Integrating Human in the Loop AI with customer support

Author: Nykki Yeager is a customer-centric consultant who writes about customer support, customer success, and management.

---

Artificial intelligence (AI) has captured our imaginations for years, at least as early as 1872. But AI isn’t just in our stories anymore. Now, it’s a daily reality.

With Alexa and Siri, many of us are using AI in our personal lives. And from spam filters to more complex and specialized products, almost all of us are using some form of it at work. Customer support teams are leading the innovation charge here: one survey from Tata Consultancy Services revealed that 31.7% of major companies are currently using AI in customer service.

With many companies already investing in AI for their workforces and more new companies building AI-focused solutions popping up seemingly every day, it’s no surprise that customer support teams are a little bit worried. What’s the role that people will play in the future of support? We’ll answer that question and more in this article about humans in the loop.

• An Important Introduction  
• What Is Human in the Loop AI?
• What Is Human in the Loop Data Analytics?
• Why Have a Human in the Loop?
• Will Humans Still Have a Job to Do?
• Machine Learning in Customer Support
• The Role of Humans
• Managing the Systems of Support
• Don’t Fear the Technology, Embrace It

An Important Introduction  

Before we get deep into the specifics, we need to cover the basics. Put simply, AI is the capability of a machine to imitate intelligent human behavior. Machine learning (ML) is a specific technique commonly used to train an AI system.

At its core, the entire machine learning process uses statistical techniques so computers can receive data, identify patterns, make decisions, and “learn” by getting better throughout the process (also known as active learning). A continuous feedback loop that specifies whether decisions were right or wrong helps machines learn and adjust the approach taken with future data points.

What is Human in the Loop AI?

Human in the loop AI is a framework that uses both human and machine intelligence. Human in the loop (HITL) means that humans are involved in the training, adjusting, and testing of machine learning algorithms. People provide training data to the machine, telling it which labels apply to which data points. People then fine-tune the algorithm, for example, by introducing new edge cases or new categories to the data. Then they test the model by scoring the outputs, essentially telling the machine if its decisions were right or wrong. Over time, the model gets more and more accurate.

While machines excel at managing extremely large sets of data, people are great at making nuanced decisions. If you’ve ever clicked on the street signs in a captcha image, you’ve actually trained a machine learning model on what a street sign looks like, and not just an ideal image of a street sign in a noiseless vacuum, but in different colors, from different angles, obscured by other objects — in the real world. The more examples a machine is given of what a street sign looks like in different settings, the better it becomes at recognizing a street sign on its own.

What Is Human in the Loop Data Analytics?

Human in the loop data analytics is the required involvement of a human in the process of data analysis. Machines can learn how to analyze qualitative and quantitative data and make assumptions based on their findings; however, the involvement of humans ensures that the findings made by AI are accurate — especially when it comes to qualitative data analysis. Despite natural language processing being highly intuitive, humans can still identify issues in the analysis process.

When AI performs a sentiment analysis on human feedback, training data that’s been programmed by data scientists and engineers allows the machine to make predictive assumptions about the data; however, as skilled as these machines are through active learning and transfer learning, there will always be a need for human involvement in data analysis to ensure that the results are accurate. Since machines lack the emotional intelligence of humans and the ability to understand nuances, humans are able to analyze the data to check for inaccuracies and make adjustments to their model if needed. 

Why Have a Human in the Loop?

Having a human in the loop is beneficial because they serve as quality control, help AI avoid bias, augment rare data when new situations arise, and more. Here is a full list of why we need humans in the loop.

• Quality control: It’s one thing to build a highly intelligent ML system, but the need for human quality checks is incredibly important to ensure humans on the other end of the network are truly satisfied with their experience. Humans (e.g. a machine learning professional) are then able to identify areas of opportunity and fine-tune their machine algorithms to be better and continue improving, with human input and intelligence driving these decisions based on existing human annotations. 
• Avoid bias: Artificial intelligence can develop bias based on historical data. Humans are needed to correct any learned biases and ensure the integrity of the machine learning applications. 
• Augment rare data: As much as a machine can grow and become more intelligent through active learning, there will always be new situations (new data) coming in that the machine is unfamiliar with or hasn’t been programmed into the machine learning algorithm yet. Humans are needed to step in and address these new and complex issues in order to properly resolve them in a way that makes sense for the customer and the business.
• Training data: HITL helps to create training data by allowing the trainee to fully immerse themselves in the event or process on an ongoing basis. This transfer of skill and intelligence turns into real world machine learning (also known as human in the loop machine learning).
• Speed: Because human in the loop machine learning uses the strengths and skills of both humans and machines, the speed in which AI can progress is at its peak. It’s possible to shop models faster and serve customers quicker than ever before.
• Accuracy and reliability: As human intelligence is inputted to ML systems, it increases the overall performance and reliability of the machine learning systems. Humans supervised machine learning is needed to ensure human level precision is consistent throughout.

Will Humans Still Have a Job to Do?

The short answer: yes. The longer answer is that our jobs will be different. We’ll work together with machines to continue serving customers faster and better than ever. And that’s the premise behind human in the loop.

