Opens in a new tab
vmblog logo 2024 wht (updated)

2027 Predictions: AI Can’t Fix a Process You Don’t Understand

Share: 

Frank Vega | Published: November 17, 2026
vmblog predictions 2027

By Frank Vega, Founder and CEO, The Efficiency Group, developer AI Process Mapper

We’ve reached a point where organizations are asking AI to do just about everything. Write a report. Analyze a document. Generate code. In the business process world, organizations are asking AI to build a workflow and then automate it.

The technology is impressive, and the pressure to use it is enormous. But there are several problems with this approach that aren’t getting enough attention, especially when it comes to business processes.

First, not every business process belongs in ChatGPT, Claude or Copilot. And just because AI can automate something doesn’t mean that process should be automated. Second, AI users don’t understand to use the technology or what it gives you. The prompts matter; how they’re drafted, how they’re interpreted, how they’re presented. Consistent outputs is not a hallmark of today’s AI. Third, the technology is moving so fast that what you know on Monday is out of date by Wednesday.

I believe that in 2027, we will start to see organizations recognizing that having a license to ChatGPT, Claude, or Copilot isn’t enough. They will see that to truly use AI effectively; they need capabilities that make AI easy to use in the first place to address their business challenges across the lifecycle. In the realm of business processes, they will recognize that process intelligence is a critical foundation for AI adoption and overall business operations. Ultimately, before asking what AI can do, they’ll need to understand what their organization is currently doing, why they’re doing it that way, and what happens when something goes wrong.

You Can’t Automate What You Don’t Understand

Most organizations have two versions of how work gets done.

There’s the official version, documented in policies, procedures and organizational charts. Then there’s the real version, shaped by experience, institutional knowledge, informal workarounds and the people who know how to keep things moving.

Those two versions aren’t always the same.

An employee may know that a particular approval requires an extra phone call. A nurse may understand that a handoff between departments needs additional verification. A government worker may know which exception requires escalation, even though it isn’t clearly explained – or even documented – in the procedure manual.

These details matter. Yet they’re often invisible to the people making technology decisions.

When organizations feed existing guidance into an AI tool and ask it to document the process, they risk automating an incomplete, inaccurate, out-of-date, or just plain wrong picture of reality.

AI can produce a convincing workflow diagram in seconds. That doesn’t mean the workflow is accurate, complete, compliant, applicable to the specific business or even safe.

Next year we can expect more organizations to use AI-assisted process mapping to establish a reliable picture of how work happens

in their organizations before introducing automation. That means examining tasks, decisions, handoffs, exceptions and responsibilities, then validating the results with the people who perform the work. We can expect this work to be “democratized” across an organization to be more than just a technology function but an operational function; people doing their work will need capabilities that help them understand their work quickly and accurately so they can work better and support AI transformation.

It’s not the most glamorous part of AI transformation but it is the most important.

Government Will Face an Institutional Knowledge Problem

The federal government’s workforce changes have created an urgent test of this principle.

Following the workforce reductions and reorganizations associated with the Department of Government Efficiency (DOGE) and broader federal initiatives, many agencies are operating with fewer employees and significant changes in responsibilities.

The immediate temptation may be to use AI to compensate for reduced staffing. But what happens when the people who understood how a process worked are no longer there?

Government operations involve complicated approval structures, regulatory requirements, public accountability and exceptions that aren’t always captured in written procedures.

If an agency automates a process without understanding those dependencies, it could accelerate errors, create compliance problems or delay services that citizens depend on. AI can help agencies reconstruct workflows from existing documentation, identify gaps and preserve operational knowledge. But those process maps must be checked against actual requirements and reviewed by experienced personnel. The GSA’s Federal EOA Playbook published this past June, specifically calls out process mapping as a key step in “solutioning” Federal process improvement. But, even with generic AI support, process mapping is labor intensive, costly, manual work.

My prediction is that government agencies will fully recognize process mapping as an essential step in rebuilding operational capacity, not simply an exercise in documentation. They will further recognize the need to use appropriate AI tools to make process mapping, analysis, and documentation itself more efficient, affordable, reliable, and accessible.

The goal shouldn’t be to make fewer people do the same broken work faster. It should be to redesign the work so it can be performed reliably with the resources available.

In Healthcare, Efficiency Cannot Come at the Expense of Safety

Healthcare presents an even more consequential example.

Consider what happens when a patient moves from an emergency department to an inpatient unit. Information must be documented, medications reconciled, orders reviewed and responsibilities transferred.

There are numerous opportunities for delays, duplication and mistakes. AI may help streamline these activities, but only if the underlying process is understood.

Automating a flawed handoff doesn’t eliminate the flaw. It can make the problem harder to detect because the process now appears to be operating efficiently.

That’s why healthcare organizations need to map workflows before deciding where AI belongs. They must identify which activities can be automated, where clinical judgment is required and what safeguards are necessary when exceptions occur.

This is where traditional process improvement disciplines become even more relevant. Lean Six Sigma teaches us to examine variation, eliminate waste and measure outcomes. AI can accelerate that analysis, but it cannot replace the judgment needed to determine whether a process is safe and effective.

In healthcare, success isn’t simply measured by how many minutes a workflow saves. It’s measured by whether patients receive the right care, at the right time, without preventable errors.

The Next Phase of AI Will Begin Before Automation

I believe 2027 will bring a much-needed adjustment in how organizations approach AI.

Instead of beginning with the technology, leaders will begin with the process.

They’ll ask where work breaks down, which decisions require human oversight, what institutional knowledge is missing and whether the current workflow should exist in its present form. AI-assisted process mapping will make this work faster and more accessible, allowing organizations to document, analyze and improve operations without relying exclusively on lengthy manual studies. AI-assisted process mapping will also support the full process improvement lifecycle seamlessly and easily making it invaluable for organizations to learn, understand, and adapt their business functions to ever-evolving realities.

But the discipline behind process improvement will remain essential. People still need to validate the findings, challenge assumptions and measure whether changes produce better outcomes. AI-assisted process mapping will significantly help people in taking these actions but won’t replace them

AI is an extraordinary tool. It can help us see problems faster and solve them more effectively.

But it cannot tell us whether we’re solving the right problem unless we’ve done the work to understand it. So next year, seek first to document, then understand, and then identify which processes are worth automating in the first place.

##

ABOUT THE AUTHOR

frank vega

Frank Vega is the founder and CEO of The Efficiency Group, a process-improvement firm using Gen AI to create BPMN models in minutes. Under his leadership, TEG has supported organizations across healthcare, government, supply chain, finance, and manufacturing, including initiatives with the National Institutes of Health, the Department of Energy, the Department of Defense, and General Dynamics IT. Frank’s work centers on making Lean Six Sigma and continuous improvement accessible to non-experts, enabling faster kaizen cycles, cleaner documentation, and measurable gains in throughput, quality, and cost. He frequently speaks on the practical use of AI in process mapping and on building a culture of improvement that scales beyond specialist teams.