The Authenticx Approach
Our platform is purpose-built to help healthcare teams leverage their existing data to drive optimal outcomes for their clients and organizations. Authenticx AI is trained in-house on 100% healthcare-specific data to ensure the highest level of accuracy and trustworthiness.
Our AI models are designed with healthcare-driven deep learning techniques to deliver insights, ensure compliance, and enhance quality outcomes. Additionally, our generative AI models are tailored to address specific healthcare challenges, providing solutions that are both reliable and precise. In this article, we’ll explore how these specialized AI models are utilized within the Authenticx platform to meet the unique demands of healthcare organizations.
Human in the Loop Philosophy
We believe in the synergy between human intelligence and artificial intelligence. Our human-in-the-loop approach ensures that while AI models analyze data and generate insights, human expertise is integral to validating and refining these outputs. This method combines the precision of AI with the nuanced understanding of human judgment, enhancing the accuracy and relevance of our solutions for healthcare organizations.
Authenticx AI Models
Application | Model Type | Description |
Redaction | Deep learning | Automatically remove select personally identifiable information from conversations. |
Eddy Effect™ Signals | Deep learning | Surfaces when a desired or expected customer experience is disrupted by an obstacle, along with identifying where in the interaction the friction point was experienced. |
Safety Event Identification | Deep learning | A family of models that flags Safety Events (product quality complaints, adverse events, and special situations), along with identifying where in the interaction the Safety Event was flagged. |
Safety Event Acknowledgement | Deep learning | Determines if the agent talking with the patient appropriately addressed the Safety Event, indicating they will report it properly. |
HIPAA Compliance | Deep learning | Flags whether the agent confirmed the customer’s name and 2 pieces of PII before disclosing HIPAA sensitive information. |
Conversation Themes | LLM/Generative AI | Groups the primary topic of each conversation into a more digestible list of Themes, making it possible to report and filter by conversation topics. |
Conversation Topics | LLM/Generative AI | Surfaces 1-3 topics of conversation for every interaction with a customer, including a description of the topic. |
Conversation Summary | LLM/Generative AI | Generates a 3-4 sentence summary of every conversation through any channel with 95%+ accuracy, trained for healthcare. |
Agent Coaching notes | LLM/Generative AI | Generates coaching notes for an agent on a given conversation and surfaces recommendations for improvement. |
Starting & Ending Sentiment | Deep learning/Neural network | Labels the starting and ending sentiments of a conversation as Positive, Neutral, or Negative, by leveraging Natural Language Understanding and Data Labeling. |
Contact Type | Deep learning | Automatically identifies the primary persona present on every interaction as one of the following 6 contact types: Caregiver, Patient/Member, HCP, Payer, Internal, or Pharmacy. |
Voicemail | Deep learning | Detects interactions that are solely voicemail responses, optionally filtering these conversations out to focus on high value calls. |
IVR | Deep learning | Detects interactions that are solely interactive voice response (automated phone system tool used by call centers), and filters these conversations out. |
ℹ️ Learn more about different AI model types that power these models.
AI Model Performance
Sometimes, you may notice that a call that should have had an Eddy didn’t get labeled correctly by the AI. Or, you'll see that a voicemail that should have been filtered out by the Voicemail Model made it into the platform anyway. Why is that? Is something wrong with the model?
AI Models are Probabilistic, Not Deterministic
When you interact with an AI model (like the one behind chatbots or language tools), it’s important to understand that these models don’t follow fixed, pre-programmed rules like a calculator. Instead, they work based on probability.
Deterministic vs. Probabilistic
Deterministic systems: Always give the same output for the same input. For example, if you type 2 + 2 into a calculator, it always returns 4. There's no variation because it follows a fixed set of rules.
Probabilistic systems (AI models): Use statistical patterns from the data they were trained on to generate an output. If you ask the same question multiple times, you might get slightly different responses each time. You may notice this exact phenomenon if you ask ChatGPT the same question twice. That’s because the model calculates the likelihood of various responses and picks the one that seems most probable, but it doesn’t always make the same choice.
What that means for you
If you see that a particular call was labeled incorrectly by the AI, don’t fret! Our models are highly accurate (typically well above 90%). When the goal is to understand what is happening in your conversations at scale and where opportunities lie, the overall trends and accuracy are what’s critical.
🔍 Of course, if you notice something that looks off, please always feel free to escalate it to your Authenticx contact or [email protected].
In Summary:
AI models calculate the most likely answer based on probabilities learned from data, which means:
They can produce different results for the same input in different contexts.
They’re not always 100% accurate because they’re making highly educated guesses.
Their strength is in their ability to make decisions at scale, as well as their flexibility and ability to adapt to complex, diverse inputs.