Leveraging AI for next-level customer engagement: A guide to prompt engineering

Personalisation meets innovation

Artificial Intelligence (AI) is redefining the landscape of customer engagement. Recent innovations in AI perfectly align with the goal of creating no-code workflow solutions that empower you to craft highly personalised communications.

Twilio’s latest 2024 State of Customer Engagement Report shows that the more personalised a customer's experience is, the more they’ll spend on a business. In fact, 64% of consumers claim they’ll abandon a business if personalised experiences don’t exist… Personalisation in the age of AI has never been more important; it’s an expectation from all customers.

So, what does personalisation in the age of AI look like? Some examples include:

• Analysing customer usage patterns and suggesting personalised upgrade options in the telecommunications space.
• Scanning transactions for unusual activity and flagging a potential fraud case for review in the finance sector.
• Identifying appointment requests and scheduling them based on doctor availability and patient preferences in the healthcare space.

In this blog, we'll delve into the latest AI tools for CX and marketing leaders, through the lens of prompt engineering; how YOU can start to experiment with AI, level up your personalisation game and design groundbreaking new experiences for your customers.

AI isn’t magic; it’s a tool that can be used well, or not so well…

You may have heard the data science phrase “garbage in, garbage out”; which resonates with the swathe of generative AI tools that have been released in the last two years.

Prompt engineering is the art and science of crafting inputs that guide AI models to generate desired outputs. Inputs are things like questions you might ask to a tool like ChatGPT, “How do I write an effective opening to a blog post?”, and outputs would be the answer.

Outputs are heavily guided by inputs, and bad data leads to bad outputs. Additionally, easy things to fix like run-on sentences, double-negatives, vague instructions among other things can heavily affect the quality of your outputs. For using tools like ChatGPT, DigitalOcean has a great article on prompt engineering do’s and don’ts.

Prompt engineering is all about ensuring that you aren’t putting “garbage in”. It’s crucial to understand how to be a good prompt engineer in the realm of customer engagement.

Here’s an example of a not so good single step prompt: "Flag any messages that you think need following up".

Note the vagueness of the user prompt, “any”, “you think” and the reliance on the AI to ‘know’ the context without providing any:

Here’s a better single step prompt. Notice how the user provides not only the categories, but also the conditions in which to organise each category as a document. When searching for AI workflow tools, consider options like the Pendula’s Intelligence Suite, which allows you to incorporate resources such as FAQs and other documents to add context to your AI workflow steps.

Prompt engineering made easy

We have a suite of intelligent workflow steps available, which are our Intelligence Nodes, that can be shaped (through effective natural language prompting) to perform repeatable tasks for individual experiences, such as:

Understanding the intent using natural language processing, of a customer’s inbound requests, or inbound replies to messages you’ve sent out, and routing them in the workflow accordingly; all without code.

Analysing survey results, determining sentiment, and if needed, serving relevant parts of FAQs or other documents to customers at scale on the channel of your choice.

You can also use intelligence nodes to operate processes that improve your efficiency as a business. You could use an intelligence node to summarise all of the interactions, and system updates, that have occurred in an experience so far; for instance to record a copy of in a CRM, or to update support notes.

You could even have Pendula AI write your messages back to individual customers, for unsurpassed personalisation at scale. While this may sound outlandish, CapGemini says that 62% of customers (and always growing) are “comfortable” with Generative AI in marketing, and a similar percentage do not mind being recommended product recommendations by AI. This kind of experimentation and learning will level up your customer engagement and keep you at the forefront of Generative AI marketing.

Nodes in the Pendula Intelligence suite have been designed to allow you to instruct them in the way that serves your use case best; you can manipulate or change the prebuilt prompts, or start from scratch. The best part; you can do it all by typing in plain language, no code here. The key is understanding how to prompt effectively and efficiently.

How to craft effective AI instructions

Now let’s break down three different ways you can prompt Intelligence nodes:

• Single Step: These are single instructions given in an Intelligence Node to generate a response or perform a task without prior examples. They are straightforward and concise. These can be useful to start experimenting with the power of Pendula AI, but don’t scale as well (they are less reliable).
• Multi Step: These include multiple steps, and a few fully written examples, with defined roles, to guide Pendula AI to make the best decision in that Intelligence Node. By providing defined examples of how Pendula AI should act when given dummy data, or when given real data, Pendula AI will follow the structure and pattern of the provided examples.‍
• Daisy Chaining: You can even use the outputs of previous Intelligence Nodes to inform later Intelligence Nodes. Because Pendula flows are infinitely extensible, you can chain as many nodes as you like!

Personally, I find that using multiple steps, with clear examples of format, dramatically improves the results of my Intelligence Node based workflows, an example provided below where I have specified:

• Responding in all caps
• Putting the full name of the day before the time
• 24 hour time

…Just by providing one example. Pendula AI is smart enough to learn from these, and even nuanced patterns of examples, and the more examples (more steps) the better.

There’s no better time to start with Pendula AI

Pendula’s Intelligence Suite lets you start integrating artificial intelligence into your workflows as much, or as little, as you like, without coding. We understand that AI is a technology with customer-facing touch points that requires experimentation and a level of consideration. Our Intelligence Suite was designed with this mind.

From single prompt workflow steps, through to multi-step, daisy chained intelligence steps that take real-time data to use in their decision making, Pendula AI scales with you.

When you’re ready to step up your customer engagement strategies, we’re here with the solution. Talk to an expert about how Pendula can improve your customer engagement and take advantage of our leading edge technologies such as the Pendula Intelligence suite.