Prompt Engineering for Customer Success (Part-2)

You know the feeling—you just finished back-to-back meetings, your brain’s buzzing with ideas, and you’re scrambling to jot everything down before the next call starts. It’s overwhelming, and sometimes, despite your best efforts, key follow-ups or insights slip through the cracks.

Your Personal Assistant for Meetings

There are some fantastic tools out there that record meetings, transcribe conversations, summarize key points, and highlight action items automatically. But what happens when those tools aren’t available? Maybe the meeting participants aren’t comfortable being recorded, or perhaps you want more control over what’s captured and how it’s structured.

That’s exactly where I found myself—not satisfied with one-size-fits-all solutions. So, I turned to something unexpected: a modern take on the dictaphone method.

Rediscovering the Power of the Dictaphone

There’s a scene in Revolutionary Road (2008) where Leonardo DiCaprio’s character records his thoughts into a dictaphone, leaving the transcription to an assistant. Watching it, I thought, “Why aren’t more people doing this today? Especially with AI in our toolbox, we all have access to an assistant—even if it’s virtual.”

Capturing thoughts right as they form is critical, especially in fast-paced environments like Customer Success and SaaS where ideas and action items emerge spontaneously. But traditional note-taking—whether typing on a laptop or scribbling in a notebook—often breaks the flow of thought. With so many distractions around, it’s easy to miss something important.

Some people record voice memos to solve this, but I wondered if I could take it further by combining voice capture with AI-powered transcription—creating my own personal assistant for meetings.

From Chaos to Clarity

During my time managing Customer Success accounts, I found myself juggling meetings with clients, partners, and internal teams—often with no breathing room between them. I needed a fast way to capture my thoughts after each call, ensuring nothing was lost in the shuffle.

I tried some AI meeting tools, but they often missed the mark. Many were designed for specific scenarios—like sales calls—and didn’t always fit the nuances of my role. Worse, they lacked the flexibility I needed to capture my afterthoughts—those “oh, I should’ve said this” moments we all get after a conversation ends.

That’s when I turned to OneNote’s transcription feature. After a call, I’d hit record, take five minutes to brain-dump everything—key points, action items, lingering questions—and let the transcription feature do the rest. The process wasn’t perfect:

• It occasionally struggled with industry-specific jargon and acronyms common in SaaS.

• My pauses as I gathered my thoughts often led to awkward punctuation or fragmented sentences.

Still, it was a game-changer—having my thoughts stored in a flexible, digital format made it easier to organize them later when it was time to send out follow-ups or update project trackers. But I wanted to take things a step further.

The Custom AI Solution

At the time, I had been reading about the AI gold rush to mine YouTube transcripts for training large language models, which gave me an idea: What if I fed my rough transcriptions to a custom GPT? By giving it specific instructions about my role, industry, and preferred note structures, I could transform my rambling voice notes into polished recaps—ready for internal alignment or external follow-ups.

Here’s a glimpse of the type of instructions I used for my custom GPT:

You will act as a personal assistant, specializing in summarizing meetings primarily between implementation teams and clients focused on building web experiences integrated on the clients' websites. This GPT specializes in converting raw text from transcriptions into well-organized notes. It's designed to understand the nuances of spoken language, especially those transcribed with irregular punctuation like excessive periods due to pauses in speech. Your role is to capture the essence of these discussions, which include technical aspects of product data and web development, while also occasionally covering marketing strategies for driving traffic. Summaries should be clear and concise, formatted as bullet-point executive summaries with each point no longer than two sentences. Use general language for broad understanding but ensure specific jargon, datapoints, or web development components mentioned are accurately documented for clarity and reference. In handling text, it corrects punctuation errors and asks open-ended questions to clarify misheard words or inaccurately picked up acronyms, leaning towards making educated guesses to maintain the flow of conversation. It maintains a professional yet approachable tone, suitable for an office environment or personal note-taking, adapting its response based on the user's preference for note style when specified.

A Chef’s Approach to AI

This isn’t the most perfect or optimized prompt, but it got the job done. It masterfully transformed rough transcriptions into structured recaps that anyone could follow, with action items clearly laid out.

Of course, you never want to send things straight from a genAI model to clients without review. But being an editor rather than a writer takes a lot less effort. Think of AI as your sous-chef: it’ll prepare all the ingredients, but you need to expo the dish to the table. And if you’re a fan of cooking competition shows like me, you know how critical it is to taste what your team is putting out.

So adjust the seasoning to taste, add some garnish by tagging those responsible for action items—and bon appétit!