Which routine to automate first

The first automation decides whether there is a second. If it returns hours the team can feel, the next one is easy to agree. If it stalls, the budget, the team's patience and the owner's trust go with it.

Most firms pick their first routine for the wrong reason. It is the one someone complained about loudest this month, or the one a vendor demonstrated well. A better way is to ask the same seven questions of every candidate and let the answers pick.

Start where the hours are

List the routines people grumble about. For each one, write down roughly how many hours a week go into the slowest step, and who does that step. A rough number is fine at this stage. Three people spending an hour a day each on it is fifteen hours a week, and that is enough to rank.

Look past the hours as well. An enquiry that waits a day, or a quote that goes out late, costs work that never shows up on a timesheet. Keep the top three routines by hours and by what they put at risk. Those are your candidates.

Four questions decide if it is worth moving

How often does it happen? Daily, or many times a day, is a strong candidate. A few times a month rarely justifies building anything.

Are most cases the same? If most enquiries, invoices or bookings follow the same few paths, software can handle them. If every case is a judgement call, AI can draft, but a person still decides.

Does the information arrive in a usable form? Email, web forms and systems with a connection are ready. Paper, phone calls and memory need a step to capture them first. That step is often worth building, but it makes the move bigger.

Does it stay the same from month to month? A routine that changes every time the price list or the product range changes will break whatever is built on it. Settle it first, then automate it.

A routine that scores well on all four is worth moving. One that scores badly on how often and how clear is usually worth leaving at level 1.

Three questions decide how far it can go

What does a mistake cost? A mis-filed enquiry is cheap. A wrong figure in a client letter, a payment or a regulatory filing is not.

Can it be undone? Some mistakes are reversed in a click. Some take a day of rework. Some cannot be taken back once they leave the building.

How soon would anyone notice? A mistake that shows up at once is easy to catch. One that surfaces weeks later, or never, is the dangerous kind.

These three set the ceiling. The further a mistake can travel before someone sees it, the earlier a person has to be in the loop. A routine where mistakes are cheap, reversible and obvious can run at level 3, where a person handles only what it flags. A routine where they are expensive, permanent or silent belongs at level 2, where a person approves each run.

Two checks that can lower the ceiling

Personal information. A routine that handles health, employment, legal or financial details about individuals stays at level 2 until a privacy review says otherwise. The approval step is what keeps a person accountable for what the software does with that data.

Acceptance. Ask whether the people doing the work, and the clients on the other end, would expect this to be automated. A routine people would object to fails however well it is built.

A worked example

Take three routines in a firm of forty people.

Inbound enquiries. Dozens a day, most of them one of five kinds, arriving by email and web form. A wrong category is cheap, easy to fix and noticed at once. It scores well on every question. It is a strong first move, to level 2 now and to level 3 once it has run steadily for a few months.

Monthly reporting. Two hours a month pulling numbers from three systems into a spreadsheet. The rules are clear, but it happens once a month. It stays at level 1, with an assistant helping the person who builds it. Spend nothing here yet.

Client contracts. Every one is different, the mistakes reach a client, and they are hard to undo. An assistant can draft clauses at level 1. A person reads, decides and sends. That is the right level, not a gap.

The first move is enquiries. It has the most hours, the cleanest input and the cheapest mistakes, and at level 2 every run is logged, so it produces a number rather than an impression. That number is what makes the second move easy to agree.

When the answer is leave it

Some routines should not move at all. A routine that happens a few times a month, where every case needs a person's judgement, is fine where it is. The honest answer is an assistant at level 1, or nothing.

That answer is worth as much as the others. It stops money going into a build that would never pay back, and it keeps attention on the routine that will. Where automation stalls, and why covers what goes wrong when the first choice is made the other way.

The short version

Take the routine with the most hours. Check that it happens often, follows mostly the same rules, arrives in a usable form and stays stable. Let the cost of a mistake set how far it goes. Move it one level, count the result, then choose the next.

If you would rather not score them yourself, the free assessment places your routines on the ladder and names the next step for each.