How to Accelerate Customer Service Automation with Generative AI

What are the potential applications for Generative AI in customer service?

There are a number of potential use cases for generative AI in your contact center, some of which are internally facing, and some of which are externally facing.

Internal Generative AI Use Cases:

• Internal Knowledge Base Search: Contact center agents spend significant amounts of time searching internal knowledge bases for the right answer to customer questions. Generative AI can dramatically accelerate the time it takes to find and compose the right answer.
• Training: Use generative AI to train and onboard contact center agents. Some generative AIs can be configured to do a role play and (1) pretend to be a customer and (2) provide feedback to the contact center agent after the role play is complete.
• Internal Agent Assist: Generative AI can listen to or read customer questions, provide sentiment analysis, and summarize customer questions. Over text channels, generative AI can draft email or live chat responses, and agents can review, edit, and send final responses. In voice channels, generative AI can provide key talking points for contact center agents.
• Customer Intelligence: Generative AI can automatically analyze all customer communications and help you understand why customers are contacting you and where your knowledge base needs to be enhanced.
• Quality Assurance/Compliance: Use Generative AI to automatically analyze all agent communications to ensure response accuracy and compliance, and provide coaching and feedback to agents.

Externally Facing Generative AI Use Cases:

• Audience Engagement: some contact centers may interact with audience members on social media, either on Facebook or Instagram ads, or via social messaging apps. Generative AI can provide first line support for those channels and drive more visitors to your site.
• Sales: Many contact centers power “Talk to an Expert” sales functionality. Generative AI can be that expert and answer questions 24/7/365 in 100 different languages, answering in-depth product questions and providing recommendations.
• External Knowledge Base Search: Rather than powering an externally facing KB with a traditional search engine – where the customer enters a search query, and all KB documents relevant to the query are displayed, leaving the customer to find the right answer among the search results – a generative AI could respond with one answer, and also provide links to supporting documentation.
• Authentication & Triage: Some brands may choose to use generative AI to automatically authenticate customers, categorize and prioritize customer issues, and route customers to most relevant agents.
• Omnichannel automated customer service AI: Generative AI can be deployed across all your customer service channels – email, live chat, voice, and others – and provide Level 1 customer service. The AI can transfer to human agents in your contact center for Level 2 service (or for anyone who wants to interact with a live agent and potentially wait).

Which support channels can I use Generative AI in?

You can use Generative AI in all of your customer service channels, including email, live chat, social messaging apps, Slack, Discord, SMS, and voice.

What’s the best way to deploy Generative AI?

One word: incrementally.

Using generative AI to automate customer service doesn’t have to be an all-or-nothing proposition.

Like all big initiatives, it’s best to get started incrementally. Waiting only delays your learning generative AI learning curve, and puts your organization at a perpetual disadvantage to competitors that are willing and able to deploy GenAI faster.

Start with internally-facing applications first (as listed above).

As confidence in generative AI increases in your organization, start including asynchronous externally-facing applications (like email and SMS), and then move to real-time externally-facing applications (like live chat and voice).

How should I respond to “AI will never replace our human agents” naysayers?

Two words: “I agree.”

Generative AI can help you automate much of your Level 1 customer service/support.

But will Generative AI replace your customer service team in full? It’s highly unlikely.

There will always be customer service issues that require a human in the loop: new products, new customer use cases for existing products, issues that require human judgment (e.g., did the product fail because of misuse/abuse, or was there a manufacturing defect?), etc.

Generative AI won’t fully replace human agents. Instead, it will be a force multiplier and help your agents be superagents.

What are the key decisions I need to make in a Generative AI strategy?

Every generative AI strategy has a few key decisions that you need to make:

• Define Objectives: What are you looking to achieve with Generative AI? What problems are you looking to solve? You shouldn’t be implementing generative AI because it’s new, interesting, and exciting – you should have a clear objective or problem to solve.
• Budget: What is the value of the objectives you’re looking to achieve/problems you’re looking to solve? Based on that value, how much are you willing to invest in a generative AI solution (either building one internally or licensing one)?
• Build versus Buy: Should your company build its own generative AI solution, or should you buy (or license) a commercial generative AI solution? Read more about considerations around building your own generative AI solution or evaluating commercial generative AI solutions.
• Use Cases: Which use cases do you want generative AI to automate? What’s the relative priority (i.e., order) in which those use cases should be tackled? This will be your Generative AI roadmap.
• Knowledge: What knowledge / knowledge bases will the generative AI need to train on, in order to solve the above use cases?
• Integrations/APIs: If some of the use cases include self-serve capabilities (e.g., looking up an order status, initiating a return) which APIs will the generative AI solution need to integrate with? What other solutions in your contact center stack will your generative AI solution need to integrate with? (e.g., help desk, CCaaS, etc.)
• Team: Who will train and manage the AI on a day-to-day basis? A generative AI will need to be managed just like a human contact center agent. Someone (or a small team of people) will need to:
  • Train the GenAI agent;
  • Decide whether the GenAI agent is ready for internal and then external use; and
  • Monitor the performance of the GenAI agent and give it ongoing feedback

How long should it take to train a Generative AI solution?

