Before You Automate with AI, Fix the Process: A Practical Framework for AI-Ready Business Workflows
AI speeds up whatever it is pointed at, including a broken workflow. Here are six steps for deciding where it adds real value, and where rules, integration, or human review do the job better.
There is a question I keep hearing more and more in conversations about digitalization:
"Can we use AI to automate this?"
It sounds like the right question. Sometimes it is.
But I think there is a more important question that should come first:
"Do we actually understand the process we are trying to automate?"
Because AI can accelerate a good process. It can also accelerate a bad one.
If the underlying process is broken, unclear, poorly governed, or dependent on unreliable data, adding AI does not magically fix it. It simply allows the organization to do the wrong thing faster, at greater scale.
That is why I increasingly think about AI readiness as a process problem before it becomes a technology problem.
AI Should Not Be the Starting Point
Automation should not begin with the tool. It should begin with the problem.
A team might spend hours every week moving information between systems, checking documents, classifying requests, preparing reports, or reconciling data. The natural reaction today is to put an AI agent on it.
Maybe. But first I want to understand what actually happens today, where the process slows down, and where errors occur. Most of all, I want to know which parts of the process actually need intelligence, and which simply need better engineering.
That distinction matters more than it might initially appear. Sometimes the right solution is generative AI. Sometimes it is a Python script, an API integration, a better database, or an automated workflow inside an existing business application. And sometimes the best solution is to remove a process step altogether.
Not every business problem is an AI problem.
The Process → Data → AI Framework
When I think about whether a business workflow is ready for AI, I find it useful to work through a simple sequence:
Process → Data → Decision → AI → Governance → Measurement
1. Process: Map what actually happens
Before automating anything, map the process. Not the process as it appears in a PowerPoint presentation. The real process: who starts it, which systems are touched, where people manually copy information, and where it depends on someone's personal knowledge rather than a documented rule.
This is often where the first opportunity appears. A workflow may look like an ideal candidate for AI when, in reality, its biggest problem is simply that three systems do not communicate with each other. Better integration, not AI, might be the solution.
2. Data: Check what the process runs on
AI systems are only as useful as the information and context available to them. Data does not need to be perfect, but you need to understand where it comes from, who owns it, how reliable it is, and whether it contains confidential or personal information that limits how it can be used.
A technically impressive AI workflow built on poorly understood data is still a poorly designed workflow. Businesses should spend at least as much time thinking about the data flowing into the models as about the models themselves.
3. Decision: Find the actual bottleneck
Not every step in a process requires intelligence. If a field has a specific format, validate it with a rule. If information comes from a known system, retrieve it through an API. If a calculation follows a defined formula, use software.
AI adds real value where a step involves interpreting an unstructured document or classifying information that does not follow a fixed pattern. So the goal is not to replace as much of the process as possible with AI. It is to find where AI provides a meaningful advantage over rules, traditional automation, or human effort.
4. AI: Use the right capability for the right task
Summarizing information, extracting data from documents, classifying content, drafting, and answering questions against a controlled knowledge base are very different use cases. They may require different models, architectures, retrieval mechanisms, evaluation strategies, and levels of human oversight.
Services such as those available through AWS allow organizations to build AI capabilities without treating every AI initiative as a research project. But because we can build something does not mean we should build it. The business case still comes first.
5. Governance: Plan for when the AI is wrong
People often ask how accurate the AI is. It is an important question, but it is incomplete. What matters just as much is what happens when it is wrong.
A mistaken AI-generated summary might be inconvenient. A mistaken financial classification could be much more serious. The acceptable level of automation depends on the consequences of failure, and this is where human-in-the-loop design becomes important:
Data → AI → Evaluation → Human Review → Action → Monitoring
AI handles the parts where it provides leverage. The human remains accountable for the decisions where context, judgment, or responsibility matters. That is not a failure of automation. It is good system design.
6. Measurement: Check that something actually improved
An AI project can look impressive in a demonstration and still provide very little business value. So after implementation, measure the outcome: processing time, error rates, time spent on repetitive work, cost, and ease of compliance.
The goal is not to deploy AI. The goal is to improve an outcome.
What Makes a Business Process AI-Ready?
Based on the way I have started thinking about these problems, I would describe an AI-ready workflow through four dimensions.
- Process readiness. The organization understands the workflow, its bottlenecks, its exceptions, and its desired outcome.
- Data readiness. The required information is accessible, sufficiently reliable, appropriately governed, and usable for the intended purpose.
- Technical readiness. The existing systems can support the required integrations, automation, security, and monitoring.
- Organizational readiness. The people using the system understand what it does, what its limitations are, and where human responsibility remains.
If one of these is missing, introducing AI becomes considerably harder. And sometimes the right answer is not to implement AI yet. It is to make the process ready first.
From Automation to Augmentation
Traditional automation had a straightforward objective:
Find a repetitive task → automate it → save time.
AI makes that equation more interesting. It moves automation beyond replacing manual steps and toward augmenting human capability: supporting decisions, making knowledge more accessible, and freeing people's attention for the work that needs it.
And that brings me back to something I wrote about in my previous article on vibe coding. AI is not making understanding less important. It is making understanding more valuable.
When the cost of building a prototype falls dramatically, the scarce resource becomes knowing what is worth building, what constraints matter, how to evaluate the result, and where the system should not be trusted.
The Bigger Lesson
Process, data, automation, and AI are not separate disciplines. They are layers of the same problem:
- A business process creates data.
- Data enables decisions.
- Technology supports those decisions.
- Automation reduces unnecessary effort.
- AI can extend what the system is capable of doing.
- And people remain responsible for deciding what the system should ultimately achieve.
That is why I don't think the future of AI belongs simply to the people who know how to use the latest model.
I think it belongs to people who can look at a messy business problem, understand the process underneath it, understand the data behind it, choose the right technology, and still ask the most important question:
"What are we actually trying to improve?"
Before we automate with AI, we should fix the process. Because the smartest AI system in the world cannot compensate for a problem nobody has properly understood.
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article