A smooth deployment with a generative AI agent depends on choosing the best use cases to target first. Choose the right ones, and you’ll prove the agent’s value to the business quickly. But choose badly, and you won’t know for sure how big the benefits could be. It isn’t just the question of temporary underperformance. It’s about what comes next. If you’ve chosen the wrong use cases to target, you won’t have much luck fine-tuning, and that could leave you on shaky ground with internal stakeholders.
How do you choose which use cases to target?
Identifying the right cases starts with a thorough examination of your contact center’s interactions to understand why customers get in touch and the steps required to resolve their issues.
Based on our experience deploying the CXP (Customer Experience Platform) with GenerativeAgent® in enterprise contact centers, ASAPP has developed a reliable, data-driven process to analyze your interactions and identify good candidates for automation. For each intent, we work with your team to determine the containment and resolution rate, assess the potential for labor hour savings, and establish what steps are required to resolve the customer’s issue.
With the results of this analysis, you’re well equipped to make informed decisions about which use cases you’d like to target first. This is a collaborative process. We provide analysis and expert guidance. You make the decisions at every step.
Regardless of which AI agent you’re deploying, it’s helpful to remember that the overarching goal is to identify the use cases that
- Your current automation cannot contain
- Your AI agent will either be able to resolve on its own or with the assistance of a human in the loop
- Will lower costs by reducing required labor hours
Identifying customer interactions with high automation potential
We start with a collection of your actual customer interactions. Using natural language techniques, we analyze the conversations to assess the potential for automation with GenerativeAgent. To do that, we work with you to answer these questions.
Which intents have low containment with your current automation?
There’s not much point in starting with use cases that your existing automation already handles well. The goal is to automate interactions that currently require human labor to resolve. Simply shifting interactions to GenerativeAgent from your existing bots, virtual agents, or IVR won’t deliver any savings.
With that in mind, it’s a good idea to target intents with low containment. Those could be intents that your existing automation attempts to handle, but too often fails to contain and resolve. Or it could be intents that are currently routed to human agents from the beginning.
Why are these interactions not contained?
Understanding why each type of interaction is not contained can help you zero in on the best candidates for automation with an AI agent. For some intents, the reasons for low containment are related to the limitations of your existing automation. Those are likely good candidates for a generative AI agent.
Other intents with low containment might require specialized skillsets, human judgment, or relationship-building. You’ll want to keep assigning those intents to human agents, at least for now. You’ll have plenty of opportunity to reconsider them as you expand your list of use cases in the future.
When low containment = high automation potential
These are the conditions in which low containment translates into high potential for automation with a generative AI agent:
- Customers need more specific and detailed answers than your existing automation can provide.
- The resolution requires applying complex policies that are difficult to code into a bot flow.
- An API returns too much data for a deterministic bot or virtual agent to handle.
- The conversation can take many different paths, making it impossible to map into a deterministic flow.
- Your IVR or virtual agent can answer the main question, but struggles with the follow-up questions many customers ask.
Rethinking your routing
Your routing method won’t determine which use cases you should target. But it’s a good idea to consider how incoming interactions will be routed to your generative AI agent before you decide on your first use case. If customers will first interact with an IVR, virtual agent, or other routing solution, you’ll need to plan for that technology to identify the intents that you expect the AI agent to handle. If the right interactions are not routed to the AI agent, you won't be able to evaluate its performance or its impact on your contact center. So, it’s important to plan ahead for accurate routing.
What’s the potential savings in labor hours for each intent?
Interaction volume is a key factor in labor hour savings. Customer intents that drive high volume are often good candidates for your first use cases. But don’t automatically disregard lower-volume intents, especially those with large Average Handle Times (AHT). If an intent requires your agents to spend a lot of time combing through knowledge articles and complex policies to come up with a clear answer, the labor hour savings could be substantial even with lower volume.
Which intents were resolved by human agents without escalation?
For each intent, we determine what percentage of interactions successfully resolved the customer’s issue without a transfer or having the customer abandon the interaction. Interactions that your agents resolved in the first contact without escalation to a specialist or supervisor are likely good candidates for your AI agent to resolve, too.
Considering use cases that require a human in the loop
Most generative AI agents can transfer a customer to a human agent when necessary, but there’s usually no carryover of context. Combine this with waiting in a queue for an available human agent, and the result is often repetitive and time-consuming. It’s a disruptive and disconnected experience.
Some AI agents handle the transfer more seamlessly, passing conversation and history to the human agent. GenerativeAgent does even more. It can consult a human agent for information or guidance—without transferring the customer.
It’s important to keep in mind that reducing required labor hours without sacrificing customer satisfaction is a primary goal for any generative AI agent solution. So, use cases that require escalation or consultation with a human in the loop could still be good choices, even for your initial launch. If the AI agent can reduce the burden on your team by handling the bulk of each interaction for a particular intent, the time savings could be substantial. If it leads to faster resolutions with less effort, it’s a win for your customers, too.
