Five Agent Assist applications for contact centres

In this article, we’re looking at agent assist approaches for customer care centres. Agent assist has become increasingly popular as a method to improve agent performance while maintaining or even improving customer service. The popularity has increased as systems have become more flexible and open with data APIs, and the accuracy of AI systems has reached a point where suggested answers and actions are reliable and trustworthy.

Humans are the best agents

The most flexible method (but also the most expensive method) to answer a customer query at a contact centre is to use a skilled human operator on the other end of a chat, email, or phone call. Customer queries that are not covered by, or are too complex for, automated deflection are ultimately presented to human advisors. 

Human operators are powerful. For the highest customer satisfaction, every interaction would be answered by highly trained customer service agents. They:

• can understand the intricacies of customer situations,
• flexibly adjust responses to suit nuanced situations,
• use humour to diffuse tension, and negotiate effectively in difficult interactions.

A fully manual approach is untenable for most high-volume contact centres with limited budgets. Bundled with the hourly labour cost for customer service agents are the expenses around hiring, offices, equipment, training, management, and quality monitoring. A completely human-powered contact centre requires investments that most companies would baulk at.

AI Agent Assistance

Agent Assistance Technology is AI-powered technology that augments human agent behaviour in the contact centre, using automation, machine learning, and software systems to improve the performance of agents while they answer customer queries. Agent assistance approaches can target the improvement of customer satisfaction scores, increases in agent efficiency, or both at once.

Where agent assistance approaches work best is in the automation of tasks that agents perform that are suited to computer automation. Reduction of the work that agents are performing allows them to process more queries per day (adding efficiency), and typically, automated systems can work quite accurately for many tasks, improving customer satisfaction. 

We’re examining four different methods where agent assistance can be implemented.

• Automated customer information retrieval
• Pre-Preparation of suggested responses for agents
• Automation of the contact wrap up process
• Real-Time Translation
• Sentiment Analysis (to be discussed in a separate and dedicated post)

AGENT ASSIST: Information Retrieval

When an agent starts to answer a query on any channel, their first task is to gather and understand the context for the query and the current customer situation.

• Who is the customer?
• What are they trying to achieve?
• What has happened to them thus far?
• What is their emotional state?

The highest performing agents will ensure they fully grasp the customer requirements before providing their answer.

Unfortunately, gathering the relevant customer information, contact history, and helpdesk articles can be time consuming, often involving interactions across multiple applications. Agents can have several windows open per customer. When speaking to multiple customers on messaging channels, manual management of this data becomes time consuming and the chances of errors increases.

Automation and Agent assistant technology can reduce or completely eliminate the time spent by agents searching for customer information. Once the customer has been identified through a booking ID, account ID, or email address, all of the most relevant information can be automatically extracted and presented to the customer service agent. Additionally, if any text classification or natural language processing is applied to determine the intent of the customer’s question itself, the information presented to the agent can be customised to the question asked.

For example, let’s examine a situation where a customer starts a chat session with their email address (john@gmail.com) and a question: “I need to cancel my booking for March, reference 22334”. In this situation, with a well integrated customer service automation system, when the query is presented to an agent, the system can already have fetched all historical contact information for john@gmail.com, all of the booking information for booking reference 22334, and, with an understanding of the query, also specifically present the cancellation terms for booking 22334, along with any standard terms and conditions in template form for the agent to use.

Using AI for information retrieval reduces the need for agents to look in multiple systems and improves the accuracy of responses when speaking to multiple customers, resulting in lower handling times and better customer satisfaction.

Agent Assist: Real-Time suggested Responses

Once the system has actually presented all of the relevant information and the query to an agent, it can assist in the composition of the appropriate responses during the interaction.

For a customer service agent, answer composition is typically process driven for known issues. For our example with john@gmail.com, in the case of cancellations, the response to the customer may depend on the time left until the start of their booking, the particular booking type, the value of the booking, or the customer type. 

An agent assistant system can codify the agent knowledge into a flowchart of decisions to generate an appropriate and customised response to each customer situation. These pre-generated answers will not be perfect, but will perform a significant portion of the agent effort while answering customer queries. Agents can then review and edit the response before sending, ensuring the highest level of service and ensures that any additional nuances in the customer query are addressed.

With well-designed agent assistance, the customer receives the speed benefit of AI systems, but the nuanced understanding of the human operator.

