Why We Interview Employees Before Recommending AI

When a business asks where AI could create value, it is tempting to begin with the owner, CEO, or leadership team.

We do begin there.

We just do not stop there.

At The Scale Crew, one of the core principles behind our Business AI Readiness Assessment approach is that we interview people at different levels of the organization before recommending where AI should be implemented.

There is a simple reason:

Employees are not just the eventual users of a new system. They are a source of evidence about how the business actually works.

If we want to understand where AI belongs, we need to understand the work first.

That starts with the people performing that work every day, because they often know things the process documentation, executive dashboard, and leadership meeting cannot tell us.

The Org Chart Is Not a Process Map

A business may have clearly defined departments and responsibilities.

Sales sells.
Operations delivers.
Finance bills.
Customer service supports.
Marketing generates demand.That only tells us who owns different functions.It does not tell us exactly how work moves through them.
Processes cross department lines.
Information moves from one person to another.
Customers create exceptions.
Software systems fail to communicate.
Employees develop shortcuts.
Managers create approvals to solve old problems.
New tools get added.
Old tools remain.
Someone builds a spreadsheet.
Someone else creates a template.
An employee becomes the unofficial person everyone asks when a particular situation occurs.

Over time, the process on paper and the process in reality begin to separate.

If you are evaluating AI from the process on paper, you may be evaluating a business case that does not actually exist.

Employee interviews help expose the real one.

Employees Know Where the Friction Lives

Ask an executive which processes are inefficient, and you may get a useful answer.

Ask an employee to walk you through exactly what happens from the moment a task begins until it is finished, and you may discover twenty smaller problems inside it.

They may tell you:

The customer information arrives in the wrong format.
A field has to be copied manually because two systems do not integrate.
A report has to be rebuilt every Friday because the software cannot produce it correctly.
A supervisor has to approve routine exceptions because of a rule created three years ago.
Employees search old emails because the official documentation is outdated.
Three people maintain separate spreadsheets containing slightly different versions of the same information.
Customers repeatedly ask questions because an upstream communication is unclear.
The CRM technically contains the information everyone needs, but nobody trusts a particular field because it is rarely updated.

This is not just noise around the process or complaining.

This is the actual process, and as the leader you never really saw it.

So if we do not understand that environment, we cannot responsibly recommend how AI should interact with it.

Current AI Research Supports Looking Beyond Deployment

The workplace research around AI increasingly points in the same direction.

Gallup reported in 2026 that only 30% of U.S. employees were using AI at work at least weekly. Its research found manager support was strongly associated with frequent AI use and employees perceiving AI as valuable.

In Gallup’s Q1 2026 U.S. workforce data, among organizations that made AI available to employees, employees who strongly agreed that their organization’s AI technology integrated well with their existing systems and processes reported substantially higher frequent AI use than those who did not strongly agree. Manager support showed a similarly large relationship.

Microsoft’s 2026 Work Trend Index reached a related conclusion from a different dataset. Microsoft found that organizational factors such as culture, manager support, and talent practices accounted for twice the reported AI impact of individual effort alone among the AI users it studied.

Technology availability matters.

But technology does not operate outside the organization around it.

Workflow fit matters.
Management matters.
Culture matters.
Rules matter.
Training matters.
Information matters.
The people doing the work matter.

We Are Not Interviewing Employees to Ask Them Which AI Tool to Buy

This is an important distinction to be made here.

An employee interview is not a technology shopping session.

We are more interested in what happens before AI enters the conversation.

We want to understand the work.
Walk us through what happens.
Where does it begin?
What information do you need?
Where does that information come from?
What do you have to check?
What do you copy?
What do you create manually?
What do you repeatedly search for?
Who do you depend on?
Where do you wait?
Where do customers get confused?
Where do mistakes happen?
Which parts require judgment?
What happens when the normal process does not work?
What have you created yourself to make the process easier?

These questions will reveal something a generic AI survey rarely will:

Context.

Someone saying, “This process is inefficient,” is useful.

Someone walking us through the fourteen steps required to complete it tells us something we can investigate.

Why Interviews Instead of Just Surveys?

Surveys are useful.

They are especially useful when we want consistent information from a larger number of people.

But surveys have limitations during discovery.

We can only ask questions we already know to ask.

Interviews allow us to follow the thread.

An employee says:

“Then I check the spreadsheet.”

Which spreadsheet?
Why is that information not in the primary system?
Who updates it?
How often?
What happens when it is wrong?
Who else uses it?
Why was it created?
Is this spreadsheet part of the documented process?

A five-minute tangent about one spreadsheet can reveal that an entire department has built a parallel information system around software leadership believes is functioning correctly.

That discovery can completely change a technology recommendation.

AI discovery requires curiosity because the important problem is frequently one question deeper than the first answer.

Contradictions Are Part of the Assessment

There is another reason we interview multiple levels of the business.

We want to compare perspectives.