Machine Learning in Customer Support?

For customer support teams, machine learning can help with many aspects of daily work. For example, when a customer asks a question a machine learning model can determine the best answer and either serve it to the customer directly or recommend it to a support agent, decreasing time spent looking for a solution and increasing accuracy.

However, as described earlier, machine learning models need to be taught. Before a model can give reliable answers it needs to consume a lot of historical customer conversations. ML models are limited by the quality of this historical data. If the conversations that are being used aren’t thorough and accurate – the ML output won’t be either. The old adage of “garbage in, garbage out” certainly applies here. And this also means that support agents are still much better in nuanced or one-off scenarios, long-tail issues, and new issues that haven’t been addressed before.

Machine learning solutions can also analyze conversations to add tags and categorize feedback to help save time and improve accuracy when reporting on customer feedback, whether you’re looking at high-level trends or diving into specific issues. When machines help with the legwork of tagging tickets, then support teams can spend more time socializing trends with product teams and working on taking action.

The human in the loop framework refers to machine learning, but humans also have a role to play when working with other applications of AI, like chatbots.  There are times when a customer just wants a fast, easy, correct response. AI can do that. Then there are times when a customer’s issue is so nuanced, or they’re so emotionally escalated, that they need or want to speak to a human. Very transactional, straightforward inquiries are perfect for chatbots, but bots should be set up to pass more complex conversations to people.

Key Points:

• ML saves time on finding solutions and improving accuracy when reporting on customer feedback.
• ML needs to consume a lot of quality historical customer conversations to excel.
• Bots should be set up to pass complex conversations and new issues to support agents.

Learn more about AI and how it can benefit your customer support team >>

The Role of Humans

The common theme throughout all of this is that with more work being done by machines, the role of customer support teams becomes even more elevated.

Leveraging automation for routine tasks, support professionals can spend more time giving personalized attention to the inquiries that are routed to them. With the simpler inquiries taken care of by the technology, the inquiries that make it to a human will be the most complex or escalated, and therefore even more important to handle with the grace that only a highly emotionally intelligent and knowledgeable support professional will possess. Not only does this highlight the important skills of people in support, but when you take away more of the transactional volume, customer support teams can feel more challenged, more skilled, and more fulfilled.

And when they’re not wading through large volumes of inquiries and operational inefficiencies, teams are freed up to move into more strategic, proactive roles, where they can look at the big picture and identify opportunities to improve the customer experience, both in support as well as across the business. By using technology to collect and analyze data on customer insights, teams can amplify the voice of the customer and pass more complete feedback to other teams. This increases the visibility, influence, and impact of support teams in their companies and across the industry.

Another big role of humans will simply be to be there. For the foreseeable future, customers will at least want the option of speaking with a live agent. Think about the level of frustration you have when you’re in an IVR phone tree and all you want to do is talk to a human, but you can’t get a hold of anyone. Customer support professionals need to be available, and businesses need to be wary not to replicate poor IVR experiences in chat and other digital channels.

Key Points:

• AI handles common (repetitive) tickets so customer support teams have more time to think strategically about inefficiencies, pass VOCs to other teams, and use their skills to address more complex issues.
• Customer support teams experience more job satisfaction and less burn out with the help of AI.
• For now, the expectation to have the option to speak to a human is still prevalent and highly important in maintaining quality customer service.

Managing the Systems of Support

Just as technologies like ticketing systems have done for support in the past, new AI technology will help us automate the work around the work and instead focus on connecting with customers, building relationships, and making a bigger impact on our businesses. But just as there’s overhead required to manage a piece of technology like a ticketing system, teams of the future will need to spend more time managing the AI. 

This requires a systems-approach, not just structural work like feeding sample data into the algorithm and continuing to train it, but continued evaluation of the overall results delivered by an AI solution to guard mistakes and bias that can happen when AI is given incorrect or biased data. For example, Amazon’s AI recruiting tool developed bias against women due to historical data that gave favor to male candidates. Humans need to be there to check in on the results generated by AI to ensure it’s working as intended.

Key Points:

• As AI systems grow, there will be a growing need for humans to manage these systems.
• Humans will always be needed to check-in and regulate AI to ensure there are no mistakes or biases in their operation. 

Don’t Fear the Technology, Embrace It

As technology keeps improving, customer expectations keep getting higher, especially among millennials. According to Microsoft’s 2016 State of Global Customer Service Report, millennials had 68% higher expectations for customer service in 2016 than they had just one year prior. 78% of young consumers expect service agents to already know their contact and product information when they contact them. To meet customer expectations, it helps to embrace the technology that is increasing the expectations.

When it comes to customer support, AI has a role to play, but people won’t be eliminated. Humans will still be needed to design and guide the human experience that powers customer support. So use the unique strengths of both AI and people to complement each other and unite as partners working toward the same goal: taking care of customers.

Human in the Loop AI for Your Customer Support Teams

AI has the power to elevate your customer support teams and grow your business. Idiomatic uses human in the loop AI to improve the customer experience and make the experience of your customer support team much easier.

Request a demo to learn about our AI solutions for customer support teams.