In a previous article, we looked at the considerable potential for chatbots in banking strategy. In this follow-up, we’ll walk you through our recommended approach to implementing impactful personalized engagement with bots, including:
- Use case selection
- Our recommended journey
- A platform-based approach to conversational banking
- Launching implementation
In-the-moment customer insights
Digital banking solutions struggle to engage customers with insights because they require users to discern the information themselves. Chatbots may be able to solve for this by offering analysis and advice without asking customers to do the work. For example:
Spending: Conversational bots may be able to reach out to customers with insights on spending decisions they’re about to make (banking customers have been asking for this functionality for a while).
Investment performance: Bots could serve as a medium for real-time financial advice, as well as facilitate a purchase or a sale through a couple of chats instead of asking the user to enter a lot of data, thus improving UX significantly.
Easier transactions: Bots can help customers perform high-frequency tasks such as bill payments and fund transfers via a brief chat, rather than asking them to review several pages of financial transactions.
Advice when you need it: Conversational bots are in the unique position of being able to reach out to customers proactively at just the right moment to advise them on purchasing a house, cross-selling a housing loan, researching house insurance and even purchasing household goods.
Quick service recovery: Conversational bots can react much faster than most other channels when it comes to service recovery. They can ping a user instantly and start a conversation, then either offer a solution or arrange for a callback—all while ensuring that customer engagement happens quickly.
Our recommended journey
It’s easy for most banks to think of customer service as the best place to implement conversational bots, but the real potential for chatbots lies in customer transactions, analysis, advising and service recovery.
Taking a platform approach
As use cases evolve and more are conceptualized, banks will realize the need to build a catalogue of bots that serve different needs. At Mindtree, we visualize a conversational app marketplace, with bots for various business functions designed rapidly to conform to enterprise standards.
To make this happen, companies must invest in a robust bot platform. This will enable them to implement a factory model for building conversational bots in a scalable, agile way that meets all the necessary regulations.
Key Building blocks for a bot platform
Interaction layer on conversation channels: This involves customizing the front-end experience on Facebook Messenger, Skype, Slack, WeChat and Chatter—to name just a few. Each channel offers different features that combine conversation with the app experience.
Bot framework: This serves as the heart of the platform by:
- Orchestrating the entire conversation
- Controlling the transaction flow across the interface layers and their respective frameworks or connectors; AI framework; enterprise systems through APIs for transaction and automation (e.g., fetching goal performance data); third-party data (e.g., social feed, stock performance check); analytics engine for peer comparison
- Ensuring that nonfunctional requirements are catered to, including security and state management
AI framework: This helps orchestrate understanding of—and interaction between—NLP engines including LUIS, Wit.AI and Pandorabot. Each has a different configuration, integration and training options, at the same time enriching the AI through continuous learning.
Data analytics engine:
- Use a sandbox to test and validate hypotheses (e.g., what advice should be given to which customer).
- Build a view of the customer to increase understanding of users. This can be based on internal banking data or third-party data, such as social media, demographics, peers and so on.
- Run algorithms for peer comparison and advice, among other analytics.
API layer: APIs enable integration across layers and ensure that existing investments are leveraged. They also allow for seamless integration with the interface layers and their respective frameworks or connectors; the AI framework; the enterprise itself for transaction and automation (e.g., pulling goals performance); third-party data such as social feeds and stock performance check; and an analytics engine for peer comparison.
Workshop: First you’ll identify use cases that need to be implemented. These should range across functional areas, including customer service, transactions and payments, customer insights, service recovery and other roles. Then select the most relevant use cases where enabling bots would have the greatest impact.
Pilot and training: Once you’ve selected use cases, it’s time to pinpoint corresponding training sets. This is followed by an evaluation of different channels and AI engines, thereby letting you define the architecture.
Next you’ll develop a proof of concept or a pilot plan with the proposed architecture, so you can implement some of the building blocks to validate selected use cases. After everything has been planned, including integration with the required APIs for use cases, the pilot can be executed.
The pilot follows the “load, train, validate and retrain” cycle:
- The training set is loaded.
- The AI models are trained with training sets.
- The models are validated during the pilot.
- The models continue to be retrained and refined to achieve the desired results and thresholds.
Platform and marketplace: Once the pilot is complete, the architecture, base models and building blocks are frozen. This enables you to scale the framework to start building a conversational app marketplace.
At Mindtree, we bring all these pieces together. When you’re ready to start the conversation about transforming your digital banking, contact us.