--- title: "Which processes to automate first?" description: "The method that decides which processes get automated first: measure hours, frequency and error cost per process, rank the list, and leave the bottom alone." url: https://torn.studio/en/insights/which-processes-to-automate-first locale: en published: 2026-09-04 --- # Which processes to automate first? > **Short answer:** Start with the process that repeats every week, follows the same rules every run and eats the most hours. Measure three things per process: hours per week, frequency, and what an error costs. Rank by hours times hourly cost, automate the top of the list — and leave processes that change every month untouched. Selection is the part of an automation effort with the widest margin for error: a well-built flow on the wrong process never pays for itself. The question is settled by measurement, and the method fits in three numbers per process. This article covers the numbers, which processes tend to land at the top, and which to leave alone. ## Measure three things per process Hours per week: how much time the process actually takes, measured with the people who run it. Frequency: how often it runs — a daily process beats a monthly one on the same hours, because interruptions and context switches cost on top of the clock. Error cost: what it costs when a step goes wrong, in redone work or missed business. The ranking becomes hours times hourly cost, adjusted upward where errors are expensive. ## Which processes fit automation? - The process repeats weekly or more often — one-off work never carries a build cost. - The rules look the same every run, and the exceptions are few enough to list. - The process crosses systems: data moves by hand between CRM, accounting and spreadsheets. - The knowledge of how it is done lives in one person’s head — automating it documents it at the same time. ## Which processes should you leave alone? Processes that change every month, because every change is rebuild time — Swivrr budgets 15–25% of build cost per year in maintenance for stable flows already, and an unstable flow sits above that. Low-volume processes, where the hours never catch up with the build cost. And processes where every case needs a judgement call: there the human is the point, and what gets automated at most is the preparation around the judgement. ## How the review does the selection for you Torn Studio’s review is this method as a service: two weeks, €2,200, fixed price. It maps the processes with the people who run them, puts hours and money on each one, and ends in a ranked list with fixed quotes for the top three. The rest of the list stays unpriced on purpose — a process that lands far down gets the honest verdict that it should stay manual, and the document is yours to take anywhere. **The review’s two weeks run from mapping to measurement to a ranked list with fixed quotes for the top three processes.** - **Mapping with the people who run the processes** — 1 week - **Hours, frequency and error cost per process** — 0.5 weeks - **Ranked list and fixed quotes for the top three** — 0.5 weeks The review’s two weeks as the studio delivers them: mapping, measurement, and a ranked list with fixed prices at the top. **Read next** - [When does automation pay for itself?](https://torn.studio/en/insights/when-automation-pays-for-itself) - [Automating in-house or hiring: where the line sits](https://torn.studio/en/insights/automating-in-house-or-hiring) - [AI automation agency services at Torn Studio](https://torn.studio/en/services/ai-and-automation) - [What AI automation costs at Torn Studio](https://torn.studio/en/pricing/ai-automation) ## Common questions ### How many processes should we automate at once? One to three. The first automation teaches the organisation how flows are operated and who owns the exceptions, and that knowledge should settle before the next wave. The review prices three for the same reason: enough to show in the working week, small enough to land. ### Which processes are a poor fit for automation? Three kinds: processes that change every month, where every change is rebuild time; low-volume processes, where the hours never catch up with the build cost; and processes where every case needs a human judgement — there, at most the preparation gets automated. ### Does a process have to be completely free of exceptions? It has to be predictable. Occasional exceptions are handled by the flow handing over to a person in those cases, and exactly that handover is built in from the start. Once every tenth case is an exception, the build cost climbs fast — and the measurement shows it before a build is ordered. ### Who in the organisation should point out the processes? The people who run them. A leadership team knows which processes irritate; the people doing the work know where the hours actually go. The review therefore maps with the doers, and the measurement tends to reorder the list leadership would have guessed. ### How quickly does an automation show in the working week? Build time for a single flow is one to four weeks on the published Swedish market, and the effect arrives with deployment: the hours the process took are freed the same week. The review’s measurement makes the effect checkable — the before-hours are documented. ### What do we do with the processes at the bottom of the list? Leave them, and reread the list in six months. A frequently changing process may have stabilised, a low-volume process may have grown. Falling build costs move the line over time too — the list is a document with a date on it, and it ages. ## Sources - [Vad kostar AI-integration — Swivrr](https://www.swivrr.se/priser/vad-kostar-ai-integration) — Evidence for the maintenance figure of 15–25% of build cost per year, published for the Swedish market. - [Vad kostar automation — Snabba Pucks](https://snabbapucks.se/automation/vad-kostar-automation) — Evidence for the one-to-four-week build time for single flows on the published Swedish market.