Don’t Start With AI. Start With the Work.

If you brought me into a room and asked, "Where should we use AI?" I would probably answer your question with another one:

Where is the work breaking down?

Before I want to see the technology, I want to see the work. Show me how a request enters the organization. Show me who receives it, what happens next, where the information goes, who makes the decision, what gets copied manually, what waits in someone's inbox, and what happens when the person who normally handles the process is unavailable.

That conversation may not feel as exciting as watching an impressive AI demonstration. But increasingly, the evidence suggests that this is where the organizations getting meaningful value from AI are separating themselves from everyone else.

McKinsey's 2026 global AI research found that nearly three quarters of what it classified as AI high performers said they had fundamentally redesigned workflows because of AI, compared with about one quarter of other respondents. Those high performers represented only a small portion of the full survey, and the research leaned toward larger organizations, so I would not simply transfer the percentages to a church or a local nonprofit. But the organizational lesson is worth paying attention to: the organizations reporting the strongest value are not simply dropping AI into whatever process already exists. They are reconsidering how the work itself gets done.

That distinction is where I want to begin.

Start With Friction

Organizations become remarkably good at accommodating friction.

A monthly report takes four hours, so eventually everyone accepts that the report takes four hours. One person knows how to assemble the board packet, so everyone works around that dependency. Meeting notes are taken faithfully, yet decisions and action items still disappear between meetings. A communications team recreates similar material again and again because previous work is difficult to locate.

Eventually the workaround becomes the workflow.

And when that happens, one of the worst things we can do is assume that adding AI will automatically make the process better.

Sometimes it will.

Sometimes it will simply make a poorly designed process move faster.

This is why I would put the work on the table first.

The Chayil Workflow Lens

Notice that AI Opportunity does not come first.

That is intentional.

Not Every Inefficiency Is an AI Opportunity

This is one of the places where I think responsible AI advisory has to resist the excitement of the moment.

Sometimes the organization does not need artificial intelligence.

It needs a checklist.

Sometimes it needs a shared folder people can actually find.

Sometimes leadership has never clarified who owns the process.

Sometimes the real problem is that three people believe somebody else is responsible.

Sometimes information has been trapped in one employee's head for six years and what the organization really needs is documentation.

Sometimes a decision has been waiting for leadership, and no technology can make the decision leadership keeps postponing.

Artificial intelligence can be extraordinarily useful. It can also become a very sophisticated way to avoid addressing a basic organizational problem.

That is why I would rather ask, "Why is this work difficult?" before I ask, "How can AI make it easier?"

The order matters.

Put the Workflow on the Wall

Imagine we are looking at volunteer onboarding.

Someone completes a form. An administrator reviews it. Information may need to be entered somewhere else. A department leader needs to know the person is coming. Communication goes out. Training may need to be scheduled. Documents need to be distributed. Follow up happens, or sometimes does not happen.

We could immediately begin asking what AI might automate.

I would rather map the process.

What triggers onboarding? Who owns it? What information enters the process? Which steps are repetitive? Where do delays happen? What requires approval? What happens when information is incomplete? How do we know the person is actually ready to serve?

Only after we understand that do we ask where AI might appropriately assist.

Perhaps it helps organize incoming information. Maybe it prepares a routine communication for human review. It may summarize responses, retrieve the correct training information, or help prepare materials.

And there may be portions of the process where AI has no business at all.

That is a legitimate answer too.

Now imagine the same conversation around a nonprofit grant cycle. The team may spend hours gathering background information, locating previous language, reviewing requirements, coordinating contributions from several people, drafting, checking, revising, and tracking deadlines.

Again, AI may be useful in several places.

But if nobody owns the process, if source material is scattered, if staff do not know which version is current, or if nobody is responsible for verifying the final submission, AI has not solved the real problem.

It has entered the problem.

Map the Judgment Too

When teams map workflows, they usually map tasks.

Complete the form. Send the email. Enter the information. Prepare the document. Schedule the meeting.

I want organizations to map something else too.

Map the judgment.

Where does someone interpret context? Where does compassion matter? Where does organizational authority matter? Where does a leader weigh competing priorities? Where could an error affect a person, a relationship, a financial decision, a beneficiary, or the mission itself?

Those points matter because an AI enabled workflow should not quietly remove human judgment simply because a system is capable of completing a task.

Sometimes the most important point in the entire workflow is the one that says:

Human reviews here.

That is not inefficiency.

That is design.

Productivity Is Not the Same as Organizational Value

This distinction is becoming important in the research too.

Employees are reporting meaningful benefits from AI in their own work. Gallup's 2026 data found that 52 percent of U.S. employees reported using AI in their role, while 47 percent said their organization had integrated AI tools. One in five employees still did not know whether their organization had integrated AI at all.

That tells me something important.

Organizations can have plenty of individual AI activity without having a shared organizational approach.

And there is a larger gap appearing between people getting faster and organizations getting better. McKinsey's 2026 research describes many organizations still struggling to translate individual productivity improvements into broader financial and organizational impact. The stronger performers distinguish themselves not merely by using more AI, but by redesigning workflows, establishing leadership commitment, measuring impact, and building operating discipline around the technology.

That is the difference between adding a tool and building capability.

From Experiment to Repeatable Practice

One employee finding a clever AI shortcut is useful.

It is not yet an organizational workflow.

For that practice to become something the organization can depend on, somebody has to decide how it works, what tool is appropriate, what information may enter the process, who owns the result, what requires review, what standard the output must meet, and what happens when the technology does not behave as expected.

Another person should be able to understand the process.

The organization should not become dependent on the one employee who figured it out first.

This is where Chayil's strategic and operational lenses meet.

My work naturally presses into organizational readiness, leadership, people, strategic fit, behavior, and the questions underneath the process. Terrance K. Cosby Jr. brings the complementary operational lens of execution, standards, accountability, process discipline, training, and repeatability.

Together, those perspectives force two different but necessary questions.

Should we be doing this?

And:

Can this be executed well, repeatedly, and responsibly?

Strategy and execution have to meet somewhere.

The workflow is often where they meet.

Teach People to See the Work Differently

One of the outcomes I want from AI workflow work is not merely a redesigned process.

I want the team to learn how to see.

When work is highly repetitive, when information has to be reorganized over and over, when people repeatedly transform one type of material into another, when knowledge is hard to retrieve, or when highly capable people spend hours performing low judgment preparation work, there may be an opportunity worth exploring.

But the second set of questions matters just as much.

What information is involved? What could go wrong? What requires verification? Where does human judgment remain? Who owns the result? What expectations or policies apply?

That is how workflow thinking begins moving naturally toward governance.

And it is why Chayil can enter this work at the workflow itself. An organization may not need a massive AI transformation initiative. Sometimes we need to put one process on the wall, understand why it is difficult, and decide whether AI genuinely belongs there.

The goal is not to sprinkle AI across the organization.

The goal is to strengthen the work.

Before We Leave This Conversation

Suppose we find the right opportunity.

The workflow makes sense. AI has an appropriate role. Human judgment is preserved. Ownership is clear. The team understands the new process, and we have decided how we will know whether it is actually working.

Another set of questions now becomes unavoidable.

Who decided which AI tool was approved?

What information are people permitted to enter?

What happens when someone finds a new tool that leadership has never evaluated?

Who determines when human review is required?

What happens when the output is inaccurate, inappropriate, confidential, or inconsistent with the organization's responsibilities?

At some point, individual judgment is not enough.

The organization needs shared expectations and a structure for making those decisions.

And writing a document called an AI policy is only part of that answer.

That is where I want to pick up our conversation next.