What to Automate First When You Only Have One Budget

The first automation decides whether there is a second

This is the part nobody tells you: the first project is not really about the money it saves. It is about whether the business ends up believing automation works. Pick something visible and mechanical and you get a team that starts suggesting the next five. Pick something ambitious and half-finished and the subject quietly dies for two years.

So the selection criteria are not purely financial. You want the highest ratio of obvious benefit to delivery risk, which is rarely the biggest opportunity on the list.

A ranking that takes twenty minutes

List every repetitive task in the business, then score each one from 1 to 5 on four axes.

Multiply the four scores. Anything above 200 is a strong first candidate. Anything under 60 can wait, however loudly it is being complained about.

The four that pay back fastest

Across the small-business projects we have delivered, the same four keep winning.

1. Invoice and document handling. Extraction from PDFs into the accounting system. High frequency, zero judgement, and errors cost real money.
2. Order or enquiry intake. Anything arriving as free text and being typed into a system. Usually the single largest time sink in a small operation.
3. Scheduling and dispatch. Especially where availability depends on who covers what. This is a rules problem people are bad at and computers are good at.
4. Recurring reports. The Monday morning pack that takes someone three hours. Boring, weekly, and easy to verify against the old version.

The two that usually disappoint first-timers

Not because they are bad ideas, but because they are bad first ideas.

• Anything requiring your people to change how they work. If success depends on staff adopting a new habit, you are running a change project with some software in it, and it needs different handling.
• Prediction and forecasting. Demand forecasting is genuinely valuable and genuinely hard, needs clean history, and its benefit is statistical rather than visible. Save it for when the plumbing is trustworthy.

Scope it so it can fail cheaply

A good first phase is small enough that being wrong costs weeks rather than quarters. In practice that means one process, one intake channel, one output system, four to eight weeks, and a number you agreed before it started.

If a proposal for a first automation runs longer than three months, ask what could be delivered in six weeks instead. The answer is usually “most of the value”.

Run it in parallel before you trust it

For the first fortnight the automation should run alongside the manual process and the outputs should be compared. It feels like duplicated effort. It is the cheapest way to find the 5% of cases nobody mentioned in the specification — the customer with two delivery addresses, the product code that changed last year, the supplier who sends everything in one PDF.

Every one of those is discovered in parallel running or discovered in production. The first costs an afternoon; the second costs a customer.

What to measure so phase two is easy to justify

Decide the measurement before you build, because measuring afterwards is always disputed. Three numbers are enough: minutes per unit of work, errors per hundred units, and elapsed time from start to finish.

Take a baseline for two weeks before anything changes. That baseline is worth more than any projection in a proposal, and it turns the next budget conversation from an argument into arithmetic.

Frequently asked questions