"We want to automate our processes" is a good resolution and a bad project brief. The real question is always the same: which process first? Here's the prioritization framework we've been using for years.
The typical starting point: after a walk through the company, fifteen candidates are on the table — from the quote PDF to holiday management. Budget and attention are enough for one, maybe two. Whoever decides by gut feeling or by the loudest department now will very likely automate the wrong thing.
The bad news first: there's no process catalogue that knows the right order for you. What's the most expensive manual step in a tax firm is a footnote in machine building. The good news: you don't need a catalogue — just three numbers per candidate and an afternoon of honesty.
Step zero: fill the candidate list honestly
Before anything can be prioritized, the list has to be right — and it doesn't emerge in the conference room. The most productive method we know: physically follow a single case through the building. One receipt, one order, one enquiry — from arrival to completion. Every point where something gets printed, retyped, copied over or passed along by shouting is a candidate. After two such walks, the list is fuller than the budget will ever be.
Second source: ask the annoyed. "What's the dumbest activity in your week?" is rudely phrased and still reliably delivers the best answers. The people who execute a process daily know its breaking points better than any process diagram.
The framework: frequency × rule-density × pain
We score every candidate on three axes, each from 1 to 5 points, and multiply the values. Sounds banal — the value lies in the fact that multiplication is merciless.
- Frequency. How often does the process run? Several times a day = 5, once a quarter = 1. Automation is a unit-cost business: the effort is fixed, the benefit scales with every repetition.
- Rule-density. Does it run the same way every time, with clear rules — or is there judgement in every case? A process you can teach a new colleague in an hour is a 5. One that "takes experience" is a 2.
- Pain. What does the process cost today — in hours, errors, revenue left on the table, or in the frustration of the people running it? An honest weekly calculation helps more here than any gut feeling.
Why multiply instead of add? Because a one in any dimension rightly kills the project. The annual financial statement hurts (pain 5) but is rare (frequency 1) — automation barely pays. Travel expense approval is frequent but full of judgement — a model can suggest, a human still has to decide. The winners are almost always the unspectacular processes: daily, dull, annoying.
The second advantage of multiplying: it forces a discussion about the scoring instead of about opinions. Whether a process's rule-density is a three or a four can be settled on the concrete case — whether someone "finds it important" cannot. We've seen scoring rounds in which exactly this shift changed the tone of the entire digitalization debate.
In practice we run this as a 30-minute exercise with the people who actually execute the processes — not just with management. The scores are rough, and that's fine: it's not about decimal places, it's about making rank 1 and rank 10 visible.
For the pain score, an intermediate step pays off: don't estimate, count for a week. A simple tally sheet — how often did the process run, how long did it take — beats any discussion. We've seen it several times: the perceived front-runner landed in fourth place after a week of tallying, and an inconspicuous everyday process came out on top.
Example 1: the document bridge in a tax firm
Getting receipts, payslips and notices from paper and email into the firm's software: in a tax firm this runs daily (frequency 5), by rules that are the same every time — amount, date, client, category (rule-density 5) — and it easily ties up two hours a day per staff member, around ten working weeks a year (pain 4 to 5). Hardly any process we know scores a higher product. Which is exactly why document automation became its own productized offer at our studio.
The pattern behind it transfers well beyond tax firms: a document arrives, fields are read, matched to the right case, filed in a structured way — and a human only checks the uncertain cases instead of all of them.
Example 2: small jobs in a plumbing and heating business
Second example, different industry: in plumbing and heating businesses, small jobs slip through — quickly done, never cleanly recorded, never billed. The process "record and bill the work" runs constantly (5), is highly rule-based (5), and the pain is especially honest to measure: it's revenue that has already been earned and is missing anyway. Our SykaSoft interface starts exactly there — capture on the fitter's phone, handed over in a structured form to the software the business already uses.
What's interesting about this case is the pain dimension: it shows up in no hourly calculation, because the work has long been done — only the invoice is missing. Gut-feeling prioritization systematically overlooks exactly these processes. The fitter feels no daily pain in not writing things up — on the contrary, it's more convenient. Only the multiplication of frequency and forgone revenue makes visible that one of the business's most expensive processes sits right here.
By the way: automating doesn't automatically mean AI
Because in 2026 every automation discussion lands on AI immediately, for clarity: in three out of four cases the lever is a clean interface, an importer or a cron job — technology that has worked for twenty years and is correspondingly cheap to get. AI enters the picture for us when unstructured input has to be read: a photographed receipt, a voice memo, an informal email. The criterion stays the same as for every feature — does it save clicks and hours, or is it a show?
What doesn't come first
Just as important as the winners' list is the courage to leave the rest alone. Four candidate types we regularly sort out:
- Rare processes — even if they're spectacularly annoying. Something that hurts once a year is cheaper than any software.
- Processes full of judgement. Where a human has to weigh things up in every second case, you only automate the paperwork — and that's rarely the actual problem.
- Processes that are changing right now. Stabilize first, then automate. Software freezes the status quo.
- Prestige projects. The AI dashboard for management loses to the receipt importer in every honest scoring — you just have to do the maths honestly.
Sorting something out isn't a rejection forever, by the way. Processes migrate: what still changes monthly this year may be stable next year — then it's allowed back on the list. The prioritization is a snapshot, not a verdict.
The pragmatic start
Once the first candidate is set, start small: a patch next to the existing system, not the big rebuild. The first automated process delivers two things no concept paper can — trust in the team and real numbers on what automation actually saves at your company. Both make project number two easier, cheaper and less political.
And: measure beforehand. The tally sheet from the prioritization is also your baseline. Whoever knows the process cost eleven hours a week before automation can put an honest number next to it after eight weeks — and with that has the argument for project number two that no slide deck can replace.
How we scope such projects — a two-hour conversation, then a written fixed-price sketch with an order of magnitude — is on our studio page. And if your rank-1 candidate is one of the two above: for documents and SykaSoft, the shortcut exists as a ready-made package.
