FirstWatch AI Functionality
FirstWatch is pleased to introduce generative artificial intelligence (“Gen-AI”) capabilities to help you explore your data more thoroughly and derive important insights into your operations and performance. To provide this functionality, FirstWatch has partnered with AWS and Anthropic. AWS, the world’s leading cloud services provider, is well known for its thorough, proactive, and clearly-documented approaches to data security in the cloud, while Anthropic’s Claude Sonnet is a best-in-class large language model (LLM) across a wide range of industry benchmarks.
FirstWatch’s initial “AI Beta” functionality focuses on helping you explore your past trigger data (for as far back as it is saved) through ad hoc questions. For example, if you have a trigger that lists all of your incidents, you might ask a question such as: “How many responses did each of my units have per month in 2024?” Or, as another example, you could ask: “For each unit in 2024, provide a count of the responses for which it took more than 30 seconds for the vehicle to be en route.”
The way that the functionality works is that the question you ask, along with certain information about the trigger (such as what column information is available for the trigger), will be passed along to the Claude Sonnet model. Claude will then be asked to create an SQL query statement that can be executed against your trigger data to answer the question that you have asked.
If Claude is unable to create an SQL query based on your question, it will let you know and may ask for additional specific information to help it form a query. Assuming that Claude is able to create the query, however, and that the query returns one or more records in response, you will be shown:
- A natural language version of the SQL query that Claude generated, so that you can review its logic and make sure that it understood your question correctly;
- A summary description of the data that was returned by the query; and
- A listing of records returned by the query, which you can then choose to view in the Events page.
Once this information has been provided, you can choose to ask a follow-up question, if you like, or start over to ask an entirely different question or to switch to a different trigger.
Protecting Your Data with AI
When it comes to machine learning and generative AI models, a common and legitimate concern is whether an organization’s sensitive data will improperly be used to train a model. This is definitely NOT the case with FirstWatch’s AI functionality. Together with AWS and Claude, FirstWatch has instituted several safeguards ensure that your sensitive data can only be accessed by authorized individuals:
- User- and organization-based access permissions. FirstWatch’s AI Beta application observes the same data access permissions as all of FirstWatch’s other applications. If a particular user is unable to access a given trigger or can only view a de-identified version of the trigger data, then the same restrictions will apply when a user wishes to explore trigger data via the AI Beta functionality.
- Data encryption in transit and at rest. As part of the AI Beta functionality, trigger data is temporarily loaded from the FirstWatch data center into AWS storage. To ensure appropriate protection, the data are encrypted when they are uploaded to AWS, and then encrypted again when they are stored. Data are then swiftly deleted once they are no longer needed (typically within a couple of hours after use of the AI Beta application).
- HIPAA-compliant data protection. To ensure that sensitive health-related and personally identifiable information receives adequate protection, AWS offers HIPAA-compliant cloud services and can provide a business associate agreement upon request. FirstWatch relies exclusively on HIPAA-compliant services for its AI Beta functionality.
- No AI training with trigger data. For the AI Beta’s SQL generation functionality, FirstWatch is relying on a “static” or “frozen” instance of the Claude Sonnet model (that is, a version of the model that is no longer being trained) that resides within the AWS cloud. Additionally, AWS has configured this instance such that (a) no data can ever be passed from the model back to Anthropic, the company that created and trained Claude, and (b) no information that is passed to Claude Sonnet or returned from Claude sonnet can be written to log files or permanently stored in any other way.
In short, FirstWatch understands the critical importance of protecting your data from unauthorized access and has designed the AI Beta application to with security at top of mind.
What Kind of Questions Can I Ask?
You can ask all kinds of questions about your trigger data. The categories below provide examples of different types of questions that you might ask to get you started. Note as well that you can combine questions from multiple groups – for example, grouping incidents by month and then computing average response time by month.
Apply Filters
- Show me records for Battalion 4
- Only include incidents involving a traffic accident
- Only show records for 2024 to date
Perform Calculations
- What’s my average dispatch time?
- How many hospital transports have we provided this year?
- What’s our average number of responses per incident?
Aggregate Records
- Group incidents by dispatcher
- Group incidents by month
- Group responses by unit