Extending GenerativeAgent capabilities with HILATM
The ASAPP CXP (Customer Experience Platform) expands automation potential with GenerativeAgent with an innovative Human-in-the-Loop Agent workflow. When GenerativeAgent hits a roadblock, it can ask a human advisor for information, guidance, approval for a particular action, or even to perform a task. When it gets what it needs, it continues resolving the customer’s issue on its own.
What tools did the agents use to resolve the issue?
Once we’ve identified which intents are good candidates for automation based on potential labor hour savings, containment, and resolution, we examine the tools the agents used to resolve the customer’s issues. Specifically, we detect which systems and knowledge articles the agents accessed during the interactions.
To make this determination, we rely on the responses the agents provided to the customer at each step. If a response contains general information about products, services, or policies, we conclude that the agent relied on the knowledge base (or other designated source of truth) for that information. If a response includes customer-specific information, we conclude that the response required the agent to access another system, such as your CRM. Some responses require information from both the knowledge base and another system.
Understanding the steps the agents took to resolve the customers’ issues is important because GenerativeAgent will need to follow the same steps.
Which tools are already available for the AI agent to use?
For each task an AI agent must perform, it needs the right tools. For some tasks, it will need accurate knowledge articles on relevant topics. For others, it will need APIs to access other systems and retrieve the necessary information.
Our team generates a list of APIs that will be needed based on each task the agent performs. For example, if a customer asks to cancel their account, the agent might need to perform multiple tasks, such as account cancellation, service cancellation, and arranging for equipment return. An API would be required for each of these tasks.
The power of API transformation
Most AI agent providers require APIs specifically built to work with their solution. But chances are, your APIs weren’t designed for an AI agent. That can create a major deployment hurdle and a lot of work for your team, not just for your initial launch but every time you add a task that requires an API.
With GenerativeAgent, there’s no need to rewrite your APIs. Because the CXP comes with an API transformation layer, GenerativeAgent can use your existing APIs just as they are.
Learn more about the ASAPP CXP API transformation layer.
At this stage, the ASAPP team will work with you to determine which knowledge articles and APIs you already have and which ones you will need to create. Naturally, interactions that rely on tools that already exist can be deployed more quickly and with less effort on your part. These are likely your best bet for first use cases.
If you need to create tools, which ones should you prioritize?
If you lack necessary APIs, we can help you decide which ones to build first based on the impact they’ll have on GenerativeAgent’s ability to resolve the customer’s issue successfully. For each intent, some APIs will be required for many of the interactions, while others will be needed only occasionally. With that information, you can invest your time and resources in building the APIs that will yield the biggest returns.
No API? No problem with GenerativeAgent
You might have a use case for your AI agent that lacks a necessary API. With GenerativeAgent, that’s no problem. The ASAPP CXP comes with an innovative Human-in-the-Loop (HILATM) workflow. In place of the missing API, GenerativeAgent can use the HILA workflow to ask a human advisor in your contact center for the information or action it needs – and then continue with the rest of the use case on its own. In this case, it’s helpful to think of the human in the loop as a biological API that allows you to launch the use case and realize value.
We’ll follow the same collaborative process to determine which knowledge articles already exist and which ones you’ll need to create. And of course, we’ll provide the usage data to help you prioritize them. GenerativeAgent can ingest content from other sources, which could reduce the burden of creating new knowledge articles. For example, if some of the necessary content exists on your website, we can scrape that information. We’ll work with you to determine the most efficient and effective option.
You choose which use cases to target first
With the results of this detailed analysis, you’ll have all the information you need to choose which use cases to automate first with your AI agent. And you’ll be equipped to plan how to expand to additional use cases over time.
When you’re ready to expand, you’ll follow the same process to choose which use cases to add.
How will you know you’ve chosen the right use cases?
Prior to launch, we work with your team to thoroughly test the use cases you’ve chosen. That testing will include simulations of realistic scenarios, including API calls, knowledge retrievals, and multi-turn conversations with different customer personas. By launch date, you’ll be confident that GenerativeAgent will perform well for your chosen use cases.
After go-live, we continue working with you to measure, monitor, and fine-tune GenerativeAgent to optimize performance. The ASAPP CXP comes with a robust suite of tools that will equip your team to continue monitoring, testing, and optimizing GenerativeAgent on your own.
A CX leader's guide to getting value from your AI agents
Whether you’re just getting started with an AI agent or seeking better results with one you’ve already deployed, this guide offers clear direction for realizing greater value. Gain clear, actionable strategies for identifying obstacles that degrade your returns, proven tactics for tuning your AI agent to improve performance, and how to know when to scale up for bigger returns.
Learn more about Getting value from your AI agent
