Generative AI models have transformed how we interact with technology, offering impressive capabilities in text generation, analysis, and decision support. However, the quality of the model's output depends largely on the training data and context provided.
This article explores a few possible reasons the Revenue AI bot's reply may deviate from an expected response and what you can do to improve your results.
Context Windows
One of the fundamental constraints of Generative AI systems are their context window - the amount of input data they can process at once. Think of this like a moving spotlight on a stage: while the spotlight can illuminate a specific area brilliantly, anything outside its beam remains in darkness.
Ask a Conversation
When collaborating with the Revenue AI bot at a Conversation level, the bot will have access to the full conversation transcript in order to provide detailed answers about the conversation. Additional context outside of the Conversation (such as previous conversations, Salesforce data, etc) is not included at this time.
Ask an Opportunity
When collaborating with the Revenue AI bot at the Opportunity level, the bot will have access to the most recent 125 activities associated with the Opportunity and Conversation Summaries.
Audio and Conversation Quality
When a conversation is recorded, the audio must go through several transformative steps before an AI can work with it. First, the audio is captured through a microphone, which may vary in quality and sensitivity. This raw audio then needs to be transmitted over the internet and across the carrier network. Finally the recording is converted into text through a process called transcription.
- In an ideal situation, with speakers in a quiet room using quality microphones, the AI receives an accurate transcript that closely matches what was actually said.
- However, if there's background noise or audio degradation, some words might be missed or misinterpreted, leading to gaps or errors in the transcription. The AI then has to work with this imperfect text, which can affect it's understanding and responses.
To maximize the effectiveness and accuracy of Revenue AI, Ensure your users are using appropriate hardware (headset/microphone) which minimize background noise and maximize speaking clarity.
Transcript data and single-sided call recordings
Additionally, if a meeting or call participant doesn't speak any words or the relevant conversation is a single-sided call recording, some context may not be captured. For example, if you ask if an individual was present for a particular meeting, but they never spoke, the bot would not have the context necessary to know they were a participant in the meeting.
Model Training and Knowledge Limits
Generative AI systems operate within the bounds of their training data. Imagine an AI model as a very knowledgeable person who went into isolation on a specific date - their knowledge cutoff date. They know everything up to that point in great detail, but nothing after it unless someone specifically tells them about it.
The most straightforward way to work around knowledge limitations is through external context. Think of this like bringing a newspaper to our isolated friend. You might need to briefly explaining the industry a company operates in, key market trends, or common pain points in the industry. You might also find it helpful to explain industry or company specific terminology that took place in the conversation for a more accurate output.
Submitting Feedback
If you encounter problems with bot accuracy and your experience is not explained by the reasons listed above, please submit feedback in-app using the feedback buttons found on the bot's response.