Drafting Prompts: Best Practices and Guidelines

Crafting effective prompts is crucial for maximizing the accuracy and efficiency of eDiscovery AI. A well-constructed prompt guides the AI in understanding the specific context of your case and the criteria for determining document relevance. Follow these best practices when creating your prompts:

1)      Be Clear and Specific

a)      Clearly define what constitutes a relevant document in the context of your case.

b)     Use precise language to avoid ambiguity.

c)      Provide specific examples of relevant and irrelevant content when possible.

d)     Example: "A document is relevant if it mentions the acquisition of Company X by Company Y between January 1, 2022, and December 31, 2022. This includes emails, memos, and financial reports discussing the terms, negotiations, or due diligence related to this acquisition."

2)      Include Key Terms and Concepts

a)      List important terms, phrases, or concepts related to your case.

b)     Explain any industry-specific jargon or abbreviations that might appear in relevant documents.

c)      Example: "Key parties include: Company X, Company Y, Consulting Group Z. Discussions of ‘due diligence’, ‘stock purchase’, M&A’, and ‘board approval’ are related to the acquisition. The abbreviation 'DD' may be used for due diligence."

3)      Specify Exclusions

a)      Clearly state what types of documents or content should be considered Not Relevant.

b)     Address potential false positives that eDiscovery AI should avoid such as closely related but irrelevant content or exceptions to the relevance criteria.

c)      Example: "Exclude general news articles about the industry that don't specifically mention the acquisition. Also, exclude routine administrative emails unrelated to the acquisition process."

4)      Provide Context

a)      Give eDiscovery AI background information about the case or investigation.

b)     Explain the significance of certain document types or content in the context of your review.

c)      Example: "This case involves allegations of insider trading related to the acquisition. Pay special attention to communications between Company Y executives and external financial advisors in the months leading up to the public announcement."

5)      Address Nuances

a)      Explain any nuanced situations where relevance might be less straightforward.

b)     Provide guidance on how to handle documents that are partially relevant.

c)      Example: "If a document contains both relevant information about the acquisition and unrelated content, consider it relevant. Documents only discussing potential acquisitions of other companies are not relevant."

6)      Iterate and Refine

a)      Start with a basic prompt and refine it based on sample results.

b)     Be prepared to adjust your prompt multiple times to improve accuracy.

7)      Collaborate with Subject Matter Experts

a)      Work with legal teams and subject matter experts to ensure the prompt accurately reflects the case requirements.

b)     Have multiple team members review the prompts to ensure they are clear and comprehensive.

8)      Test with Diverse Samples

a)      Apply your prompt to a diverse set of sample documents.

b)     Include both clearly relevant and clearly irrelevant documents in your test set, as well as some borderline cases.

Remember, prompt creation is an iterative process. Regularly review and refine your prompts based on the AI's performance and any evolving case requirements. A well-crafted prompt is key to achieving high accuracy in your eDiscovery AI.