Complete Guide to Chatbot Containment Rates

1. Track total and escalated interactions

Count the total number of user interactions with the chatbot during a given period. This includes all conversations, queries, or tasks initiated by users.

Then track the number of interactions that are escalated to human agents or flagged as unresolved by the chatbot. 

‘Escalations’ could include direct handoffs to human agents or cases where users explicitly request help.

2. Calculate the containment rate

Use the formula:

Containment Rate = (1− [Escalated Interactions / Total Interactions]​) × 100

For example, if your chatbot handled 900 out of 1,000 interactions without escalation, you would calculate: 

Containment Rate = (1−1000/100​) × 100 = 90%

3. Or use analytics tools

Leverage chatbot analytics or customer service platforms that automatically track and report metrics like containment rates. These tools often provide additional insights into reasons for escalations and user satisfaction.

By continuously monitoring and optimizing based on this metric, you can enhance your chatbot’s efficiency and user experience.

Use LLMs instead of intent classifiers

High-containment chatbots are typically powered by LLMs rather than intent classifiers. 

LLMs excel at understanding nuanced, free-form language and adapting to varied phrasing, making them ideal for handling complex or unexpected queries. 

They eliminate the need for rigid intent mappings, allowing the chatbot (sometimes called an LLM agent) to flexibly respond to a wider range of user inputs with precision.

As our COO explains in his article, Botpress uses LLMs over intent classifiers, unlike our competitors.

Design for adaptive conversations

Rather than relying on static workflows, chatbots should be built to dynamically adapt to user behavior in real time. 

By leveraging contextual awareness and decision-making algorithms, the bot can adjust its approach mid-conversation. This ensures it remains relevant, even when users shift topics, provide incomplete information, or ask follow-up questions, improving overall resolution rates.

The easiest way to design a dynamic chatbot is to use agentic AI; that way, your bot can autonomously decide how to best approach an issue based on the tools at its disposal.

Leverage retrieval-augmented systems

Combining retrieval-augmented generation (RAG) with integrations like CRMs and knowledge bases unlocks the full potential of a chatbot. 

By pulling real-time, domain-specific information from reliable sources, the bot can answer complex questions and complete multi-step tasks without escalating to human agents.

1. Improve intent recognition with LLMs

Switching from traditional intent classifiers to LLMs will likely improve your containment rate.

Unlike rigid intent mapping, LLMs can understand nuanced, free-form language and adapt to diverse phrasing. This advanced understanding reduces misunderstandings and equips chatbots to handle complex, unpredictable conversations with accuracy and fluidity.

2. Expand knowledge bases

Keeping your chatbot’s knowledge base updated is key to satisfying interactions. 

Regularly add new FAQs, product information, and solutions to common issues. This equips the chatbot to handle a wider range of queries, ensuring users receive accurate and timely answers.

3. Increase your suggest pathways

Before escalating to a human, your chatbot tries to resolve queries independently. If it’s not succeeding, try adding more pathways for users to opt for.

Adding fallback options, like clarifying questions or alternative solutions, gives it more opportunities to address user needs. 

4. Build in support throughout the process

If your flow includes complex workflows – like booking appointments or troubleshooting – ensure that these can be smoothly accomplished by your user without a human. Chatbots that can guide users through these processes prevent the need for human intervention.

5. Clearly explain what your chatbot is for

Set clear expectations from the start. Guide users on what the chatbot can handle and where it excels. This avoids frustration by steering conversations toward areas where the bot is most capable, setting it up for success.

6. Monitor performance

Track key metrics like unresolved queries, escalation rates, and satisfaction scores to measure how well your chatbot is performing. Use these insights to pinpoint areas for improvement and evaluate the success of updates.