Use this whenYou’ve picked a workflow and need to decide how it should change before you automate it.
Why it exists
Most AI projects start with a tool and go looking for work to apply it to. That usually automates steps that shouldn’t exist, on top of information nobody trusts. Being able to automate a task isn’t a reason to keep doing it.
The framework
- Step 1DataWhat information does the job need, and can people and agents use it?
- Step 2ProcessWhich steps still need to exist?
- Step 3PeopleWhose responsibilities change, and can they make that change?
Data
What information does the job need, and can people and agents use it?
Data comes first. Look at what comes in, what has to come out, and where it goes next. If the information isn’t clean, connected, and owned by someone, an agent just moves the confusion faster.
Process
Which steps still need to exist?
Follow one piece of work from start to finish. Note where it waits, repeats, or loses information. Some steps can go away entirely. Some belong with an outside provider.
People
Whose responsibilities change, and can they make that change?
Once the information and steps are clear, decide who owns what. Redesign stalls when the fix depends on a group that wasn’t in the room. Sometimes responsibilities have to change before better software will help.
How to use it
- Pick one real workflow and map it with the people who do it.
- Start with the data: trace what comes in, what goes out, and where it goes next. Fix ownership and access before anything else.
- Run each step through the process lens. Mark it keep, remove, move, or redesign.
- Decide whose responsibilities change, and bring those people in.
- Only then decide where AI or automation fits. Choose the smallest experiment, and decide up front how you’ll know it worked.
Mark every step
- Keep: the step earns its place and stays as it is.
- Remove: nobody would miss it. Being able to automate a step is no reason to keep it.
- Move: it belongs with a different team, system, or outside provider.
- Redesign: the step matters, but how it is done has to change.
Signs the workflow is not ready for AI
- Nobody owns the data the job depends on.
- The people who do the work are not in the room.
- The process changes every time someone runs it.
- There is no way to tell whether the change worked.
- If any of these is true, fix it first. Automating it now only makes the problem faster.
In practice
When we mapped work across capital programs and corporate real estate, making the underlying data readable by AI and accessible to the people building tools became a major engineering priority. Some processes could simply go away. Others needed a change in who owned what before any tool would help. What remained was the work we built agents for.
Where it breaks down
The lenses don’t tell you what to build. They help you avoid building the wrong thing. They also need the people who do the work in the room. Mapping from a conference room produces a tidy diagram of a process nobody follows.
Try it with AI
I want to redesign one workflow before adding AI to it. Work through three lenses, in this order: 1. Data: What information does the job need, and can people and agents use it? 2. Process: Which steps still need to exist? 3. People: Whose responsibilities change, and can they make that change? Ask me one question at a time to map how information moves through the workflow. Then mark each step keep, remove, move, or redesign, and suggest the smallest experiment worth running. Don't recommend software until the map is done.