Over the coming weeks I’m giving a series of 6 workshops about my promptEen prompt is de instructie die je aan een AI-model geeft zoals bijvoorbeeld ChatGPT. Het is hoe je communiceert met het systeem: wat je vraagt, hoe je het vraagt en... Meer method ECHO. I’m really looking forward to it, but it took some doing. Because not everyone was (and is) immediately convinced of what I’m trying to do with the ECHO Method. A good moment, then, to really walk through it.
What is ECHO?
ECHO stands for Ecosystem for Controlled Human/AI Output. Nice name, right? I think so too. I wanted it to be called ECHO because GenAI is an “echo” of what people do. See how people (human) come first? Yeah, that’s also a deliberate choice and as far as I’m concerned, a human-centered approach to GenAI is important.
But what IS ECHO?
Sorry, I digressed. ECHO consists of several parts that work together:
A component design architecture
A prescribed structure you use to build prompts. The structure consists of these components: context, perspective, role, instructions, output instructions, output example and a self-check. Not all mandatory, but all pure.
Modular and reusable components
Because the components are pure, you can reuse them nicely. Instead of always writing a new prompt from a blank page or prompts that are so contextual you can only use them once, ECHO helps you reuse your components (e.g., role: you are a communications advisor).
A component library
You store those reusable components in a component library so you can both manage them centrally and use them centrally. A guardrail changes? Then you adjust it in the component library and not somewhere in a Word doc.
Automatic or manual
That component library ensures that (if properly metadated), you can assemble prompts from individual components. That can happen in multiple ways. Users can indicate what they need and the system assembles the right components, or a prompt designer uses the components to put together ready-made prompts and for users the front end is then a prompt library. Handy.
A quality process and lifecycle management for your prompts
Yeah yeah, governance is boring. But suppose 200 people work at your company and each of them uses 3 prompts. That’s already 600 prompts. You want to manage those, because the quality of the prompts also has (alongside source, GenAI itself) impact on your output (apart from the fact that output often becomes input again).
There’s more, like the relevant roles you need to make this work and other relevant details. But then you have a picture.
So the idea of the ECHO Method is, if you completely flatten it, that prompts also need governance.
I see that governance everywhere on the other potential GenAI failure points: source (data), the LLMs themselves and the final communication (content), but not on the prompts, while they’re just as much a failure point.
Why ECHO exists
Partly because of what I just explained: your prompts are also a GenAI failure point. But besides that, I realized, as I saw more and more prompts in the wild, that the lack of a standard was causing quite divergent prompt quality. And without a standard it’s also difficult to compare prompt quality between prompts.
That’s not handy and not manageable, not auditable, not traceable and therefore there’s no governance. And you need to want that, especially if you operate in a regulated setting (banks, insurers, pharma, government etc.).
So I made a component design architecture (blah blah, an optimal structure). Based on industry best practices and existing research, but also my own knowledge and experience from the past 25+ years about what a good brief looks like. I translated all that to how that good brief should be filled in for an LLM: voila, a prompt standard.
I say voila and it felt voila. But it was mostly a lot of reading, searching, trying out, comparing and all the other stuff that ensured the prompts the standard delivers are actually effective. Let me reveal that 3 a.m. can be a productive time and Saturday got a new meaning during this period.
What do you get from ECHO?
This is what happens when you start ECHO-prompting:
- You can (if you want) reuse components
- Structured prompting is more effective than unstructured prompting
- You can more easily compare prompts with each other
- You can more easily figure out where something goes wrong in your prompt
- It’s a cognitive aid that helps you get all the relevant parts that should be in a prompt, into it
ECHO-prompting isn’t meant for your fun little chat with ChatGPT or Claude, by the way. I have those too and I do that without any standard whatsoever. It’s also not necessarily suitable for super creative tasks like brainstorming.
ECHO is meant for output that’s as consistent as possible and the place where that really contributes to efficiency is within business and especially within those regulated sectors. You deploy ECHO when GenAI use becomes embedded in your business processes, because that means you want the same thing for your prompts as you want for software.
I don’t want ECHO
Don’t do it then! But I do recommend that you (at least) adopt ECHO’s principles:
- A prompt standard (or multiple, ECHO isn’t the answer to everything)
- Reusable prompts (because who wants everyone separately tinkering away on hundreds to thousands themselves)
- And therefore at minimum a prompt library (central place where you store prompts and make them findable)
- Governance on your prompts (so you know what’s happening where, by whom and how)
- Structured prompts (because that’s more effective and more consistent than unstructured prompts)
With the work, experiments and research I’ve done over the past 6 months, I’ve learned that what I wrote down back then works in practice. I haven’t been able to test everything yet, of course, but the fact that I’ll soon have trained about 80 colleagues in the ideas behind ECHO makes me very happy. Because it wasn’t just hard work (that was fun), but also a lot of uncertainty (is it right?) and at times serious resistance.
After the workshops I’m going further with developing ECHO, because I’m not done yet. Apart from going deeper on existing parts, I’ve also been writing down new insights and parts over the past period (think orchestration and agentic deployment).
Don’t stress….it’s just me!
I’ve spent over 25 years working in content strategy and digital transformation, which means I’ve seen enough technology hype cycles to be skeptical and enough genuine innovation to stay curious.
Want to talk shop? Do get in touch!
I have a newsletter and it's bearable. Subscribe to read my (Gen)AI articles!
