Knowledge base, automation

Automating business processes: which order actually works?

Not every task a computer can handle deserves priority. You set the order with a simple calculation, frequency times cost of errors times wait time, not with whatever annoys you most right now.

Why 'easiest task first' is the wrong question

A lot of advice on automating business processes starts with the question of which tasks a computer can or can't handle. By now that's the wrong first question. Almost every repetitive task in a small or midsize business can be automated today: booking invoices, sending quotes, updating stock, answering customer questions. The technology is rarely the problem. The question that really matters is the order in which you tackle those processes.

If you start without an order, you usually pick the process that happens to be most visible, or the one the owner is most annoyed by. That feels logical on a Monday morning, but it rarely gives the best result for your first automation. I see this pattern in nearly every conversation I have through /en/contact/. There's a list of irritations: an Excel file that keeps freezing, a colleague who gets the same question three times a day, two systems that don't talk to each other.

The tendency is to start with whatever hurts most right now. But pain is a feeling, not a number. For a good first choice in automating business processes, you need a calculation, not gut feeling. On the pillar page /en/automation/ I explain how automation works in the broad sense. In this article I go deeper into the order. What do you tackle first, and why?

The calculation that does work: frequency times cost of errors times wait time

I set the order in which you automate processes with three factors that you multiply, not add. A process that scores something on all three weighs more than a process that scores extremely high on one and zero on the other two.

  • Frequency: how often does this process happen, per day, per week or per month? A task that comes back twenty times a day weighs more than a task that happens once a quarter, even if that one time takes more effort in one go.
  • Cost of errors: what does a mistake in this process cost? Think of an incorrectly sent invoice, a missed delivery date or a double booking. Some errors take five minutes to fix, others cost a customer or a fine.
  • Wait time: how long does a customer, colleague or next process sit idle before there's an answer or result? A process someone has to wait a day for carries a different weight than something that's handled within a minute.

Multiply these three factors and a surprising winner often comes out on top. It's not the process that feels most like a millstone around your neck. It's the process that scores fairly high on all three axes at once. That's exactly why this calculation is more useful than a simple list of irritations.

Why the process that annoys you most is rarely number one

Irritation is usually based on one factor, not three. Retyping a customer name in two systems happens often (high frequency) and feels tedious because of it, but the cost of errors is low. A typo in a name costs a correction email at most. The wait time is low too, since it happens right away. Still, this is often the process business owners want to fix first, simply because it repeats every day right in front of them.

A duller process, like checking incoming invoices for discrepancies, gets less attention. It doesn't happen every minute, so it doesn't feel urgent. But the cost of errors can be high. A missed discrepancy costs money, sometimes a lot. And wait time matters too when a supplier has to wait for approval. In the calculation this process often scores higher than the annoying retyping, while in daily experience it's barely seen as a problem.

There's also recency bias at play. Yesterday's incident feels more urgent than last year's structural problem. If you base your business process automation on what went wrong this week, you'll almost always choose wrong. The calculation with frequency, cost of errors and wait time corrects that distortion, because it doesn't look at how today feels but at the pattern over time.

Step by step: ranking your own processes

This calculation isn't a complicated exercise that takes weeks. Most small and midsize businesses can do it in an afternoon, as long as they follow the right steps.

  1. List all recurring processes, not individual tasks. A process has a clear start and end, for example 'order comes in to invoice sent', not 'answer an email'.
  2. Give each process a score from 1 to 5 for frequency, cost of errors and wait time. Do this together with someone who runs the process every day, not only from the owner's perspective.
  3. Multiply the three scores for each process. Sort the list from high to low.
  4. Only then do you look at feasibility. Can this process be handled with an integration between existing systems, with workflow automation, or does it need an AI agent that makes decisions?
  5. Start with the process that scores high and is doable within a reasonable timeframe. A process that's theoretically number one but needs three systems that don't talk to each other yet is sometimes better as a second project.

This order keeps you from starting a project that draws a lot of attention but delivers little, and from skipping the process that structurally costs the most money or time without anyone saying it out loud.

An example: three processes side by side

Say a small business has three candidates for its first automation. First, manually retyping order details between the webshop and the accounting system. That happens daily, an error costs a correction at most and the wait time is minimal. Second, answering the same question about delivery times over the phone again and again, also daily, with a low cost of errors and a short wait for the customer. Third, the yearly preparation of the VAT return, which only happens a few times a year but can lead to an additional assessment or a fine if something goes wrong, and where an accountant sometimes has to wait days for documents to come in.