Suppose leadership tells us:

“Our employees spend too much time responding to customers.”

Employees tell us:

“Responding is easy. Finding the correct information to respond with is what takes forever.”

Those are different problems.

Leadership may be seeing a response-time problem.

Employees may be experiencing a knowledge-access problem.

The eventual solution might still use AI, but its purpose changes.

Instead of automatically answering customers, perhaps the better first use case is helping employees retrieve and synthesize information from approved sources.

Or perhaps AI is not necessary at all because the underlying information architecture needs to be fixed first.

Neither conclusion can be reached confidently from the original statement alone.

The contradiction creates the investigation.

Employee Interviews Can Reveal What Should Not Be Automated

This is just as important as finding automation opportunities.

Some work looks repetitive from a distance but contains important human judgment.

A task might take fifteen minutes, but during those fifteen minutes an experienced employee notices unusual customer behavior, recognizes an exception, evaluates risk, or uses contextual knowledge that has never been formally documented.

If we reduce the process to:

“Employee receives information and enters result”

We may incorrectly assume everything between those two points is mechanical.

An interview allows us to ask:

What are you thinking about while you do this?
What makes one case different from another?
When do you stop and ask someone for help?
What would make you uncomfortable allowing a system to decide automatically?
What mistakes would be expensive?
What information affects your judgment but is not documented?

The answers to these types of questions will help determine where AI can act independently, where it can assist a person, where human review should remain, and where automation may not be appropriate.

NIST’s AI Risk Management Framework similarly emphasizes engagement with end users and relevant stakeholders, human-centered design, and understanding context when evaluating AI systems and their potential impacts.

Employees Also Reveal Adoption Risk

A system can perform beautifully in a demonstration and fail inside the company.

Why?

Because implementation is not simply technical.

Employees may not understand why the system exists.
They may believe it threatens their role.
They may not trust its output.
They may know it fails on exceptions the implementation team never discovered.
They may already have a faster workaround.
They may have received no guidance about what information can safely be entered into the system.
Their manager may not use the technology.
Their performance metrics may reward the old workflow.
The AI tool may require employees to enter additional information before it saves them any time.

Microsoft’s 2026 Work Trend Index found that 45% of AI users surveyed said it felt safer to focus on their current goals than to redesign work with AI, while only 13% said they were rewarded for reinventing work with AI even when results were not met.

That is an organizational problem, not a model-performance problem.

An assessment that ignores employees can miss it entirely.

Employee Input Does Not Mean Employee Rule

There is an important boundary here.

Interviewing employees does not mean every complaint becomes a technology project.
It does not mean the majority opinion determines the AI strategy.
It does not mean leadership gives up strategic authority.
And it does not mean employees always understand the economic or technical implications of the changes they want.Different levels of the business possess different information.
Leadership has context employees may not have.
Employees have operational context leadership may not have.
Managers see patterns across teams.
Technology and data reveal constraints people may not know exist.
Financial analysis determines whether solving the problem creates enough value to justify the investment.

The point of an assessment is to bring those perspectives together.

We are gathering evidence, not votes.

Sometimes the Employee Interview Changes the Entire Recommendation

A company may enter an assessment expecting custom AI development.

The evidence may show that its existing software can already solve the problem with better configuration.

Another may think it needs automation.

Employee interviews reveal the process should be simplified first.

Another may believe its biggest opportunity is customer service.

The evidence shows employees across operations are losing hundreds of hours to an internal knowledge problem.

Another may think its data is ready.

Employees explain that the official system is only half the story because critical information lives in email and shared spreadsheets.

Another may assume employees will resist AI.

Interviews reveal that employees are already using AI independently and are frustrated that the company has provided no standards, tools, direction, or training.

The purpose of discovery is not to confirm what we believed when we walked in.

It is to improve the recommendation.

Sometimes that means AI.
Sometimes automation.
Sometimes integration.
Sometimes custom software.
Sometimes process repair.
Sometimes doing nothing until another problem has been solved first.

A good assessment must be allowed to reach any of those conclusions.

The People Closest to the Work Hold Part of the Evidence

AI creates a strange temptation.

Because the technology is sophisticated, businesses can assume that evaluating it must begin with sophisticated technical questions.

Which model?
Which platform?
Which vendor?
Which agent framework?
Cloud or on-premises?
Custom or off-the-shelf?

These questions matter later.

The first questions are usually much more human.

What are people trying to accomplish?
How does the work actually happen?
Where does it break down?
What information do people need?
What decisions are they making?
Where is time being lost?
What should become easier?

That is why we interview employees before recommending AI.

Leadership tells us where the business is trying to go.
Managers help explain how work is coordinated.
Employees help show us what actually happens when the work gets done.
The systems, data, and economics give us additional evidence.

Together, those perspectives allow us to diagnose before we recommend.

So when a business is deciding where to invest in AI, that distinction can be the difference between buying an impressive technology and honestly solving a problem that actually matters.