Proto, a global leader in AI customer experience (AICX) solutions, finishes Q1 2024 with the release of its new AICX Platform with 15+ GPT features and new usecases driven by clients.
This release comes after a fundamental redesign of how AI customer experiences can be enhanced safely with large language models, rather than simply plugging ChatGPT into chatbots. The Proto AICX Platform also includes novel efficiency features, such as its own chatbot that can assist with customer support automations and answer questions about performance results.
The new Proto AICX Platform, built on client feedback, offers four modules — chatbots, livechat, ticketing, and CRM — all united with a consolidated dashboard and messaging inbox:
Proto's enterprise and government clients are completing their migration into the new platform, with a contracted growth increase of 68% thanks to the AICX Platform’s advanced capabilities. With six new enterprise and government clients in the last month alone, Proto is well-positioned for future growth.
Balancing GPT with proprietary AI for quality, languages & privacy
While large language models like ChatGPT offer efficiency gains for customer experience automation, Proto has chosen a more nuanced approach for its chatbots. Recent missteps by chatbots from DPD and Air Canada underline the risk of relying too heavily on enduser messages for generative AI training data.
Even though GPT is powering efficiency gains in its AICX Platform, Proto is mindful of its chatbots' reliance on LLMs alone. This is due to three reasons:
- Chatbot quality. It became all too clear this year from the offensive language and wrongful information of the DPD and Air Canada chatbots, respectively, that such errors can cause both reputation and financial damage. Proto limits its GPT-based chatbots to training data from client-provided documents, knowledge bases, websites, and qualified chat transcripts — as well as balancing generative AI with rule-based dialogue for essential information relating to payments and legal policies. Doing so mitigates against the risk of malicious inputs by endusers and information hallucinations by the chatbot.
- Local and mixed languages. Large language models like ChatGPT are trained primarily on English data. As a result, they can be deficient in handling local and mixed languages such as Tagalog and Taglish in the Philippines. In this example, Proto's proprietary proLingual™ engine fills this multilingual gap with an average 12% higher accuracy than comparable natural language processing engines for text chatbots, and a 5% lower word-error rate for voice chatbots.
- Data privacy. To ensure compliance with the letter and spirit of evolving national data privacy legislation in the wake of generative AI, Proto switches by default to its proLingual™ engine from LLMs in dialogues that may contain personal information in order to guard against potential errors in data anonymisation. This disallows the risk of personal identifiable information (PII) such as names, account numbers, and addresses being fed into LLMs for retraining (which would effectively expose this PII to the public). For an additional layer of security, Proto’s GPT infrastructure is on-premise or cloud hosted on Microsoft Azure, ensuring a closed model.
Balancing out Proto’s GPT dependencies are the platform’s AICX Add-Ons, offering complementary capabilities for data controls, custom analytics, and local language understanding:
Clients driving new usecases beyond customer support
With Proto's rapid training and integration capability, clients and partners are testing new AICX applications, such as internal knowledge sharing for employees and indoor wayfinding for travellers. This expansion extends into the domain of analytics, where OJK Indonesia and the Bank of Ghana — the financial regulators of their respective countries — are integrated social media sentiment data into the Proto AICX Platform to gain a holistic, real-time view of correlated public and private complaints in order to rapidly detect market misconduct.
Proto’s government vertical is also undergoing a major usecase expansion, where the new Internet Transaction Act in the Philippines has mandated the integration of chatbots and case management systems between government agencies to provide a one-stop service for citizens in the interests of more effective consumer protection. This integration is overseen by Proto’s client, the Department of Trade & Industry of the Philippines, with strategic insights and guidance from the supervisory technology experts at the University of Cambridge SupTech Lab.
This progression of the AICX Platform and the trust of its clients is propelling Proto’s measured application of generative AI into customer experience, with upcoming global deployments beyond Proto’s traditional emerging market base. These new deployments build upon best practices learned from multilingual customer experience automation in challenging markets with a lack of local language training data and expanding AI readiness.
Test Proto’s client chatbots deployed into countries such as Malaysia, Philippines, Rwanda, and Zambia:
.png)
While the new platform and growth is boosted by GPT, Proto remains committed to its proprietary AI engine for local and mixed languages, in order to ensure long-term data privacy, customer inclusion, and chatbot quality.
Moreover, this new technical capability will push beyond the enterprise usecases to continue delivering the societal benefits of financial inclusion, consumer protection and countering disinformation. This is line with Proto’s mission for conversational AI to include everyone — regardless of gender, ethnicity, education or language.
