Do you ever feel like artificial intelligence (AI) is everywhere? It’s in our email inboxes, on our favorite websites, and even all over our social media pages. Yet, we’ve only begun to explore its potential to transform our daily lives and the world around us.
Here at ProTrainings, we’re tapping into AI’s potential to make health and safety training more accessible, effective, and user-friendly because, at the end of the day, life matters. Our AI customer support chatbot, powered by a large language model (LLM), is designed to solve users’ problems as accurately and efficiently as possible. Though the journey has its challenges, we’ve already seen remarkable outcomes—and we’re just getting started.
Here, we’ll share some of the challenges we’ve faced while developing our AI support chatbot, the real-world impact it’s already making, and how we believe it will shape the future of lifesaving training.
The Challenges of Building AI Chatbots
Creating a reliable AI customer support chatbot is no small feat. It requires constant learning, adaptation, and collaboration—both from the chatbot and from us. While we’ve made significant progress – particularly in using retrieval augmented generation (RAG) to guide the chatbot in answering customer inquiries accurately – we’ve also encountered several hurdles.
Referencing the Right Knowledge Base
One of our biggest challenges has been ensuring the chatbot references the correct, most up-to-date information from our knowledge base. When it does, it answers questions correctly. However, sometimes the chatbot relies on outdated training data instead, leading to inaccurate responses.
For instance, if someone asks about OSHA compliance requirements, the chatbot might give an outdated response unless it’s explicitly directed to check the latest guidelines. To address this:
- Preloading Answers for Common Questions: For questions with consistent answers—like “How do I reset my password?”—we’ve preloaded answers to ensure accuracy.
- Dynamic Lookups: We’ve instructed the chatbot to always look up answers for certain categories, such as certification requirements, to avoid relying on old data.
Still, the chatbot’s training data can be useful for timeless inquiries, such as principles of effective customer service. For example, it often recognizes when users are attempting to cheat on tests and tactfully refuses to provide answers, even without explicit programming for each question.
Understanding Inconsistent Responses
Not all LLMs are deterministic, so there’s no guarantee that the same prompt will produce the same response every time. This is because LLMs have many “weights,” or variables, that determine their behavior, many of which are created by reinforcement learning through thousands and thousands of prompts from developers and users.
Just like you can’t open up a person’s brain to see what they’re thinking, we simply have no way to fully understand what causes LLMs to make decisions about certain responses.
For example, sometimes minor language changes — such as asking “Can I do this?” instead of “How do I do this?” or saying “Please do this” instead of “Do this now” — make a huge difference to how the LLM responds.
Right now, we don’t understand exactly why this happens or how to avoid such inconsistent responses. However, observing these patterns does help us identify areas for improvement in our developmental processes.
We’ve worked to minimize inconsistencies by engineering prompts that balance context with simplicity. Moving forward, we’re exploring options like specialized chatbots for different types of inquiries, ensuring more precise answers for every user.
The ProTrainings Chatbot in Action
Despite these challenges, our chatbot has already delivered impressive results, improving both efficiency and accessibility for lifesaving training.
Real-World Examples of Success
- Understanding Intent: When a user enters a typo in their email address or submits a poorly worded question, the chatbot uses context clues to resolve the issue—often more effectively than a human agent could.
- Example: A customer typed, “Is this certified by OSHA,” and the chatbot clarified that while OSHA doesn’t certify courses, our training meets OSHA requirements.
- Providing Contextual Responses: The chatbot tailors its answers based on the user’s location on our website.
- Example: If a customer on the Pet First Aid Certification page asks about pricing, the chatbot provides specific fee details for that course.
- 24/7 Support: Since launching the chatbot, it has handled 2,460 chats outside of regular business hours—even on holidays and weekends—providing critical support when human agents aren’t available.
- Language Accessibility: The chatbot’s ability to communicate in multiple languages enables us to assist users worldwide, helping bridge language barriers in lifesaving training.
- Example: A user in Spain received detailed guidance on accessing their CPR certification entirely in Spanish.
- Human Collaboration: By reducing repetitive inquiries, the chatbot has allowed our human agents to focus on more complex projects, leading to an 84% drop in chats forwarded to humans. This shift has empowered our team to enhance their skills and work on strategic initiatives.
For us, AI chatbots aren’t just a cost-cutting measure. They’re about providing a better experience for our end users while freeing up our human agents to focus on higher-level tasks.
While the overall number of chats has remained roughly the same following the launch of our chatbot — 3,804 chats in August and September versus 3,674 chats in October and November — the number of chats that have had to be forwarded to human agents during that time has decreased from 1,094 to 181. That’s an 84% drop!
With all that extra time, our human agents have been able to work on complex projects instead of answering the same five questions over and over. Not only are they putting their time to better use, but they’re also developing new professional skills — all without sacrificing quality support for our customers.
The Future of Health & Safety Training
In the health and safety training industry, staying current with rapidly evolving standards is essential. At ProTrainings, we’re committed to leveraging AI to:
- Provide up-to-date training that reflects the latest medical and safety guidelines.
- Improve access to training resources for individuals and organizations globally.
- Ensure customer support is available anytime, anywhere, in any language.
By embracing AI, we’re not just improving our services—we’re redefining what it means to equip people with lifesaving skills. Whether it’s delivering CPR instructions in multiple languages or helping users navigate certification requirements seamlessly, our chatbot is a testament to the power of technology in saving lives.
The Tip of the Iceberg
AI has come a long way in revolutionizing customer support and training, but this is just the beginning. At ProTrainings, we’re committed to pushing the boundaries of what’s possible—because when it comes to saving lives, there’s no room for complacency.
Follow us on LinkedIn to stay updated on our journey and see how we’re making health and safety training more accessible and impactful for everyone.