To leverage and scale AI, we need to get the work out of people’s heads. But defining our work is more difficult than most realise. In this article I share a simple rubric that can be used to support our colleagues so they know how to break the work down and what else to include so AI can execute it.
This is part of a series on building workflow capability for enterprise AI:
Why enterprise AI will fail without workflow capability: why people need to be able to define their work before AI can help.
How to design an enterprise AI workflow capability program: how to build that capability across an organisation.
This article zooms in on a key resource that people need to effectively define their work.
To get value from AI, you first have to define the work you want it to do. And that turns out to be harder than it sounds. Because being able to do something is very different from being able to explain how you do it. It's a skill most of us have never needed, and it's the skills gap causing most of the disappointing AI results.
In [How to design an enterprise AI workflow capability program], I outlined how to close this gap across an organisation. This article zooms in on one of the resources that makes it work: a simple rubric for defining work.
What people miss when they define work
When people try this, I see the same two things missed time and again:
First, they stay too high level. “Analyse spreadsheet for trends” feels like a clear instruction to the person writing it. But it’s actually several smaller steps bundled into one. Hand it to a new starter (or AI) and they can’t tell what those steps are, because everything that sits within that instruction only exists in the author’s head.
Second, they describe the ‘what’ but not the ‘how’. Unless we provide context, AI has to fill in the gaps. And we know the consequences of that. So as the experts, we must provide that context. How do we execute each step? What decisions are we making? Against what criteria should we follow? What exceptions should we consider? Which format we use. This detail is usually so ingrained, people don't realise it's missing.
The good news is we don't need a complex methodology to fix this. A simple rubric gives people exactly what they need: one test for how far to break the work down, and a few questions to ask about each step.
And this works whether you're defining work for a colleague or for AI.
Here's how it works.
Part 1: How far do we break the work down?
Without thorough analysis, there is no scientific way to know whether we’ve broken a workflow down into enough detail for someone else (or AI) to execute each step.
However, a simple heuristic is to consider whether the workflow can be executed alone and within a few hours. Super crude, but just like Amazon’s ‘two-pizza team’ rule, this will get us in the right ballpark.
Now, using this lens, let’s consider to what extent we need to break down the parts of our workflow so that someone else (or AI) could do the work effectively.
To make this concrete, let’s use an example workflow - the following is a workflow someone might follow to make a video.
Workflow:
Brainstorm video ideas
Research high-ranking titles
Write script
Design storyboard
Set up camera and lighting
Record video
Edit video
Write description
Publish to YouTube
Share on socials
Now, to effectively define the work, we need to break each of these down further. Why? Because this might involve some back and forth between people (scriptwriter, on screen talent, videographer etc.) and may take a few days to deliver.
So, let’s break Brainstorm video ideas down into steps.
Steps:
Gather audience pain points from comments, customers and the team
Research what’s already working on YouTube
Shortlist ideas and agree them with the team
But is that enough granularity? I think we should go even further. So, let’s break down the Research what’s already working on YouTube step into actions.
Actions:
Enter relevant search terms into YouTube
Review top results
Identify standout videos
List recurring patterns
Save the strongest ideas
How do we know this is granular enough?
Well, I could imagine one person doing this within a few hours.
Part 2: What does each action need?
Just like there is no scientific way to know whether we’ve broken a workflow down into enough detail for someone else (or AI) to execute the steps, the same could be said for knowing how much detail to add to each action.
However, a really simple heuristic is to think about a capable new intern who has recently joined your team. They’re enthusiastic, but have never worked in this role before and have no prior experience at your company.
What would they need to do the work without pestering us for answers?
Well, it can be helpful to map the requirements against four categories: context, instructions, guardrails and blueprints/templates. These categories translate the core principles of performance design into something practical i.e, aligning the work, guiding the action, defining the boundaries, and providing a structure for consistent execution.
So, let’s consider what that might include for the Identify standout videos action, from the step we just broke down.
1. Context
Audience interests - video ideas should align with these
Channel’s typical content - standout videos should outperform these
2. Instructions
Compare each video’s views with the channel’s recent average
Flag any that are unusually high
Note what they have in common (topic, title, format)
3. Guardrails
Videos must be relevant to audience
Performance must be judged relative to channel
Ideas should be adapted rather than copied
4. Blueprints and templates
Table showing title, channel, views, typical views and underlying idea
Notice that context and guardrails are exactly what people tend to leave out. They’re so ingrained that we don’t realise they’re missing.
When we start to put this level of structure around each action, we begin to see what’s needed to define the work.
Not only that, but writing clear instructions forms the basis of AI prompts. So, if the tasks in question could be done by AI, we’ve actually started prompt engineering!
How cool is that?
Part 3: Which part of a workflow should we start with?
I can already hear the objections - how on earth can we do this for every action?
Well, we can’t!
But thankfully, not all parts of a workflow are created equal. Meaning, the effective execution of some parts within a workflow are more important than others.
For example, if our goal is to make a video, it’s probably more important to include the right content than to have perfect lighting. But writing the script and setting up equipment are both parts of our workflow. And if we look at them without that extra perspective, they could be perceived as having the same importance - it’s just a flat list of jobs to be done. But the outcome of the workflow is more dependent on great content than great video production.
So, when considering to what extent we design our work, we should triage parts of a workflow based on their importance.
A simple way to do this is to ask: if this part went badly, how much would it affect the overall outcome? The parts with the biggest impact deserve the full treatment. For the rest, less detail will do.
The rubric at a glance
Put this all together, and you have a simple rubric to apply to any workflow:
How far do I break it down? Can this workflow be done on my own in a few hours? If not, you’re probably looking at several workflows and need to pick just one and break that down further.
What does each action need? Have I provided context, instructions, guardrails and templates or blueprints?
Where is it worth the effort? How much does this part affect the outcome? Design the most important ones in depth.
Once you know what to look for and once you’ve practiced this, defining work gets a lot easier.
Over the next few posts, I’ll share what this looks like in practice:
how to design the workshop to help people practise, and
how to build tools to guide people through it.
And if you’re an L&D professional who wants to master workflow improvement and become the person your business turns to as AI changes work, join my email list today.