Agent assistant systems that generate suggested customer responses can be used on email, chat, call, and messaging channels. 

Agent Assist: Contact Wrap up Automation

The steps that an agent takes when a customer query has been completed vary widely between contact centres. At EdgeTier, when helping contact centres, we’ve seen everything from a single drop down form right through to a manual copy and paste of chat transcripts into four disparate systems (albeit, that was the worst we’ve seen!). 

It’s incredibly tempting to contact centre administrators to “add another field” or “expand the number of contact reasons” for agents, but such actions can build over time to reduce operational efficiency.

Collecting information from agents at the end of an interaction is important for reporting, quality, and management, but quickly becomes an issue if the time taken by agents to complete the process starts to expand, or the quality of the data starts to drop. It’s incredibly tempting to contact centre administrators to “add another field” or “expand the number of contact reasons” for agents, but such actions can build over time to reduce operational efficiency. Handling time (AHT) is always measured, and there’s a natural tension between collecting the best possible data and moving to the next customer query.

Automation can help. Prompting agents with a reduced set of contact reason labels can be achieved with natural language processing (NLP) and text classification of the contents of the interaction. Contact log summaries can be automatically generated with language AI systems. API integrations can help to ensure that agents spend no time copying/pasting data between systems and interactions are automatically inserted into the right systems immediately after completion.

Initiatives for wrap up automation primarily target time savings for agents, but have a secondary impact of improving data quality and agent experience.

Agent Assist: Real-time Translation

Often underestimated as a hidden advance in AI technology, the accuracy of machine translation systems has leapt forward in recent years, particularly amongst European languages. With translation costs as low as $20 per million characters from providers like Google, the technology is now within reach for those with ambition to try it. 

In our experience for many applications and across multiple industries, machine translation technology is now suitable for seamless text-based communication between customer service agents and customers.  Translations are not completely perfect, but the imperfections will not inhibit understanding. Customer satisfaction, where EdgeTier has used real-time translation, has also not been adversely affected.

In contact centres that embrace translation technology, there are widespread changes to hiring practices and team scheduling. Corner languages can now be provided with the same 24/7 care as core languages without additional team members fluent in every language, and specialized hiring teams are no longer required to hunt down native speakers.

For example providers, Google Translate provide general purpose translation APIs, Unbabel are a good example of contact centre specific providers. Open source language models have started to be released (here is a completely open source model with 200 languages from Meta), and we think that it’s likely that translation technology will become more commoditised as the technology continues to improve.

Outcomes and Pitfalls

As with all automation initiatives, the implementation of an agent assistance program in a contact centre requires planning and skilled execution. A host of factors will influence the chances of success, including but not limited to:

• The information infrastructure – how easy is it to access systems with customer information, are there APIs, how many systems are there.
• The agent desktop solution – is the agent desktop in place suitable or flexible enough to present information, suggestions, or manage automation?
• Agent buy-in – Agent assistance can be seen as a threat, whereas, in reality, a properly implemented solution can dramatically improve the agent experience.
• Management buy in – It’s likely that not every part of an agent assist project can be implemented without sponsorship from senior leaders. There will be implementation required from the contact centre, software vendors, and the IT and security departments.

Ideal outcomes

• A lower cost base, resulting from reductions in agent handling times (reduced lookups, wrap times, and reply times) across the contact centre.
• Increased throughput at the contact centre from efficiency gains for each agent.
• Improved customer satisfaction driven by higher quality responses with less errors by customer service agents.
• Lower wait times for customers due to the improved throughput across the centre.
• Improved reporting accuracy and data visibility on contact reasons.

Potential Pitfalls

• A lack of investment to keep automations up to date with changing systems or policies, such that the suggested assisted responses are out of date or not useful.
• Not taking agent opinions and experience into account to ensure buy-in at the contact centre, and to continually improve the assistant system.
• Implementing very limited assistant systems because they are not integrated into the internal APIs and CRM.
• Implementing machine translation systems without ensuring compatibility with the specific vocabulary and language mix of the contact centre can lead to inaccurate translations and unexpected situations!

Where do you start with agent assistance?

If you’re interested in implementing any of the agent assistance initiatives in this post, we’d love to speak. At EdgeTier, we’ve used agent assist to reduce handling times by up to 80% in some cases, and we’d love to tell you about it.