If We Were Designing This Business Today, Would We Still Work This Way?

Somewhere in your business there is a step that exists because of a decision nobody remembers making.

Maybe every quote goes to a manager for approval. Maybe a report passes through three people before a client sees it. Maybe one person handles a certain kind of request because they have always handled it.

Ask why, and the answer is usually some version of: that is how we do it.

That answer is not laziness. It is what happens when a business runs for a long time. Work gets shaped by decisions made years ago, under conditions that may no longer exist, by people who may no longer be there.

The assumption hiding underneath

The assumption is that the current design is necessary. That the steps exist because the work genuinely requires them.

Often that was true once. A rule was written because something went wrong. A check was added because the information could not be trusted. One person became the bottleneck because they were the only one who knew the answer.

Then the business kept the step and forgot the reason.

You can usually spot this by watching what people do around a process rather than inside it. Someone keeps a private spreadsheet because the real system is awkward. Someone knows which approvals can be chased and which cannot. A team hires another person, not because there is more work, but because the work takes longer than it should.

Those are not signs of a lazy team. They are signs of a design that people have learned to live with.

This mistake has been made before, expensively

It is worth knowing how badly this has been handled in the past, because the same temptation is back.

In 1990, Michael Hammer published an article in Harvard Business Review with a blunt title: Reengineering Work: Don't Automate, Obliterate. Companies were spending heavily on computers and getting very little back. Hammer's explanation was that they were using new technology to speed up old processes instead of asking whether those processes should exist in that shape at all.

His argument was not that automation is bad. It was that automating a badly designed process gives you a faster badly designed process, and locks the design in place while you are at it.

The example he became known for was an accounts payable department. The company wanted to cut costs by installing a better system for matching invoices to orders. Then it looked at a competitor running the same function with a far smaller team. The difference was not a faster matching system. The competitor had removed the invoice from the process entirely.

One company was making a slow step quicker. The other had asked whether the step needed to happen.

That is the distinction worth holding on to. When you automate a step, you quietly decide that the step should exist.

What AI actually changes

For most of the last thirty years this was difficult advice to act on, because the constraints that shaped the work were real.

Reading a hundred documents took human hours. Writing a first draft took someone who could write. Answering a question about a past project meant finding the person who remembered it. Checking work against a set of rules meant someone who knew the rules.

Those costs shaped everything around them. You do not build a fast process on top of an expensive step. You build queues, batches, approvals and specialists instead.

AI changes the price of some of those constraints. Reading large volumes of text, producing a first version, checking work against documented rules, answering questions from material the business already has: these used to cost human time and now cost considerably less.

That does not mean AI should take those jobs. It means the reason your workflow was built that way may have expired, and nobody has gone back to check.

This is the useful question, and it is a different question from "where could we use AI":

If we were designing this business today, knowing what technology can now do, would we still design the work this way?

The evidence says this is where the money is

This is not only a philosophical point. There is recent evidence about which AI efforts actually show up in the accounts.

McKinsey's State of AI survey, published in March 2025, looked at a range of things organizations do when they adopt AI. Of everything tested, fundamentally redesigning workflows had the strongest link to earnings impact. Yet only around 21 percent of adopters had fundamentally redesigned any workflow. Companies getting the most value were close to three times more likely to have done it.

Read that twice. The most effective thing is also among the least practised.

That is not a coincidence. Buying a tool is a purchase. Redesigning a workflow is a management decision that touches roles, approvals, responsibilities and sometimes people's sense of what their job is. One can be done in an afternoon. The other cannot.

Speeding up the wrong step changes nothing

There is an old idea from operations that explains why so many improvements disappear without a trace.

In any process, one step sets the pace. Everything upstream of it produces work faster than it can be absorbed. Everything downstream waits. That step is the constraint.

If you improve a step that is not the constraint, total output does not change. You have made one part of the process faster and built a bigger queue in front of the real problem. The work still leaves the business at the same rate.

This matters enormously for AI projects, because the steps that are easiest to speed up are often not the constraint. Drafting is easy to accelerate. Summarising is easy to accelerate. Neither is usually the thing holding up delivery.

So follow it one step further. If drafting takes an hour instead of a day, what is now the slowest part? Frequently it is the review. The partner, the director, the one person who signs things off. You have not removed the bottleneck. You have moved it, and pointed more work at it.

That is not a reason to avoid the project. It is a reason to know, before you start, what will become the constraint afterwards, and to decide what you are going to do about it.

Redesign means more than installing something

Redesigning the work is not a technical activity. It usually means changing four things.

What the steps are. Not faster steps. Fewer, or different ones. The competitor in Hammer's example did not speed up invoice matching. They stopped producing the invoice.

Who decides. Many approvals exist because a manager could not see the work. If the work is now visible and checkable, the approval may be protecting nothing. Some approvals are genuinely about judgment or liability and must stay. The point is to tell the two apart deliberately rather than by habit.

Where the knowledge lives. If one person is the only route to an answer, that is a design, not a personality. It can be changed.

What good looks like. If nobody can say what a correct output is, no system can be built to produce one, and no reviewer can be relieved of checking everything.

Questions worth asking about your own business

Pick one process that matters. Something with real volume or real cost. Then ask:

Why does this workflow have the shape it has? What was true when it was designed?

Is that still true? What has changed about cost, speed, information or availability since then?

Which steps exist to produce something, and which exist to check, chase or move information?

If a step disappeared tomorrow, what would actually go wrong? Be specific. "It would be risky" is not an answer.

Which step sets the pace? If you speed up everything else, what happens to that step?

Who is the only person who can do a particular part? What would it take for that to stop being true?

If you were starting this business today, with the same clients and today's technology, would you build it this way?

The point

Most businesses did not choose their processes. They accumulated them. That is normal, and it is not a failure of management. It is what running a company for years actually looks like.

What has changed is that some of the constraints that shaped those processes have become much cheaper to remove. That creates a genuine opportunity, and a genuine trap. The trap is to take the current design as given and make it faster. You will spend real money and the business will work the same way, slightly quicker, with a new bottleneck where the old one used to be.

The opportunity is to treat AI as a reason to reopen decisions that have not been examined in years. Not all of them. One or two that matter.

The businesses getting value from this are not the ones with better tools. They are the ones willing to ask a harder question first.

If you are looking at your own business and suspect some of it was designed for conditions that no longer apply, that is exactly the conversation I enjoy.