Search
Contact UsAccess the On-Demand Webinar Playback of “How AI-Enabled Super-Agents Improve CX” with Kate Legget, Vice President and Principal Analyst Service Application Development and Delivery, Forrester Research, with Steve Nattress, Director, R&D and Jacki Tessmer, Vice President – Product Marketing, Enghouse Interactive.
Blog #2 of 4: Seeking the Root Cause and Affecting the Right Outcome
For the most part today, AI’s ability to look at data-points to understand what each customer is concerned about, or has issues with, is currently very much uni-dimensional.
To address that, there are limitations to be overcome and diverse approaches to be considered. Most AI-based platforms or processes leverage a finite and discrete data set from which to extrapolate ‘understanding’. They typically focus on repetitive requests, cataloging and assessing the actions taken or the responses given, when helping to resolve those situations. Then, it uses this limited ‘understanding’ to try to provide a very narrow range of suggested actions to help guide agents in future situations:
i.e. when to ask the customer for clarification, or to route customers to a human agent with enough conversational insight to know if a generalist or a specific subject matter expert (SME) would be best, or when to trigger a Knowledge Base (KB) search.
This works, but only when using the right dataset.
In this still-emerging field (more on the current state, discussed later), various approaches can be used; from the esoteric and academic including Deep Learning Neural Networks to the more practical and useable Natural Language Processing (NLP) and the Reinforcement Learning and Control approaches.
Interesting Reading(1):
Looking at these different approaches, it’s been seen that various types of deep learning are highly capable at discerning correlations in huge datasets that they work with. However, when trying to develop or understand inferences and other higher-level concepts, deep learning generally falls flat, which has prompted researchers to combine various types of approaches to find what works best.
From a practical perspective, it’s been found that natural language processing (NLP) techniques can be used to more accurately analyze and assess the customer comments and classify them. Enghouse’s experience has shown that by leveraging semantic analysis, and industry-specific linguistics, terminologies, phraseologies and other proprietary algorithms and processing, NLP can better identify what the customer is experiencing and then, it can propose a possible solution from a specific set of options.
A typical example would be to use that extracted insight to interface with an associated knowledge base and then recommend a process, script, or document to help deal with the situation or need at hand. In addition, by using reinforced learning with the knowledge base, over time the proposed answers will significantly improve their relevance as more data (both positively and negatively ranked) is categorized.
The Reality Today: Limited Information and Comprehension = Probable Options
The broad promise of AI has existed for over 70 years when the first computer and computing processes were developed by Presper Eckert (Univac, late ’40s) and then furthered by Alan Turing (Turing Machine, 1950)…but yet, they remain unfulfilled. AI struggles with broad comprehension and being able to discern overarching intent and meaning.
The limiting factor is not just the need to use large enough datasets, it’s AI’s inability to contextualize the information and understand the interrelationships and interplay of words and inferences – in the way that humans do.
The Limitations: Data Needs Context
In this film clip(2) from «La Femme du Boulanger (The Baker’s Wife)« we see and hear that the baker is scolding his cat while speaking lovingly to his wife …yet we humans understand that the baker is actually scolding his wife through his comments to the cat while deferring to his wife.
AI today does not.
It sees 2 people, at a table, can possibly infer their gender, and notes that there is some sound from the image.
Why the Difference?
It’s because humans don’t analyze situations from the perspective of discrete data points. The reality is that they have life-long learnings from which to pull – from maternal/paternal guidance and nurturing, situational learning; academic travails; 3rd-party experiences; conflicts survived, joyous moments enjoyed, stressful situations passed, discipline received, rewards achieved, unethical situations lived, etc. As a result, humans inherently know how to, directly/indirectly and inferentially assess every situation, either cognitively or sub-consciously. Leveraging these learnings, they can determine what something means along with what the potential ramifications are now and possibly in the future… As a result, each person when seeking to resolve a situation, sees it from a unique perspective (multi-variate data) and makes their decision differently (diverse factor analysis)… and generally, they may suggest different strategies to resolve the situation (variable approach). AI today will try to choose a standardized approach to solve potentially highly diverse needs.
The Promise: With context, AI seeks to improve – more quickly than ever
Ultimately, the holy-grail will be achieved where AI will be able to propose a range of scenarios and agent support scripts that are developed ‘on the fly‘ in real-time – using the customer’s comments – to help best address the situation, without reliance on a narrowly-focused standardized approach.
But to achieve that objective and be transformative in their effectiveness, AI-platforms must broaden the range of information they gather while enhancing their analytical focus. They need to aggregate data from across the complete spectrum of CX activities, from online with context-aware assistance, to chat and messaging solutions using chatbots and live agents, to voice with intelligent IVR, and email with linguistic routing and intelligent auto-response and finally to analytics with AI insights. Then assess how all these data-points interrelate and what actions were taken, how customers responded while considering how quickly these issues were resolved, and how well the customers rated the overall experience and whether they would make a repeat purchase.
It’s doable. But right now, this is still best done in support of agents, who can make the final decision of what to say to the customer and how.
Watch for the next blog, coming October 21st:
Constant Learning Improves AI Recommendations
Help make sure your organization delivers the customer experience that exceeds their expectations. Doing so, will transform your contact center from a cost-center into a revenue generator.
Access the On-Demand Webinar Playback of “How AI-Enabled Super-Agents Improve CX” with Kate Legget, Vice President and Principal Analyst Service Application Development and Delivery, Forrester Research, with Steve Nattress, Director, R&D and Jacki Tessmer, Vice President – Product Marketing, Enghouse Interactive.
See how your organization can benefit from AI-Enabled Super-Agents