Another Kind of 5S: Fix the Process Before You Automate It

A recent discussion about artificial intelligence and automation got me thinking about another kind of 5S.

Most people who have spent time around Lean manufacturing or continuous improvement are familiar with the traditional 5S approach to workplace organization: Sort, Set in Order, Shine, Standardize and Sustain. Each step builds on the previous one to create a workplace that is easier to manage, safer and more predictable.

When I look at processes, I use a different set of five S’s:

Stabilize — Make the results reasonably predictable.

Standardize — Agree on how the work should be done.

Simplify — Remove unnecessary steps, decisions and handoffs.

Systematize — Build the agreed way of working into the tools and workflow.

Scale — Repeat it successfully as volume or the organization grows.

The sequence matters.

And that sequence is becoming increasingly important as organizations look for opportunities to use AI.

Stabilize

Before improving or automating a process, I want to know whether it behaves reasonably consistently.

That does not mean every situation has to be identical. Most real processes contain some variation.

But there is a major difference between legitimate variation in customer requirements and a process where the outcome depends on who happens to be doing the work that day.

If one employee takes 20 minutes to complete something and another takes two hours, why?

If one person asks the customer for six pieces of information and another asks for nine, which is correct?

If work regularly comes back because required information is missing, where is the failure occurring?

If today’s backlog is three hours and tomorrow’s is three days, what is driving that variation?

Before asking how to automate the process, I want to understand what is preventing it from producing reasonably predictable results.

Otherwise, there is a serious risk that we simply make an unstable process move faster.

Standardize

Once the process is reasonably stable, the next question is whether we agree on how the work should be done.

That sounds obvious. In practice, it frequently isn’t.

Ask five people to explain the same process and you may get five slightly different answers.

One employee maintains a spreadsheet because the system does not provide information they need. Someone else keeps notes in Outlook. Another person has developed a shortcut that everyone knows about but nobody has formally documented.

There may be written procedures, but the procedures do not reflect how the work is actually done.

This matters when we start talking about automation.

A system needs rules.

If the organization has never decided what the rules are, the technology team is eventually forced to make those decisions during implementation.

At that point, a process problem has quietly become a technology problem.

Standardization does not mean removing judgement from every job. It means being clear about where judgement is genuinely required and where people are simply using different methods because nobody has established a better common approach.

Simplify

This is the step I think organizations are most likely to skip.

Once we understand how the process works and have established a consistent method, we should ask whether all of it needs to exist.

Why are we collecting this information?

Why does this approval exist?

Why is somebody copying information from one system into another?

Why are three departments involved?

Why does the customer have to provide information we already have?

Why are there twelve exceptions to the rule?

Why are we producing a report nobody appears to use?

There is a significant difference between automating work and eliminating work.

Eliminating unnecessary work is usually better.

If a process contains ten steps and three are unnecessary, automating all ten might look like an improvement.

Removing the three unnecessary steps and automating the remaining seven is a better one.

The same applies to AI.

If AI is being used to read an email, determine which information the sender forgot to provide, find that information somewhere else, translate their terminology into ours and then enter everything into a system, that may be an impressive technical solution.

But there may be a simpler question:

Why didn’t we collect the correct information in a consistent format at the beginning?

Sometimes there is a good answer.

Sometimes there isn’t.

Systematize

Only after stabilizing, standardizing and simplifying do I want to start building the process deeply into systems and technology.

At that point we understand the process much better.

We know what should happen.

We know which information is required.

We know which rules apply.

We know which exceptions are legitimate.

We have removed at least some unnecessary complexity.

Now we can make a much better decision about technology.

Sometimes the answer is AI.

Sometimes it is workflow software.

Sometimes it is robotic process automation.

Sometimes it is a database rule, a dropdown field or a barcode.

And sometimes the answer is simply a better form.

One of the risks in technology projects is using a very sophisticated solution to solve a relatively simple problem.

