Where automation actually delivers ROI — and where it doesn’t

The insight most people miss

One of the most surprising ROI wins in practice comes from what I call “low-value but high-friction administrative loops” — particularly approval chasing, status reporting, and exception handling in procurement and finance workflows.

Individually these don’t look expensive. But when you automate the handoffs between people, not just the tasks themselves, you often unlock cycle-time reductions of 30–60% without changing the core system at all. The task is fine. The waiting between tasks is the problem.

Most automation projects focus on making individual tasks faster. The real gains are in eliminating the dead time between them.

The most common mistake

The biggest automation failure I’ve seen, repeatedly, is automating a broken process exactly as it exists. Organisations take a slow, unclear, exception-heavy process and digitise it. The result is a faster version of the same mess — just harder to fix because now it’s “systemised”.

A real example: we inherited a 14-step invoice approval process. The request was to automate it. We didn’t — we redesigned it to four steps first. The automation then took two weeks instead of three months. And it actually worked when it went live because the process underneath it was sound.

The second common issue is overbuilding edge-case logic too early. Simple workflows become fragile mini-applications that only the person who built them can maintain. When that person leaves, the whole thing breaks.

Questions to ask before you automate

• Is this process stable and well-defined enough to codify?
• How often does it run, and what does the manual cost actually add up to across a year?
• What’s the error rate, and what does each error cost downstream?
• Will the people affected by the change actually adopt the new system?
• Can you explain the “happy path” in one sentence? If not, it’s not ready.

The rule of thumb

Strong automation candidate: High-frequency, rules-based, with clear start and end points and a low exception rate. Invoice processing, approval routing, compliance documentation, status reporting.

Automate only parts of it: Interpretive, exception-heavy, or relies on tacit human judgement at each step. You can automate the scaffolding but not the decision.

Don’t automate yet: The process is unclear, inconsistently followed, or actively changing. Fix the process first or the automation will embed the dysfunction.

Common mistakes that kill automation projects

• Automating a broken process — you get broken outcomes, faster
• Building automation the team doesn’t trust and works around
• Underestimating integration work — connecting to real systems is almost always harder than the workflow itself
• No handover plan — automation that only the builder can maintain
• No measurement baseline — if you didn’t measure before, you can’t prove the improvement