TL;DR
- Most AI projects start too late, with the tool instead of the workflow.
- The SOP is rarely the same as the real process.
- Map the work, remove waste, then automate.
- Sometimes the better question is whether the workflow should exist at all.
- AI deployment only works if people actually adopt the new way of working.
1. Most AI Projects Start in the Wrong Place
Someone sees a demo.
Someone hears that a competitor is using AI.
A tool gets bought.
Then the team tries to automate work nobody can clearly explain.
Before asking:
Where can we use AI?
Ask:
How does this work actually get done?
Who touches it? Which systems are involved? Where are the delays? What gets copied between tools? What happens when the normal path breaks?
If you cannot answer those questions, you are not ready to automate the workflow.
2. The SOP Is Usually Not the Workflow
Most established businesses already have documentation somewhere.
But processes drift.
People build workarounds. Exceptions become normal. Old approval steps survive. Spreadsheets quietly become part of operations.
Eventually there are two processes:
The one management thinks exists.
And the one people actually use.
AI needs the second one.
The best way to find it is often simple:
“Show us the last time you actually did this.”
Not the policy. The real invoice, customer record, support case, or order.
That is where the useful details appear.
3. Don’t Automate a Bad Workflow
Once the workflow is visible, do not immediately add AI.
First ask what should disappear.
You may find:
- duplicate data entry
- redundant approvals
- reports nobody reads
- unnecessary handoffs
- meetings created because systems do not surface the right information
A better sequence is:
Eliminate → Simplify → Standardize → Automate → Apply AI
Sometimes the best AI decision is not using AI at all.
4. The Bigger Question: Should This Workflow Exist?
Optimization asks:
How can AI make this faster?
The more important question is:
If we designed this process today, with AI available from the start, would we build it this way at all?
Take a weekly report.
If AI can continuously monitor the data and surface only what requires action, maybe you do not need a faster report.
Maybe you do not need the report.
That is the difference between adding AI to an old process and redesigning the process around what is now possible.
5. Run Quick Wins and Transformation in Parallel
You do not need months of process mapping before getting value from AI.
Run two tracks.
Use AI now for horizontal work like email triage, meeting notes, document search, knowledge retrieval, and routine research.
At the same time, study the workflows that actually make your business different.
Those deeper workflows often hold the bigger ROI.
Quick wins build confidence.
Workflow redesign creates leverage.
6. Deployment Is Not Adoption
You can build a technically perfect AI system and still fail.
People may not use it.
Or they try it, hit one annoying exception, and go back to the spreadsheet.
That means change management starts during discovery, not after launch.
Bring the people doing the work into the redesign.
And if AI turns a four-hour task into 30 minutes, ask the next question:
What should that person do with the other three and a half hours?
That is where the business value starts.
Read the Full Article
The full article goes deeper into workflow discovery, task mining, process mining, AI-native redesign, governance, adoption, and how to decide where AI actually belongs.
Final Thought
Do not start with:
Where can we install AI?
Start with:
How does this business actually create value?
Then figure out what to eliminate, simplify, automate, or redesign.
Business outcome first. Workflow second. Technology third.
That is signal.
Everything else is tooling.
Thanks for reading Signal Over Noise,
where we separate real business signal from AI noise.
where we separate real business signal from AI noise.
See you next Tuesday,
Avi Kumar
Founder: Kuware.com
Subscribe Link: https://kuware.com/newsletter/