a significant contrast to DBS, where customers typically experience a much longer onboarding process filled with friction, such as the need for address verification and physical card activation.

If Trust Bank can continue to iterate based on customer feedback, they could achieve greater success in penetrating retail digital banking market segments than DBS, thanks to their agility in reacting and adapting to change. For DBS to respond more quickly to the threat posed by challenger banks like Trust Bank, they will need to continue their digital transformation journey or risk becoming irrelevant due to an inability to adapt to changing customer needs and market conditions.

  • in zh
  • 1 min read

進入市場（GTM）策略

在此博客文章中，我將撰寫比較兩家銀行業組織的進入市場（GTM）策略：作為現成者的新加坡發展銀行（DBS）和作為破壞者的Trust銀行。在過去的幾年中，技術的進步和消費者行為的變化導致了重大的變化。像DBS這樣的現有參與者面臨來自在新加坡擁有完全銀行執照的多家當地數位銀行的挑戰，其中包括由FairPrice集團和渣打銀行聯合發起的Trust銀行。

GTM策略是一項綜合計劃，將銀行的各個不同功能整合起來，以將產品和服務提供給指定對象。該策略提供了一種全面的觀點，說明銀行及其生態系統的各個部份如何以一種整合、同步的方式一起運作。在這次的比較中，我將分析每家銀行的GTM策略支持業務目標的有效性，並且還將對價值創造、品牌定位、獨特銷售點、客戶群、業務模型、產品、價格、銷售渠道、行銷個案等進行檢驗。

DBS - 現有者

作為新加坡資產總額最大的銀行之一，DBS通過全面的GTM策略在銀行業建立了信任和良好的聲譽。在他們的GTM策略的“為何”方面，他們的主要目標專注於維護他們在Digibank上的現有客戶群，同時擴展如PayLah錢包等新服務。該銀行通過提供個人、中小型企業（SME）和公司銀行接軌的全方位服務來為客戶創造價值。其品牌被定位為“全球最佳銀行”，得到Euromoney的認可，也同時被評為“亞洲最安全的銀行”。DBS通過強調其對在亞洲地區經營業務的深入理解，並得到洞察和連接力的補助，提供了獨特的銷售點。儘管該銀行從50多年的歷史中獲益，但它也受到其遺留系統和組織官僚主義的約束，需要經過多個階段的數位轉型以保持競爭力。

在他們的目標市場區段上的“誰”，DBS的客戶基礎相當多樣化，包括在新加坡工作的年輕移民，以及社區中對數位細節抱持怀疑態度的老年人。該銀行的目標是滿足各種客戶需求，例如提供一個安全的地方來存放每月的薪水，並通過ATM和網路銀行提供輕鬆的資金存取，用於取現和匯款。客戶還在尋找一種方便的方式來支付公用事業費和管理他們的財務，這包括預算、追蹤開銷和金融規劃。此外，有些客戶需要獲得信貸，用於購買房屋的抵押貸款或應對緊急情況的貸款。

關於“哪些”有關他們的信息，DBS的價值主張是通過利用數字技術和無縫地將自己融合到客戶的生活中來“使銀行變得愉快”。他們的目標是提供簡單、快速和符合環境的銀行解決方案，並且還通過管理多個客戶旅程地圖來達成與他們區別的客戶體驗。這得益於他們長達十年的數位轉型之旅，該旅程涉及對員工技能的投資以及將其基礎設施重新架構成微服務為基礎、原生雲、韌性和可擴展。儘管在儲蓄帳戶上並未提供最佳的利率，但人們仍然願意將錢存入DBS，因為他們對其數位基礎設施充滿信任。該銀行的業務模型主要基於服務費，收入來源來自貸款、