A Top Technology Initiative Article. June 2023.
With the spring conference season wrapping up, watching the frenzy around Artificial Intelligence (AI) and its applicability in the profession has been amazing. As noted in my April column, there are various innovative uses of AI and privacy risks for your client data. Recently, Google released its SAIF (Secure AI Framework) regulations based on six principles to enable the creation of high-quality, secure artificial intelligence.
We believe that the privacy risk to client data is high, but some consultants seem to be ignoring this exposure to client confidentiality. Thought leaders and novices are working to convince you they have AI figured out. AI applications are evolving so rapidly that it is difficult to imagine all the possibilities.
If you have developed a use of AI that you’d like to share, please email me at email@example.com. I will attribute the idea to you and add it to my warehouse of AI ideas. I intend to develop a cookbook of AI prompts and applications that will have practical application in your firm.
As a reminder, colleague Brian Tankersley and I have recorded five podcasts on the topics of AI with ChatGPT4, Microsoft AI, DALL-E & AI Competitors, AI Truthiness & Hallucinations, and Large Language Model (LLM) considerations. Please ensure you have checked out these AI podcasts and our podcast discussions of various other products at The Technology Lab.
What Are the AI Trends That Have Recently Surfaced for Accounting?
Remember, your imagination, a systemic approach, and the application of prompts for the AI Engines are crucial to your success. Vendors of products suggested these uses in their platforms to me in the last month.
- Applying tax planning strategies
- Analyzing audit data (Fieldguide.io released AI in their platform)
- CAS trend analysis for cash flow, inventories, and other metrics
- Explaining concepts and services to clients in their language and style
- Lease analysis
- Form creation and extraction
- Data interpretation
- Document analysis and retrieval
- Workflow optimization
- Macro creation
- RPA creation
- Draft procedures
- Draft engagement letters
So far, the best way I have found to structure my AI prompts is in a spreadsheet, the sledgehammer of accounting. I have created prompts to generate bulleted lists, presentation slides, and other items that I will illustrate below. Remember that most Large Language Model (LLM) Generative AI platforms like ChatGPT are like an assistant you must train. Providing feedback and guidance helps you get better results.
What Are Some Suggestions for Training AI Models?
First, be aware that my methodology is evolving, but I created a prompting model that includes: User Input Controls, ChatGPT Controls, and ChatGPT Commands. Each of these areas has developed to minimize my typing. Frankly, I’d rather speak to ChatGPT, but tools like Speak to ChatGPT, TalkBerry, or Speechify don’t seem to work well for me. But, neither does ordering from a drive-through window since most workers can’t understand my voice over the intercom.
Further, recognize that our K2 organization tends to write CPE materials, so our ChatGPT prompts tend to support those efforts, eliminating writer’s block and overcoming the blank sheet of paper syndrome. Each of the ChatGPT Commands (prompts) is generated after naming the Input and ChatGPT controls. Our User Input Controls include:
Specific Slide Controls
- Slide Headline/Topic–>
- Bullets/Ideas –>
Overall Presentation Controls
- Presentation Title –>
- Learning Objectives –>
- Course Description –>
Our ChatGPT Controls include the following with the ability to change the lengths stated:
- Output Qty –> 5-6
- Body Sentence Length –> 7-10
- Title Word Length –> 4-6
… and our ChatGPT Commands include:
Bullet Point Commands
- Long Bullets – Slide Title Only
- Short Bullets – Title and Idea Points
- Short Bullets – Practical Examples
- Short Bullets – Facts and Stats
- Come Up with Ideas – Company/Product – Features, Functions, etc.
- Provide Examples
- Rewrite and Rephrase Existing Slide
- Rewrite Last Thing into Bullet Points
Review Question Commands
- Simple Review Question – Title Only
- Simple Review Question – Title and Ideas
- Review Question on Recent Conversation
- Review Question on Specific Topic
Generate a Table of Commands
- Generate a Table & Headline
- Compare and Contrast w. Headline
- Generate a Table w. Ideas
Generate a Graph Commands
- Generate a Graph
- Title for a Slide
- Subtitle for a Slide
- Title and Subtitle
- Summary / Overview Slide
Presentation Overview Commands
- Intro Seed Data to Chat GPT
- How I Want My Bullet Points
- Ask for Ideas
- Upcoming Trends
The goal of these prompts is to quickly get ChatGPT to provide results in a format that is immediately useful. I have found that I can create content in 25% of the time (yes, ¼) this year compared to last year. Additionally, I believe the content may be superior.
So, What Can You Do to Learn AI?
Start. Be methodical. Be specific. Teach ChatGPT to be your assistant. Develop a firm-wide template to get consistency across your team. Learn from each other.
Each person’s AI journey will be different. There is no right and wrong way to use the tools, but there may be better ways that we learn over time. By creating your own prompting model, you will save time and get better, more consistent results. Finally, you may be surprised how much better work you can do and how quickly you complete items. If you find a better way, I hope you will share it with me, like I will share my best ways to use AI with you.