Synechron helps banks, fintechs and payment processors across the globe to build efficient solutions in cards and payments. It provides end-to-end services in consulting, design, development and maintenance of card issuing & acquiring platforms, transaction switching, payment gateways, real-time payments, digital payments, and open banking solutions. They have built a suite of AI-powered accelerators to build payment solutions faster.
Amongst the work that Synechron does in the industry, it provides chatbox integration for customer support. As banking moves more and more from an in person experience to online, chatbots become more vital to the banking experience. While skepticism still remains high, how can these chatbots provide an equally good experience for customers
Go deeper with GlobalData
Robert Prendergast (RP): How does Synechron Solutions approach language learning in the use of chatbots?
Ryan Cox (RC): “Large language models such as OpenAI, Anthropic, and Gemini use a lot of data to train their models for use by the Chinese, Spanish, and French. We tend to work with these models when building solutions in major languages as they tend to have the most reliable output instead of training the model ourselves.
“To ensure quality, we run tests and work with native speakers to see how the language models perform. We then use prompt engineering to better direct the style, tone, and interaction within the chatbots, using fine-tuning techniques to ensure users receive the most appropriate and fluent output.”
RP: How do you appeal to customers who are sceptical towards the use of AI and Chatbots?
RC: “Understanding a customer’s goal and the value they are trying to unlock is critical. If the customer has a clear objective, we can see where AI fits and what type of AI solution would be most suitable. We then guide our clients to the most appropriate solution. Most customers are sceptical of AI and Chatbots because they don’t have a clear understanding of use-cases, our team can help customers identify AI solutions to fit their needs.
“We build trust with our customers by listening to what they want and showing them the relevant projects we’ve completed using AI. We are transparent about AI’s limitations and where we suggest keeping human oversight.”
How well do you really know your competitors?
Access the most comprehensive Company Profiles on the market, powered by GlobalData. Save hours of research. Gain competitive edge.
Thank you!
Your download email will arrive shortly
Not ready to buy yet? Download a free sample
We are confident about the unique quality of our Company Profiles. However, we want you to make the most beneficial decision for your business, so we offer a free sample that you can download by submitting the below form
By GlobalDataRP: How does Synechron approach building bots for financial institutions?
RC: “Discovery phases with proof of concepts are essential. It’s important to understand what our client wants to achieve and iteratively build it to ensure feedback at every stage. Clear and transparent communication with key stakeholders as the project plan is built and implemented is also integral.
“In parallel, we work closely with the Synechron global regulation practice and client regulation teams to ensure compliance with relevant regulations, data privacy, and security standards.”
RP: How does your system make sure that people can find what they are looking for and answer their questions without the need to interact with a real person?
RC: “Our solutions are built using knowledge bases and data repositories. Our solutions can answer questions based on verified data and facts, ensuring high-quality answers. Tailored data sets solve the problems users experience when using public large language models, such as explainability and observability.
“We use vector databases when working with a large number of documents so that our solutions can respond to queries quickly and the output remains explainable. Our vector databases can identify the source of the data and provide that source reference to the user. This explainability helps during the testing and building process and in production, as we have confidence in the output, which lowers the client’s need for human interaction within their chatbots.
“Our model validation framework uses quantitative metrics like BLEU and ROUGE to evaluate AI model performance over time. These metrics help us gauge the model’s accuracy and alignment with human-like understanding, ensuring its reliability and effectiveness.
“To ensure our solutions continually evolve and improve, we actively monitor user feedback and analyse interaction patterns throughout the process. This ongoing analysis allows us to dynamically refine our chatbot’s knowledge and response quality. We have greater confidence in areas with high-quality, explainable data, where user feedback is overwhelmingly positive, and our model validation framework shows strong performance. In these instances, we can reduce our reliance on human oversight for interactions with internal users. However, we remain committed to closely monitoring performance and feedback to ensure quality is upheld before extending these practices to external customers or clients.”
RP: How are bots tailored for each company?
RC: “We work to understand the business, the technology, the value, and the user journeys to help us provide tailored solutions for each of our clients. We integrate and ingest company data, adjust the data sources used and get an understanding of how important each piece of data is to ensure the AI reads the data in order of importance.
“We also incorporate client branding and tone as well as terminology to ensure that the output aligns with the company’s voice.”
RP: What are the current limitations in building these bots?
RC: “Data is often a challenge before you even start thinking about building an AI solution. High-quality data, with a single source of truth, and governance around that data are key to having a high-quality chatbot that can be trusted.
“It’s also critical to ensure that your model is free of data bias, which has its own set of challenges. If a model’s input and training is biassed, it can lead to unfair or discriminatory outcomes and even have a regulatory impact.
“Seamless context switching and maintaining conversational state across multiple topics or services is challenging. Ensuring transparency, explainability, and user trust is key. Handling complex, multi-turn financial transactions requiring deep domain expertise can be challenging. The models used need to understand this context in depth to ensure the output is correct. That is why building tailored models for your use-case is incredibly important.”
