Should You Automate the Workflow or Fix the Process First?
A slow process often looks like an automation opportunity.
Employees are copying information between systems. Approvals take too long. Client requests sit in inboxes. Reports require hours of manual preparation. Someone inevitably asks whether AI could make the process faster.
Sometimes it can. But speed is not always the problem.
The workflow may have too many steps. Responsibilities may be unclear. Employees may be working from inconsistent information. An approval may exist because of a policy nobody has revisited in years. Different departments may have developed their own versions of the same process.
This is a process problem.
You can’t fix a broken process with automation.
McKinsey's 2026 research on AI operating models makes an important distinction. Many organizations are using AI to accelerate existing activities while leaving the underlying workflows, governance, teams, and operating models largely unchanged.
For business leaders, the question is not just:
"What can we automate?"
A better question is:
"What is preventing this workflow from producing the result we need?"
Start With the Friction
Before changing a workflow, identify where the friction actually occurs.
Consider a professional services firm where onboarding a new client takes several days. At first glance, the solution may seem obvious: automate onboarding.
But onboarding is not one task.
A signed agreement arrives by email. Someone creates the client record. Another employee requests missing information. Finance sets up billing. A project manager assigns the work. Documents are saved in the appropriate location. Internal teams are notified. The client receives instructions about what happens next.
If onboarding is slow, the delay could come from any number of places:
Required information is frequently missing.
Employees don’t know who owns the next step.
The same information is entered into three different systems.
Every engagement requires manager approval even though only a small percentage involve unusual circumstances.
Until the organization understands the source of the friction, automating the process is premature.
The first objective is not automation. It is a diagnosis.
When the Process Should Be Fixed First
Some workflow problems are fundamentally process problems.
AI may eventually play a role, but technology should not be the first intervention.
There are several warning signs.
Ownership Is Unclear
If employees regularly ask, "Who is supposed to handle this?" automation will not resolve the underlying ambiguity.
A workflow needs clear responsibility:
Who starts the process?
Who makes decisions?
Who handles exceptions?
Who is accountable when something does not happen?
Automating handoffs before answering those questions will make accountability difficult. Lack of clear ownership creates confusion and slows the process.
The Steps Are Inconsistent
If five employees perform the same process five different ways, the organization needs to determine whether that variation is necessary.
Sometimes employees legitimately need flexibility. Other times, different approaches have developed simply because the process was never standardized.
Before automating, management should determine which steps are required, which can vary, and what a successful outcome should look like.
AI works better when the organization can clearly define what it is trying to accomplish.
The Process Contains Unnecessary Work
Organizations accumulate steps as they grow. With growth comes complexity.
A report was created for a manager who no longer works there. An approval was added after a problem several years ago. Employees copy someone on an email because that has always been the practice. Information is entered into a spreadsheet even though it already exists in another system.
Automation should not preserve work that no longer needs to happen.
Before asking how to automate a step, ask whether the step should exist at all.
The Inputs Are Unreliable
A workflow cannot consistently produce good results when the information entering it is incomplete, inaccurate, or stored unpredictably.
If employees routinely need to track down missing client information, reconcile conflicting spreadsheets, or determine which document is current, automating downstream tasks may only move the problem.
Data and information issues need to be addressed before you can reliably automate a workflow.
When Automation Makes Sense
Automation is a strong fit when the workflow already works reasonably well but consumes too much time.
The steps are understood. Ownership is clear. The inputs are dependable. The outcome is predictable.
The problem is that employees spend too much time performing repetitive work within the process.
That is where automation can create meaningful capacity.
For example, an employee may receive standardized forms by email, extract information, enter it into another system, create a folder, notify a colleague, and update a tracking record.
If the process is consistent and the exceptions are understood, much of that coordination may be a strong candidate for automation.
The same can apply to:
Routing incoming requests
Preparing recurring reports
Extracting information from documents
Updating records across systems
Creating standard notifications
Categorizing routine inquiries
Checking whether required information is present
Preparing information for human review
The key distinction is that automation removes manual effort from a process that already makes sense.
It is not being asked to compensate for a broken process.
Sometimes the Right Answer Is to Redesign and Automate Together
The most valuable opportunity may not be to automate the existing process. It may be to redesign how the work happens.
This is where workflow redesign becomes valuable.
