Prompt Engineering

As we are in the frenzied period of maximum hype in relation to all things Artificial Intelligence (AI), I am writing and continuously updating this blog post. This is where I am using ChatGPT-4.o and applying Prompt Engineering to solve a range of everyday challenges in sales.

Our focus here is business; specifically the challenges faced in everyday life by people engaged in a high-value, high-touch sell. In this context we treat Chat GPT-4.o as the 'Co-pilot' in learning how to increase sales effectiveness. The results so far have been mIxed.

Introduction

Prompt Engineering in the context of a high-value, high-touch sell refers to the strategic formulation of prompts or queries to elicit the most effective and useful responses from advanced AI systems, such as Chat GPT-4.o, particularly in enhancing sales processes and customer interactions.

1. Role of AI in Email Marketing

AI tools, like ChatGPT 4, can assist in various stages of the sales process, from lead generation to closing deals. They can analyse vast amounts of data, provide insights, generate leads, personalise communication, and even predict customer needs. This, in my experience is ineffective: what ChatGPT 4 generates as email content feels inauthentic and is bloated.

2. Prompt Engineering for ROI Models 

Prompt Engineering is the skill of crafting questions or prompts to an AI system to generate the most relevant and useful responses. It involves understanding how the AI interprets and processes language, and using this knowledge to frame prompts that lead to desired outcomes. The best example I generated with Chat GPT-4.o iterative prompts was ROI Models for salws forecasting.

3. Salesforce and AI - So Far

There has been a lot of hype recently about how AI and Prompt Engineering can speed-up the design and configuration of a Salesforce application This includes Salesforce themselves, and a number of AppExchange software publishers announcing new products, such as Einstein GPT.

This is not a review of Einstein GPT but a simple, ongoing assessment of ChatGPT-4.o as a Co-pilot in the varioius tasks related to tailoring Salesforce Sales Cloud. My experience here was mixed.

Flow Automation Design: Although Einstein GPT will provide an integrated approach to combining Prompt Engineering with Flow (Salesforce workflow ) design, I used ChatGPT-4.o to act as Co-pilot in a manual approach to describing each Action or Update in the Flow. It took several iterations of Prompting to get correct answers. This highlights the fundamental problem with Flow - you need to have a 'tech analyst mindset' to understand how to use ChatGPT-4.o as a Co-pilot in creating a Salesforce Flow.

Field Formula Creation: ChatGPT-4.o works well for figuring out formulas applied to Salesforce Formula Fields. This is conceptually no different to acting as a Co-Pilot for Excel or Google Sheet formula design.

Software Test Automation: ChatGPT-4.o can act as a Co-Pilot to deliver improved prompys with the Gherkin business language - combined with Conformiq Visualizer and Atlasian Jira.

4. AI in a High-Value Sell

This sales approach focuses on significant transactions, often involving complex solutions and a substantial investment from the client. The 'high-touch' aspect implies a personalised, consultative approach, where building relationships and understanding customer needs is paramount.

ROI Modelling: As stated above, this is where ChatGPT-4.o can act as a Co-pilot, creating ROI Models which buyers can provide a Buyside Forecast to compare with a Sellside Forecast. This works well and you can see detailed examples in my post on sales forecasting.

Lead Qualification: Crafting prompts to identify and prioritise high-potential leads from a large dataset. This highlights a more fundemental issue: what does good look like? This requires further experimentation.

Email Communications: Generating customized emails or messages that resonate with specific clients, using prompts that incorporate customer-specific data. ChatGPT-4.o is really bad at crafting emails: as note above, it produces waffle - bloated words and sentences - feels totally Inauthentic.

Market Analysis and Insights: Asking the AI to analyse market trends and provide insights relevant to a particular customer or industry. As with Lead Qualification - this highlights a more fundemental issue: what does good look like? This requires further experimentation.

Agreement Generation: Creating detailed, tailored proposals by prompting the AI with specific customer needs, industry specifics, and solution parameters. So far, ChatGPT-4.o as a Co-Pilot offers interesting 'devil's advocacy' on the non-legal aspects of terms and conditions in an Agreement.

Handling Objections: Using AI to simulate various customer objections and crafting prompts to find the best counter-arguments or reassurances. This is a work-in-progress - updates to follow.

5. Skills Required in Prompt Engineering:

Linguistic Precision: Understanding how different phrasings can lead to varied AI responses.

Contextual Awareness: Framing prompts that take into account the customer's industry, history, and specific needs.

Creativity: Thinking 'outside-of-the-box' to ask AI for insights or data analyses that might not be immediately obvious.

Technical Understanding: Knowing the capabilities and limitations of the AI tool being used.

6. Benefits:

Efficiency: Streamlining the sales process by quickly generating high-quality, relevant outputs.

Personalisation: Enhancing customer engagement through highly tailored communications and solutions.

Data-Driven Decisions: Leveraging AI's data processing capabilities for more informed decision-making.

7. Challenges:

Over-Reliance on AI: Risk of losing personal touch which is crucial in a high-value, high-touch sell.

Misinterpretation: Incorrectly framed prompts can lead to irrelevant or misleading information.

Ethical Considerations: Ensuring the use of AI aligns with ethical standards and privacy regulations.

Reflections

The answers were greatly improved by clarity or as noted above by 'linguistic precision'.

I think that the best practices most relevent (so far) are:

1. Start with 3 things in mind: clarity; context; and, creativity.
2. Linguistic precision for inputs. Clear language, shorter, be specific.
3. Split complex tasks into simpler sub-tasks. Describe the steps.
4. Create a role (persona) for identity: e.g. "As a Chief Revenue Officer ...".
5. Apply parameters: specify the desired length of the output.
6. Be iterative: add more context or key words over time.
7. Ask AI for suggestions: "How do I ...".
8. Nest prompts: two or more in new prompt (list n facts, write content on each).
9. Creatre chain-of-thought prompt: e.g. audience, USPs, channels.
10. Embrace AI non-deteminalism: same input, two or more different outputs.

More to follow ...