Innovating Investor Relations with AI: A Journey from Hackathon to Prompt Engineering

More and more each day, LLMs are becoming a part of how we create and interact

Following Q4’s “Innov8 Days” event on March 23rd, 2023, a competitive and innovative gathering reminiscent of a hackathon, Q4’s interest and excitement in AI has substantially grown. This event sparked an innovation drive within our organization, leading us to invest significant time into the potential of AI-driven features, aimed at augmenting the value we bring to Investor Relations Officers (IROs).

Embarking on the journey into AI is a venture filled with both excitement and challenges. One of the core competencies needed is a solid understanding of Large Language Models (LLMs) and their inherent strengths and limitations. For instance, we quickly realized that while AI models excel at processing language, they can struggle with numerical operations and tabular data, leading us to adapt our approach.

One cardinal rule that we’ve adopted is to prompt the AI to write as little as possible, especially when response time is vital. Coupled with an understanding of token limitations and data quality, this forms the backbone of effective AI deployment.

Our foray into AI applications introduced us to an intriguing concept — Prompt Engineering. This involves the art of crafting effective prompts to guide AI towards generating the desired output. It is akin to programming the AI and is pivotal in creating AI applications that are consistent, effective, and less prone to errors.

The Road Trip Prompt

To illustrate this, we have a creative prompt to share — the Roadtrip prompt. The task was to plan a comprehensive road trip across the southern United States, complete with an itinerary in an ICS file format. The AI’s response was an elaborate and detailed road trip plan. While not perfect, it demonstrated the capability of AI models to generate diverse outputs.

You are a master road trip planner and your task is to plan a road trip across the southern United States for a pair of TRAVELLERS looking to visit ICONIC LOCATIONS and more. You will follow a set of PARAMETERS that will guide you into the creation of the road trip in ICS file format.

TRAVELLERS:

The two travelers are equipped with a small RV. They are husband and a wife celebrating their anniversary. The husband and wife enjoy a good long walk and would enjoy any attraction that might require a walk. The husband and wife will fly out to the starting location and rent an RV there. Make the starting location something iconic along the south east side of the country. 

PARAMETERS:

1. Starting date is Monday, August 28th, 2023
2. This is an 11 night long road trip
3. The starting location is Atlanta, Georgia and the final stop is in San Diego, California
4. The plan includes events for the activities or visits as well as the driving. To explain further, the plan should include activities like "[Driving] Salt Lake City to Boneville Salt Flats". This is done to give the TRAVELLERS indication of when to start to drive to the next location in the plan.
5.  Each overnight stop will be near an iconic location and located in a safe place to park overnight
6. Focus on what the best of each stop has to offer in terms of attractions. 1-3 attractions per day is a good target to aim for. Deprioritize museums and prioritize iconic tourist destinations. 
7. Include breakfast, lunch, and dinner destinations based on amount and rating of google reviews (from your memory)

ICS:

Your task is to output the road trip in ICS format. Each event that you generate should include the location data and an appropriate title with the type of event as a title prefix. In addition to this, each event should use the time zone appropriate to the location.

ICONIC LOCATIONS:

Here is a list of locations that the TRAVELLERS feel that they must visit:
- Boneville Salt Flats
- Monument Valley

What’s fascinating is how this seemingly light-hearted experiment can be applied in an IRO’s world. Consider drafting an IRO itinerary for a Roadshow, a series of presentations to potential investors. A similar prompt could generate an itinerary that aligns with the IRO’s specific objectives, saving them considerable time and effort.

A sample of The Road Trip Prompt’s output

AI offers a wealth of advanced capabilities that can enhance many software development cases. A technique we’ve found particularly valuable is the use of micro-decision prompts. These isolated decision-making prompts can yield swift and consistent outputs, making nuanced decisions one step at a time.

I am a cuisine expert AI and my task is to choose the most appropriate tool from the following list to answer the last message in a conversation between the user and their cooking assistant. I should not make assumptions about what tool could be helpful, but rather choose a tool based off of the direct message that the user is trying to convey.

The conversation:

<insert the tail end of the current conversation between the user and their assitant>

I have to determine that, considering the conversation's final message, what is the number of the tool from this list of tools that should be used?

Tools:

1. Recipe lookup: This tool is very useful to looking up recipes on the internet
2. Ingredient detail: This tool is useful for retrieving detailed information about a certain ingredient
3. Measurements: This tool is useful for working with measurements such as converting ounces to liters
4. Nothing: This option is available for when the previous three are not applicable such as if the user is asking an off-topic question or just looking to chat. It should be used in almost all cases except when the user explicitly asks for a recipe, details on an ingredient, or measurement help.

Number:

The principle of breaking down complex tasks into smaller, manageable prompts has influenced our work in AI and our broader approach to software development. It allows us to leverage AI as a coding companion, assisting in tasks such as UI development.

As we delve deeper into the realm of AI, its applications in Investor Relations are becoming increasingly clear. It enables faster, more efficient data analysis, paving the way for innovative offerings for IROs. However, our role as engineers in validating AI outputs remains crucial to maintain data integrity. The potential of AI is enormous, and as a team, we are eager to explore the developments and advancements this exciting field has to offer.

For a bit of interesting insight, this blog post was largely written by an AI model, ChatGPT-4. By providing it with a set of guidelines or prompts, it was able to gather the necessary information and structure this post. You can see this interactive process in this ChatGPT Conversation. As we’ve learned, splitting a task into a series of prompts can be more effective than a single, seemingly comprehensive one.

In our ongoing work with AI at Q4, we’re not just testing the waters — we’re diving in headfirst. We’re excited by the possibilities and committed to pushing AI’s potential in Investor Relations. Our goal? To continually innovate and deliver the best possible experiences for our clients.