Orlix[ operations ] · · 6 min
The automation nobody asks for is usually the one that pays
Teams pitch us the demo-friendly workflow. The one that changes their quarter is almost always duller, older, and already half-documented.
When a team evaluates Orlix, they arrive with a workflow in mind. It is almost always the exciting one: the customer-facing agent, the generative report, the thing that will look good in the all-hands deck.
We have learned to ask a different question first: what work does your team do every week that everyone agrees is beneath them?
The shape of the boring winner
The workflows that pay back fastest share a profile:
- High frequency, low glamour. Invoice triage. Claims intake. Contract term extraction. Hundreds to thousands of runs a week, each one small.
- Already documented. Someone wrote the runbook years ago because the work was tedious enough to hand off. A documented process is a process you can encode as steps.
- A clear human decision inside it. There is one moment where judgement matters — approve the exception, flag the clause — surrounded by mechanical work that doesn't.
That last property is the tell. If the whole process is judgement, automation adds risk. If none of it is, a cron job would have done. The valuable middle is a pipeline with one gate.
Why demos mislead
The demo-friendly workflow optimises for novelty: it shows a model doing something nobody expected. The production-friendly workflow optimises for volume × error cost: it takes something a model can already do at 99% and wraps the remaining 1% in an approval gate.
One of our customers moved eleven document workflows into production in a quarter. None of them would demo well. Together they retired about 40 hours of weekly toil and — more importantly — produced an audit trail their compliance team now treats as the system of record.
A selection heuristic
Score each candidate workflow 1–5 on four axes:
- Frequency — how many times a week does this run?
- Documentation — could a new hire execute it from the wiki alone?
- Gate clarity — is the human decision inside it a yes/no with visible context?
- Blast radius — if a run goes wrong, is the damage bounded and reversible?
Anything scoring 16+ ships in weeks. Anything under 10 is a research project wearing a workflow costume.
The exciting project is not wrong — it's just third. Ship two boring ones first. They fund the interesting one, and they teach your team how the platform behaves before the stakes rise.