← All writing

Business Process Automation: Prove the Workflow First

Before business process automation, define good output, handoffs, exceptions, and review. Build a workflow that works before you make it faster.

Jordan Glickman·November 21, 2026·5 min read
Operations

Before automating a business process, make sure you can explain what goes in, what should come out, and how you know the work is right.

You do not need a perfect manual process. You need enough clarity to avoid turning confusion into a faster, larger problem.

I have made this point across different kinds of work, from customer support to content and outbound sales. If the team cannot agree on a good result, a tool cannot settle that question for them.

Automation should carry a clear process. It should not hide the absence of one.

What Does It Mean to Prove a Workflow Before Automation?

It means running enough real examples to understand the work and its main failure points.

Take a content process. Someone finds an idea, checks the audience need, gathers evidence, writes a draft, reviews it, and approves publication.

Before building software around it, you should know what makes an idea worth using, which claims need sources, and who can approve the finished article.

If those rules change every time someone asks, the tool will be built around guesses.

A few manual runs can reveal gaps much more cheaply than rebuilding a large automated system.

Which Parts Should You Define First?

Start with five simple questions.

  1. What starts the work?
  2. What information does it need?
  3. What should the finished result contain?
  4. What can go wrong?
  5. Who decides when the result is ready?

Write the answers in plain language. A teammate should be able to follow them without needing to read your mind.

Then use a real example to test the description. If the instructions fail, improve them. Do not assume the person following them is the problem.

Define the workflow before you automate it
1
Trigger
What starts the work
2
Inputs
What information it needs
3
Output
What the finished result should contain
4
Failure points
What can go wrong
5
Approval
Who decides when the result is ready

What Should You Automate First?

Look for a task that is frequent, clear, and easy to check.

Moving an approved request into the right queue may be a better start than asking AI to decide which client deserves priority. Formatting known data may be simpler than interpreting unclear feedback.

Good early candidateWhy it may fit
Copy approved data between toolsInputs and expected output are clear
Flag a missing required fieldThe rule is simple to check
Route work to a named ownerThe path can be defined
Prepare a draft for reviewA person can check before use

Start where a mistake is easy to see and correct. Build confidence through useful work, not through the size of the demo.

How Should You Handle Exceptions?

Give the process a safe way to stop and ask for help.

A missing field, conflicting request, or unusual case should not force the tool to invent an answer. It should route the item to a person with enough context to decide.

For example, a system may draft a standard reply to a delivery question. A damaged order with a disputed refund may need a human review.

Write down who receives the exception and what they need to see. A vague error message sent to a shared inbox is not a complete handoff.

Also track repeated exceptions. They may reveal a missing rule or a task that is not ready to automate.

Where Does AI Fit in the Process?

AI can help with work that has variation, such as summarizing notes or preparing a first draft. That flexibility also means the result needs a suitable check.

Define what the tool may use, what it must not invent, and what requires review. Give it good examples and clear source material.

Then test it on more than the easy cases.

What happens when the notes conflict? What happens when a source is missing? What happens when a customer asks for something the business does not offer?

A system that handles a polished demo is not yet proof that it handles normal work.

How Do You Measure Whether Automation Helped?

Measure the whole job, including review and rework.

A draft created in ten seconds is not a time saving if someone spends an hour repairing it. A task moved quickly is not a success if it goes to the wrong person.

Compare the time, errors, and effort before and after. Also ask whether people trust the output enough to use it.

Use a small trial with a clear review date. Keep a way to do the work manually if the trial fails. That makes it easier to learn without making the whole business depend on a new tool at once.

When Should You Wait Before Automating?

Wait when the process has no stable goal, the inputs are unreliable, or nobody owns the output.

You can still experiment. A small prototype may help the team understand what it needs. Just treat the prototype as a learning tool, not a finished operating system.

You may also find that the best fix is simpler than automation: remove a step, clarify a brief, or stop asking for the same information twice.

The goal is better work. Automation is one possible way to get there.

FAQ

Must a process be fully manual before it can be automated?

No. You can develop the process and tool together in small steps. The key is to test assumptions and keep responsibility clear.

Should we automate the most time-consuming task first?

Not always. A smaller, clearer task may deliver a safer early win. Consider complexity, error cost, and how easily the result can be checked.

Who should own an automated workflow?

A business owner for the process should own the result. Technical support may maintain the tool, but someone still needs to decide whether the work is useful and correct.

What should happen when the workflow changes?

Update the rules, test the change, and tell the people who rely on it. An old automation can keep following an old process long after the team has moved on.

The bottom line

The most useful automation projects start with clear thinking about ordinary work.

What are we trying to do? What does good look like? Where do we need human judgment?

Answer those questions first. Then build a tool that helps the team do the right thing with less effort.

Keep reading

Pieces I've written on related topics that pair well with this one:

Keep reading
Subscribe to the newsletter

Get every post in your inbox.

New writing every two weeks. No fluff. Unsubscribe anytime.

Subscribe