Email bots leverage powerful artificial intelligence (AI) and workflow capabilities to vastly reduce email processing times and accelerate customer service response times.
However, like any technology, they won’t be able to solve 100% of a company’s issues overnight. That’s because the types of emails organizations need to process are heterogeneous – spanning a wide-array of topics, languages, business units, and complexities.
How can you optimize your email bot deployment to achieve maximum value as fast as possible? By following the three key phases to a successful email bot implementation:
- Strategizing
- Deployment and Monitoring
- Re-strategizing and growth
Let’s dive deeper into these three key phases. You can use this guide as your playbook when putting together your own organization’s plan for an email bot implementation.
Step 1: Strategize
The biggest factor to ensuring success in any software project is a sound strategy. To create an effective email bot strategy, you must first establish your baseline. This involves developing an understanding of your current email processing landscape.
Use estimations to provide a high-level view of things like:
- Volume of incoming emails by topic
- Average email processing time by topic
- Back-end systems involved and complexity assessment per topic
Once you’ve established your baseline for current email processing, you’ll need to prioritize work for your email bot.
Categories for prioritization include:
- Low-hanging fruit: There may be a high volume of low-effort questions coming into your agents. Often, these are simple questions that don’t involve personal information or back-end systems (e.g., “How do I reset my password?”). Although these types of emails may not take up the bulk of your agents’ attention, many organizations automate self-service responses to these questions so they can free up their agents to process higher value emails.
- Low complexity, high volume: The next grouping emails often fall into is common questions that require a low/moderate number of back-end system interactions in order to resolve them. These are often simple questions about personal information (e.g., “When is my payment due?”). Topics like this can be automated end-to-end with workflow automation.
- High complexity, high volume: This grouping is also for common inquiries; however, these emails require a high level of automation for work to be done autonomously (e.g., “dispute a transaction”). In scenarios like this, an email bot should be used to incrementally add more value to agents over time. Start by using the email bot’s AI to guide agents to faster resolution. Then, as the AI learns, use the bot to automatically identify the topic and kick-off work for related business units via back-end workflow systems. Over time, you can further automate processes through intelligent automation capabilities like robotic process automation (RPA) and integrations.
- Everything else: These are your low volume emails that you’ll prioritize last for automation.
Step 2: Deploy and monitor
Once you’ve prioritized your email bot strategy, you’re ready to begin iteratively developing and rolling out the necessary AI and workflow components within your email bot.
As you begin to use the bot, a few areas to monitor include:
- AI confidence scores: AI models typically publish a confidence score along with their analysis as an indication of model uncertainty. This uncertainty in natural language processing (NLP) is an inevitable result due to limited training data and the many intricacies of human conversation. As training data increases, confidence scores will increase as well.
It’s important to implement your email bot to use confidence scores as thresholds for kicking off automation based on the use case. For instance, you may want a lower threshold of confidence scores to kick-off automation for internal, team-to-team use cases (e.g., 75%+). However, for customer-facing use cases where accuracy is paramount, you may want to ensure the AI is 90%+ confident before completing the work automatically.
- AI model training: Email bot implementations will be most successful when deployed using adaptive AI. This refers to machine learning models that can learn over time by retraining models with new or updated datasets. Using adaptive AI ensures that confidence increases over time, models adapt as customer behavior changes, and AI becomes tailored to the business.
When planning AI model training, it’s important to also:
- Empower the business to train: Nobody understands your customers and processes as well as your agents and employees. Ensure your email bot takes advantage of their knowledge by providing them with a visual UI in which they can highlight text from emails to train new models and use an AI toolbox that can learn their actions in the applications they use.
- Set up a pipeline for retraining: Data scientists can become single points of failure in an email bot implementation if your model retraining process requires the export of large datasets and the updating/execution of Python scripts. Set up an automated, autonomous process to pump new data into the AI model, rebuild it, and promote it so anyone can instantiate the process with the click of a button.
- Overall activity: Visibility into the activity of your email bot is key to validating your ROI expectations, planning future iterations, and identifying bottlenecks. Find an email bot solution that provides out-of-the-box reports and dashboards that aggregate results into:
- The number of emails processed by topic
- The number and types of cases created
- The automatic and manual triage rate of cases by type
- The automatic and manual resolution rate of cases by type
- The number of emails processed by topic
Step 3: Re-strategize and grow
Once you’ve implemented and deployed your email bot, use its insights to update your strategy and plan your next move.
Your growth may lead you in a few directions. Things to consider adding to your solution include:
- Topics: Businesses receive emails on hundreds of topics. Your initial email bot strategy and implementation likely addressed the top five to 10 use cases. If there are more high-volume topics coming in, use the reporting capabilities in your existing email bot to identify, prioritize, and act on them.
- Automations: There may be topics in which your email bot identifies and triages work, but still requires employee intervention to complete the task. Find out which use cases are most pervasive and prioritize the development of automations and integrations that connect back-end systems to address these tasks.
- Channels: While email is an important channel for customer service and work intake, it’s not enough for many use cases. Once you have your email bot up and running, you may want to consider adding a chat channel, web app, or mobile app to the same workflow process. Leveraging your email bot platform enables your business to use the same process across channels and limits duplication of work/effort.
Start automating work intelligently
To deliver on customer and operational expectations, organizations need a sustainable strategy for continuously reading, interpreting, routing, acting on, and responding to large influxes of emails – one that frees up their employees to focus on more urgent tasks.
Email bots provide a complete email automation strategy that accelerates response times, gets real work done, and frees employees to focus on what matters most – the customer. This playbook will help make sure your implementation is a success.
Recommended resources:
- Read the Pega Community's Pega Email Bot overview.
- Visit Pega.com to learn more about email bots.
Don’t forget
JOIN THE CONVERSATION on Collaboration Center
FOLLOW @PegaDeveloper on Twitter
SUBSCRIBE to the Pega Developer Podcast on Spotify or via RSS