The objective should not be to use the most advanced technology available.

The objective should be to solve the operational problem.

Scale

Once the process works reliably, the next question is whether it can continue working as the organization grows.

A process that works for 100 transactions per month may fail completely at 10,000.

The same is true when a company adds locations, products, employees or customers.

Processes that originally depended on one experienced employee remembering how everything works eventually become bottlenecks.

Informal communication becomes unreliable.

Exceptions multiply.

Workarounds accumulate.

At this point, systematization becomes particularly valuable because it allows the organization to reproduce a proven process without relying entirely on individual knowledge.

But again, the sequence matters.

Scaling a good process creates leverage.

Scaling a bad process creates bigger problems.

Where AI Changes the Discussion

AI is particularly interesting because it can deal with information that traditional software struggles with.

Traditional automation generally likes structure.

Fields.

Tables.

Defined rules.

Consistent inputs.

AI can often operate where that structure is missing.

It can read emails.

Interpret documents.

Summarize notes.

Recognize different terminology.

Extract information from free-form text.

Categorize requests.

Identify patterns.

Apply relatively complicated decision criteria.

Those capabilities create enormous opportunities.

They also create a new risk.

AI can make a poorly designed process appear healthier than it really is.

Imagine a company receives customer requests by email.

Customers describe the same product using several different names. Some include a part number. Some don’t. Some provide quantities in the email. Others put them in an attachment. Important information is regularly missing.

An AI system could potentially read each message, interpret the terminology, extract the relevant information, identify missing fields and prepare a transaction for processing.

That could be extremely valuable.

But before implementing it, I would still ask:

Why do customers submit requests this way?

Could a structured form collect some of this information?

Could product terminology be standardized?

Could known customer data be populated automatically?

Could unnecessary choices be removed?

Which situations genuinely require interpretation?

We might discover that 80% of the process can be handled with simple rules and structured data, leaving AI to deal with the 20% that actually requires interpretation.

That is a very different implementation.

It is probably less expensive.

It should be easier to maintain.

It may be easier to audit.

And AI is being used where its capabilities actually provide an advantage.

Don’t Confuse Complexity With Disorder

There is real complexity in business.

Customers do not always behave predictably.

Documents arrive in different formats.

Language is ambiguous.

Products have unusual configurations.

Regulations create exceptions.

Experienced people sometimes need to exercise judgement.

These are exactly the kinds of situations where AI may provide enormous value.

But there is another category of complexity that we create ourselves.

Different departments use different terminology.

Nobody agreed which information is required.

Three approvals accumulated over the years and nobody remembers why.

Employees maintain separate spreadsheets because the main system does not support the actual process.

Procedures contradict one another.

Exceptions exist because nobody has reviewed whether the original rule still makes sense.

AI can potentially navigate all of that.

The fact that it can does not mean it should.

There is little value in paying sophisticated technology to continuously compensate for problems we could simply remove.

A Better Question Before Automating

When someone proposes an AI or automation solution, I think the first question should rarely be:

Can we automate this?

Technology can do enough today that the answer will increasingly be yes.

The more useful questions are:

What problem are we actually trying to solve?

How stable is the current process?

Do we have an agreed way of doing the work?

Which parts of the process are unnecessary?

Which complexity is real and which have we created ourselves?

What is the simplest technology capable of doing the job reliably?

And where does AI genuinely add something that simpler technology cannot?

That brings us back to the five S’s:

Stabilize.

Standardize.

Simplify.

Systematize.

Scale.

Do the first three properly and one of two things often happens.

You discover that AI was not really necessary. A much simpler and less expensive solution can handle the process.

Or AI remains the right choice, but now you are giving it a smaller, clearer and better-defined job.

Both are good outcomes.

AI should help organizations manage real complexity, make better decisions and do things that were previously difficult or impossible.

It should not become an expensive way to avoid fixing disorder we created ourselves.