← All articles
AI Strategy3 min read

Before you automate it, ask whether the workflow should exist at all

Before you automate a painful workflow, ask whether it should exist at all. Delete the ritual first — then automate only the work that survives.

Before you automate it, ask whether the workflow should exist at all

Every automation conversation starts in the wrong place. A team points at a slow, painful workflow — approvals that take days, data retyped between systems, reports assembled by hand — and asks how to automate it. The better question comes first: should this workflow exist at all?

Automating a workflow you do not need is the most expensive way to keep it. You pay to build the agent, pay to maintain it, pay to monitor it, and pay every time the underlying process changes and the automation quietly breaks. A workflow that should have been deleted now has a machine defending its existence, and the team that owns it has one more system to feed.

We see this pattern in almost every company we work with: the approval chain that exists because one bad decision in 2019 made someone nervous, the weekly report nobody reads, the three-step handoff between teams that stopped trusting each other's spreadsheets. None of these are automation problems. They are process problems wearing an automation costume.

Delete, then automate

The rule we use is simple: before a workflow gets an agent, it has to survive a deletion test. If we stopped doing it entirely tomorrow, what would actually break? If the honest answer is "a report nobody opens" or "a sign-off that adds no information", the workflow gets deleted, not automated. The automation budget goes to the work that survives.

This is not a theoretical exercise. One client wanted an agent to chase approvals across four departments; the chain took nine days on average. When we asked what the approvals actually caught, the answer was: almost nothing — the same people approved the same requests 98% of the time, and the rejections were all cases the requester already knew about. The chain was a habit, not a control. We replaced four approvals with one automated check and a single human sign-off for exceptions. Nine days became four hours, and no agent was needed to chase anything.

The test works because it separates the work from the ritual. Work has a purpose you can name. Ritual has a history and a defender. If nobody can say what breaks when the workflow disappears, the workflow is the thing to delete.

The automation that earns its keep

None of this is an argument against automation. It is an argument for aiming it at work that deserves to exist. The workflows worth automating are the ones that survive the deletion test and still hurt: high volume, clear rules, real cost when they are slow. Those are the automations that pay back in weeks, because every hour they save was an hour someone was actually spending.

The discipline is to do the deletion first. Delete the steps that add nothing, merge the handoffs that exist only because two teams stopped talking, and only then automate what remains. An agent built on a cleaned-up process is simpler, cheaper, and breaks less often.

We help teams find the AI work actually worth doing — starting with the workflows that should disappear before anything gets automated — because the fastest automation project is the one you never have to build.

Field note: A client once asked us to automate a monthly reconciliation that took two people three days. We spent the first afternoon asking why the reconciliation existed. Two systems held the same data and disagreed because of a timezone setting. We fixed the setting, and the reconciliation — and the automation project — vanished in an afternoon. The agent we did not build was the best automation decision of that engagement.

Related service AI products

Building something worth shipping?

We take on a small number of AI product engagements. Tell us what you are building — we reply quickly.