If you’re building a generative AI solution from scratch, it could take months to build a reliable, non-hallucinating solution.

A good commercial generative AI solution should be able to ingest all of your knowledge in a matter of a few hours.

After the generative AI’s initial ingestion of knowledge, you should test the AI and give it feedback. As you give it feedback, the quality of the AI’s responses should quickly and steadily improve.

Response quality of best-in-class generative AI solutions can improve by about 50% within 1-2 weeks of feedback.

The number of people you need to manage a generative AI solution depends on a few things:

• The total volume of customer service inquiries
• The % of use cases the generative AI can resolve
• The state of your knowledge base before implementing generative AI – e.g., if you already have a high organized support KB and FAQ, there may be less day-to-day supervision involved

Rather than having a dedicated team manage a generative AI solution, it may be useful to have the management of generative AI be distributed across your contact center team.

How many people will I need to dedicate to managing a generative AI solution?

The number of people you need to manage a generative AI solution depends on a few things:

• The total volume of customer service inquiries
• The % of use cases the generative AI can resolve
• The state of your knowledge base before implementing generative AI – e.g., if you already have a high organized support KB and FAQ, there may be less day-to-day supervision involved

Rather than having a dedicated team manage a generative AI solution, it may be useful to have the management of generative AI be distributed across your contact center team.

What skills are needed to manage a generative AI solution?

Managing a generative AI is just like managing a customer service agent. You need to:

• Carve out some time to supervise the AI
• Be able to recognize good answers from bad/incorrect answers
• Coach the generative AI – i.e., provide an example of a good/better answer
• Monitor performance metrics

How should I measure the success of generative AI in my contact center?

To measure the success of your generative AI, you should consider monitoring:

• Average response time
• Resolution rate: the number of customer conversations with the generative AI that did not result in a transfer to a human agent, divided by the total number of generative AI conversations. Note: both the numerator and denominator of this metric should not include customers who immediately asked to be transferred to a human agent
• CSAT of generative AI-resolved conversations
• CSAT of human-resolved conversations
• Employee satisfaction of human agents
• Agent turnover rate

What kind of GenAI productivity gains should I expect in my contact center agents?

Gains may vary by industry and by the number of use cases deployed.

In general, however, you can expect your contact center agents to be 20%-30% more efficient with human-handled customer service issues. (Note: this metric doesn’t include customer service issues that are fully resolved by generative AI.)

What should I do with those GenAI efficiency gains?

You can do anything you want! Delay your next contact center hire. Start work on other long-deferred customer service projects.

Good, fully trained employees are hard to come by, and there’s often too much work for too few people. As such, many contact center managers who deploy generative AI often look to reinvest efficiency gains back into the team/business.

Leverage your AI-driven efficiency gains however you see fit! You can defer your next contact center hire or allocate resources to long-overdue customer service projects.

With skilled employees in high demand and teams often stretched thin, many managers reinvest these newfound efficiencies back into the business, improving overall performance and team morale.

How do I ensure a successful deployment of generative AI in my contact center?

A few key pieces of advice:

• Executive sponsorship: Make sure you have executive sponsorship/buy in. Define your objectives and paint a future vision of what a generative AI-enabled contact center looks like in your company.
• Manage expectations: Rome wasn’t built in a day. No generative AI will be perfect on Day 1. Remember – a generative AI is just like a new employee. It needs to be trained; it needs coaching and feedback. It needs to be supervised on an ongoing basis.
• Socialize the solution: Allow employees to play with the generative AI in a low-stakes environment. For example, some companies have deployed a generative AI on a dedicated Slack channel, allowing anyone in the company to ask it questions.
• Deploy gradually: You don’t need to deploy all use cases immediately. Go for the highest volume/highest impact use cases first. Like any major organizational change, you want to make sure you have some easy wins under your belt, early on.

How do I keep my GenAI solution up-to-date with industry trends?

If you build your own generative AI solution, your engineering team will need to stay up-to-date with the rapidly changing advances in generative AI, e.g., new LLMs, longer context windows, different methodologies for reducing hallucination, etc.

If you license a generative AI solution, your vendor will need to stay current on all those changes.

As a result, you should make sure your vendor is nimble, and their solution is agnostic of foundational generative AI models.

Continue reading Chapter 4 - Generative AI Applications in Customer Service Automation