Eliminate before you automate.

Automation has a seductive property: it makes any process faster, including the ones that shouldn't exist. And once a wasteful process is automated, it becomes considerably harder to see — it no longer consumes visible human hours, so nobody complains about it, so nobody questions it.

You have now paid to make a problem permanent and invisible. This is the most expensive mistake in modernization, and it is extremely common, because eliminating work requires a political conversation while automating work requires only a budget.

The correct sequence

Three steps, and the order is the entire point.

Step one: eliminate

Which of these steps could simply stop? Not be improved — stop. Every organization carries process that outlived its reason: approvals guarding against a risk that passed, reports three people open, reconciliations between two systems that could be one, data captured because someone once asked for it in 2019.

The test we use: if this step disappeared tomorrow, who would notice, and what specifically would go wrong? If nobody can name a consequence, you've found something. Expect defensiveness — the step exists because someone built it, and often that someone is in the room.

Step two: simplify

What survives elimination usually still carries unnecessary complexity: handoffs that could be merged, approval chains with more links than risk requires, exception paths built for a case that occurs twice a year.

Simplification is where most of the durable gain lives, and it's frequently free. It also makes the next step dramatically cheaper, because you're now automating something with fewer branches.

Step three: automate

Now automate what remains — and only now, because the thing you're automating is finally worth automating. The scope is smaller, the logic is simpler, the build is faster, and the payback period is shorter.

Teams that skip to step three routinely spend three to five times more, because they've automated every historical exception and workaround along with the actual work.

Why almost everyone skips to three

Elimination is a conversation about whether someone's work matters. Automation is a conversation about software. The second one is much more comfortable, and it can be delegated to a vendor.

There's also a budgeting artifact at play: capital for tools is often easier to approve than time for analysis. So companies buy the thing, because buying the thing is the approved-shaped activity.

This is a large part of why AI programs underdeliver. Research consistently finds that the bulk of the effort in a successful program is people and process, not algorithms — BCG puts the split at roughly seventy percent people and process, twenty percent technology and data, ten percent algorithms. Budgets are usually allocated in exactly the reverse proportion.

What this looks like in practice

Take a process map with real cost per step. For each step, ask in order: can this stop? Can this merge with another? Can this be automated? Record the answers, and the objections, because the objections are where you find out which constraints are real and which are habits.

In most engagements the elimination pass alone removes between a tenth and a third of the steps. That work is now free forever, took no software, and required no vendor.