McKinsey's research argues that the real advantage from AI comes from redesigning how work is performed and how decisions are made, rather than simply adding technology to the existing operating model.
Consider a weekly management report. The existing process might require employees from several departments to update spreadsheets, send information to a manager, reconcile inconsistencies, prepare charts, write a summary, and distribute the finished report.
A better design might connect the underlying systems, standardize the definitions, flag exceptions automatically, and generate the report.
If the required information can be pulled from existing systems, monitored continuously, summarized when needed, and presented with exceptions highlighted for management review, the organization may be able to redesign the entire workflow.
The question is no longer:
"How can AI prepare this report faster?"
It becomes:
"How should management receive this information if we were designing the process today?"
That is a much more valuable question.
Evaluate the Entire Workflow, Not One Task
One of the easiest automation mistakes is optimizing a single step without considering what happens before and after it.
Imagine AI reduces the time required to review an incoming document from thirty minutes to five. This could be a significant improvement.
But suppose the output still needs to be manually checked, reformatted, entered into another system, emailed to a manager, and approved before anyone can act on it.
The organization has improved one task.
It may not have improved the workflow.
In some cases, making one step dramatically faster can expose a new bottleneck somewhere else.
This is why workflow improvement should be measured from beginning to end.
Consider the measurable improvement:
Did the total time required to complete the process decrease?
Did employees perform fewer manual steps?
Were handoffs reduced?
Did error or rework rates improve?
Did the process become more consistent?
Was meaningful capacity created?
Did the experience improve for employees or clients?
Improving a task doesn’t speed up the process. Improving the workflow will.
Decide Where Human Judgment Still Matters
Automation does not have to mean removing people from the workflow.
Many of the strongest processes deliberately separate routine work from work requiring judgment.
AI may collect information, prepare a summary, identify missing details, categorize a request, or recommend a next step.
A person may still approve an exception, communicate with a client, make a financial decision, or evaluate a situation where context matters.
The Microsoft 2026 Work Trend Index reinforces this distinction. Its research suggests that effective AI adoption requires leaders to rethink how work is structured and deliberately determine where AI handles execution and where people provide direction, judgment, quality standards, and accountability.
For management teams, this means establishing clear boundaries before implementation.
Define where the system can act automatically.
Identify where the system should prepare information for a person.
Establish which decisions require human approval.
Determine what happens when confidence is low or a situation falls outside normal parameters.
A well-designed workflow makes those boundaries clear.
When To Fix, Automate, or Redesign
When evaluating a workflow, management can begin with three options.
Fix first when responsibilities are unclear, steps vary significantly, the process contains unnecessary work, inputs are unreliable, or the desired outcome has not been defined.
Automate when the process works reasonably well, the work is repetitive and predictable, ownership is clear, exceptions can be identified, and manual effort consumes meaningful capacity.
Redesign and automate when the current process was built around limitations that no longer apply, several steps can be combined or eliminated, information can move differently, or improving individual tasks would leave the larger bottleneck untouched.
This does not need to become a months-long process redesign exercise.
The priority is designing a better way for the work to happen.
The purpose is simply to avoid automating the wrong thing.
Measure Whether the Workflow Actually Improved
After implementation, return to the original business problem. If onboarding took four days before the change, how long does it take now? If employees spent 20 hours each week preparing reports, how much manual effort remains? If client requests were frequently delayed, are response times improving?
This is where capacity, speed, and consistency become useful measures.
Capacity: Can the organization handle more work with the same resources?
Speed: Does work move from beginning to end more quickly?
Consistency: Is the process producing more reliable outcomes across employees, departments, and clients?
A successful automation initiative should improve the operation in a way management can observe.
If the only measurable change is that another AI tool has been added, the organization has more work to do.
Improve the Workflow Before You Accelerate It
AI makes automation easier to imagine.
That does not mean every manual process should be automated in its current form.
Sometimes the right decision is to automate.
Sometimes it is to fix ownership, simplify the process, improve the data, or remove unnecessary steps first.
And sometimes AI creates an opportunity to redesign the workflow entirely.
The most important step is understanding which situation you are dealing with.
At AI Growth Advisors, we help leadership teams evaluate how work actually moves through their organizations, identify the friction slowing growth, and determine where process improvement, automation, or AI can create meaningful business value.
Sources:
