Observe.ai, a conversation intelligence platform, has released a generative AI suite meant to improve agent performance, as well as a contact centre large language model (LLM) with a 30-billion-parameter capacity. The business claims that its patented LLM, in contrast to models like GPT, is trained on a massive dataset of actual contact centre interactions.
Observe.ai said that the unique value of their model is in the calibration and control it gives consumers, despite the fact that a few comparable products have been introduced lately. Users are able to fine-tune and adapt the platform’s model to their own contact center’s needs.
According to the business, their LLM has been trained on several contact centre datasets, making it capable of AI-based activities (call summarising, automated QA, mentoring, etc.) designed specifically for contact centre personnel.
Observe.ai’s LLM is a part of their generative AI portfolio, which aims to improve agent performance across the board, from phone calls and chats to emails, complaints, and everyday discussions with customers.
Observe.AI thinks these additions will allow agents to provide higher-quality service to customers.
Our LLM has been rigorously trained using data collected from actual conversations in call centres. Swapnil Jain, CEO of Observe.AI, told VentureBeat that the company used a large corpus of data points gathered from the hundreds of millions of chats handled by the company over the last five years for the training process.
Hundreds of curated instructions covering a wide range of activities that are directly relevant to contact centre use cases were included in the instruction collection, which Jain emphasised for its significance and quality.
He said that the LLM’s capacity to provide the correct and contextually relevant replies demanded by the business was enhanced by the rigorous approach to dataset curation.
The business claims that their contact centre LLM is 35% more accurate at conversation summarising and 33% more accurate at sentiment analysis than GPT-3.5 in early benchmarks. According to Jain, these numbers will rise as a result of ongoing training.
Furthermore, the LLM was trained only on data that had been censored, which eliminated any possibility of training on sensitive material. Observe.AI emphasises that it use redaction methods to prioritise client data protection while still using generative AI’s capabilities.
Getting rid of delusions to reveal real information and background
Around 70% of organisations across sectors, per Jain’s estimates, are investigating the potential advantages of generative AI. This is especially true in areas like customer experience, retention, and revenue growth. The heads of contact centres are among the early adopters keen to use these game-changing tools.
Jain, however, thinks that generic LLMs have several problems that prevent them from being fully useful in call centres, despite their potential.
Limitations in comprehending human dialogue and real-world situations, a lack of specificity and control, and a lack of answer discrimination are all examples of these difficulties. Therefore, he said, these too general models, including GPT, often produce mistakes and confabulations, commonly known as “hallucinations,” making them inappropriate for use in professional contexts.
Generic models are educated using free data found on the web. “These models have to deal with transcription errors caused by speech-to-text models, and they don’t learn the nuances of spoken human conversation,” added Jain. It’s possible that they excel at broad tasks like conversation summarization but fail to take into account the specifics of contact centre talks.
Jain said his firm had overcome these obstacles by factoring in five years of cleaned and relevant data. In order to train the model for contact center-specific activities, it acquired data from hundreds of millions of consumer encounters.
We know what’successful’ client experiences are and how to provide them in the real world. Jain said, “Our clients can then fine-tune and tailor this to the specific needs of their business.” Our method offers a comprehensive framework for call centres to fine-tune their equipment and make sure their results match their predictions. This is inherent in every ‘glass box’ AI paradigm that provides full visibility and inspires confidence in the programme.
He also noted that the company’s new generative AI package gives agents an advantage across the board in dealing with customers.
Since the Auto Summary feature allows agents to focus on the customer by reducing post-call tasks and ensuring the quality and consistency of call notes, it works in tandem with the Knowledge AI feature to ensure quick and accurate responses to customer inquiries.
Agents get rapid, actionable, and evidence-based coaching through the Auto Coaching tool after every client engagement. This helps agents develop their skills and strives to enrich their learning experience beyond what they get from their normal coaching meetings with a supervisor.
A new standard for LLMs in call centres
Observe.ai believes that a major improvement has been made since their unique model is more reliable and applicable than GPT.
All personally identifiable information (PII) and sensitive customer data is removed from the data used for our LLM’s training. In 150 million occurrences over 100 million conversations, we prevent over-redacting sensitive material with less than 500 reported faults, which Jain called “exemplary” for the industry. Maximum information is retained for LLM training while confidentiality, privacy, and regulatory compliance are all maintained.
He said that the firm complies with all relevant regulations on the storage of client data, including data created by the LLM. Data encryption and account-specific identification are both guaranteed by assigning each client their own separate storage partition.
According to Jain, we are now living through a watershed moment in the development of generative artificial intelligence. The contact centre sector, he said, is replete with repetitive chores, but he is certain that generative AI will enable human talent to do the work ten times faster and more accurately than before.
Jain predicted that successful industry disruptors will work towards developing a “fully controllable; trustworthy with complete visibility into outcomes; and secure” generative artificial intelligence. We’re committed to developing AI that is trustworthy, dependable, and consistent so that it may be used to enhance the performance of human talent. Our goal is to develop AI that frees up human brainpower for higher-order tasks like innovation, strategic planning, and satisfying customers.