Today’s post is by Ronald D. Southard, CEO at SafeSourcing Inc

During a recent post titled “Prompts developed correctly are critical to benefiting from AI tool usage Part I,” we discussed the use of AI tools and how to write a quality prompt in order to take advantage of what AI has to offer. We also discussed the importance of deep industry expertise when doing so. Simply stated, if you don’t know, how can you ask or evaluate.

There are a large number of generative AI models that you may be familiar with like ChatGPT, Copilot, Claude, Perplexity, Bard, Bing Chat and Gemini. These are all Large Language Model (LLM) based AI tools, our SafeSourcing process for developing your prompt strategy is largely the same for each.

To begin with, maintaining a reusable prompt library is critical to benefiting from these tools. It contains all of your learning so that you are not always starting over but creating an iterative process. Once that is in place, and certainly it will be tailored to how you work so your format will be framed to begin with. This will continue to evolve as you share learning with others and develop a company focused library.

Here are ten examples of how to build and build upon your prompts. Unfortunately, I am not including the SafeSourcing secret sauce. You can contact us for that information. This will however get you started.

1. Use a Structured Prompting Framework
   1. Break complex requests into clearly labeled sections.
2. Define Context and Assumptions Explicitly
   1. The more context you provide upfront, the less need for clarification or guesswork.
3. Incorporate Examples into Prompts
   1. When you expect a certain format or style, include an example:
4. Create and Maintain a Prompt Library
   1. Organize prompts by use case (e.g., Company Research, Supplier Sourcing, Market Intelligence, etc.)
5. Chain Tasks for Complex Workflows
   1. Break larger jobs into sequenced prompts.
6. Version Your Prompts
   1. As your tasks evolve, keep versions of your prompts.
      1. What worked for one version may not for another.
7. Anticipate Model Limitations
   1. Data cutoff for static knowledge (e.g., knowledge base date of last update).
8. Use Feedback Loops
   1. Build internal validation process.
9. Automate with Custom Tools or Templates
   1. Use tools like Python scripts to wrap and execute prompts automatically.
10. Train your team in Prompting
    1. If others in your organization use AI tools, consider developing a short prompting style guide:
       1. Standardize outputs.
       2. Avoid misuse.
       3. Improves collaboration.

SafeSourcing is a Procurement as a Service (PaaS)  provider of a number of e-Procurement solutions and associated white glove services that are part of our PaaS offered SafeSourceIt™ eSourcing suite. We have the tools, we have the knowledge, we have the experience, and we have the AI prompts that deliver results.

To learn more, please contact a SafeSourcing customer services associate.