What to automate first, and what to leave alone

The rule: frequency × cycle time × error cost

Most automation advice fails because it starts with the technology. Start with three numbers you can measure this week without buying anything:

1. Frequency — how many times a month does this job run?
2. Cycle time — how many minutes does one run take, including the waiting and the chasing?
3. Error cost — what does one mistake cost in refunds, rework or an unhappy customer?

Multiply them. That product is your ranked list. The job at the top is the first build, and it is usually not the job your team complains about most — complaints track irritation, not cost.

A useful side effect: this arithmetic is also the honest answer to "how much should we pay for this?" Divide the annual product by twelve and the build should be a fraction of the monthly figure it removes.

The four that repay the build

1. Inbound reply and follow-up

Every small company has a shared inbox. The same twelve questions arrive, get answered by hand, and the answer lives in one person's head. A first build classifies the message, drafts the reply from an approved answer bank, and routes anything unusual to a named human. The saving is not typing time — it is the delay between a lead arriving and a person answering.

2. Intake through to quote

Forms, quotes and proposals share a shape: someone fills in fields, someone else re-keys those fields into a document. That re-keying is the whole opportunity. One field map, one template, one approval step.

3. The weekly numbers pack

Someone assembles the same report every week from three tools that all have an export button. The rule here is strict: the automation may collect and format, and it may write one sentence of factual summary. It may not pick the narrative. A report that argues on its own behalf gets distrusted within a month.

4. Invoice and subscription matching

The accounts payable pile is mechanical: read the invoice, match it to the purchase, flag the mismatch. Errors are expensive and the rule is written down. This is the least glamorous build and the easiest to justify.

The three that quietly cost more

• Anything without a settled rule. If two people would answer the same case differently, the automation will encode one of them and hide the disagreement. Settle the rule first; it is usually an afternoon of argument and a page of writing.
• The process that changes every month. Maintenance outweighs the saving. Revisit it in a quarter.
• Anything customer-facing with no approval step. Generation is cheap; a wrong promise sent to a customer is not. Keep a human between the draft and the send until the error rate is boring.

What "ready" looks like

Three tests, all boring:

| test | pass condition | |---|---| | Stability | the process ran the same way for three months | | Input shape | the data arrives in one predictable format, from one place | | Written rule | the decision fits on one page a new hire could follow |

Fail any one and the correct next project is not automation — it is a short piece of process writing.

Scoping the first build

One trigger, one transformation, one output, one approval. If your first version has more than three steps, cut scope until it has three. The goal of version one is not coverage, it is evidence: a measured reduction in cycle time that you can show to the person who has to approve version two.

Where to start this week

If you want a number before you want a meeting, the spend check takes five subscription lines and your headcount and tells you which line is heaviest — which is usually the process worth automating first. It runs in your browser and stores nothing.

If you would rather have the mapping done for you, book a 20-minute call. We scope one workflow, quote it in writing, and you decide whether to build it.