Based on irritation, you'd probably start with the retyping, since that's what causes daily frustration. But in the calculation the VAT preparation often scores higher. The cost of errors is much higher than with a typo, and the wait for documents can run to days. The frequency is low, but the weight of the other two factors more than makes up for it.

This example is illustrative, not from a specific client file, but I recognize the pattern every time. The process that rarely stands out often turns out, once you run the numbers, to have the strongest case for being automated first.

Not every process needs the same solution

Once you know which process goes first, the next question is which technology you use to tackle it. Automating business processes isn't one-size-fits-all. Different processes need different tools.

  • Is it about data that has to move from one system to another, without anyone needing to look at it anymore? Then an integration is often enough.
  • Does the process consist of a fixed series of steps you now go through by hand, like an approval or a notification? Then workflow automation fits better.
  • Does something need to be assessed or interpreted, like reading an incoming email and responding to it on its own? Then you look toward an AI agent.
  • Does no existing package fit the way your business works? Then custom software is sometimes the only route that holds up in the long run.

On /en/automation/ I explain more broadly how these different forms relate to each other. For this article, what matters most is that you set the order first based on the calculation, and only then look at which technology fits that specific process. Working the other way around, picking a technology first and then looking for a process to match, often leads to a solution that delivers little.

Common mistakes when choosing the order

In practice I see a few mistakes come up again and again with business owners who automate their business processes without setting the order first.

  • Starting with the most expensive or most visible project because it gets the most attention from management, without checking whether it also delivers the most gain.
  • Looking only at frequency and ignoring the cost of errors, so a process that happens often but costs little when it goes wrong ends up at the top when it shouldn't.
  • Automating a process that will change in six months anyway because a new accounting package or CRM is being introduced. Get your systems stable first, then automate.
  • Waiting until every process has been mapped in detail before tackling anything at all. That leads to analysis paralysis, while a rough ranking is already enough to get started.
  • Picking the process the owner does most themselves, instead of the process that costs the business as a whole the most time or money.

These mistakes aren't stupid. They're human. Anyone close to their own business has trouble stepping back from what's annoying today. That's why a fixed method of calculation helps. It corrects for the distortion that's simply there when you're in the middle of the process.

Work it out yourself or have someone do it

You can do this ranking yourself just fine. Make a list, assign scores, multiply, done. For businesses with a handful of processes that's an afternoon's work. It gets harder at businesses with dozens of processes spread across several departments, where nobody has the full picture.

In that case an outsider helps you score objectively, without the emotional weight an employee or owner often brings to it. Through /en/services/ I set up projects like this, from the first ranking to actually building the integration, workflow or agent that comes out of it. More about how I approach that is on /en/about/.

If you're unsure whether to figure this out yourself or outsource it, /en/knowledge-base/build-or-outsource-automation/ is a good follow-up article. If you first want to know roughly what a first integration costs before you decide, take a look at /en/knowledge-base/what-does-an-integration-cost/. Both articles connect to the step that comes after you've set the order.

From a first process to an automation routine

The first process you automate is rarely the last. Once you see the calculation works, you'll repeat the routine. Every few months you go through the list of processes again, update the scores and pick the next project. That way you get a structural approach instead of a one-off project.

Some businesses track this progress in a dashboard, so it's clear which process has been tackled and which is still on the list. You can read more about that on /en/dashboard/. For the wider context of how automation relates to AI solutions and where the limits are, /en/automation/ is the place to start reading from.

Want to read more about related choices, like the difference between an AI agent and regular automation, or what's sensible to leave to AI and what isn't? You'll find those articles on /en/knowledge-base-free-practical-guides/. If you get stuck working out your own order, you can start that conversation directly through /en/contact/.

Frequently asked questions.

What's the difference between automating a task and automating a process?

A task is a single action, like answering an email. A process is a series of steps with a clear start and end, like from order to invoice. Automating business processes works better when you look at the process level, because then you also account for the wait time between steps.

Do I need to map all my processes before I start?

No, that often leads to delay. A rough list of the ten to fifteen most important processes is enough to apply the calculation. You can always add to the list later once you've finished your first automation.

What if two processes score about the same in the calculation?

Then look at feasibility. Which process can you tackle faster and with fewer dependencies? A process that scores slightly lower but can be done within two weeks often delivers results sooner than a project that touches three systems at once.

Does this calculation also apply if I'm considering an AI agent instead of simple automation?

Yes, you set the order the same way. The difference is in the execution. An AI agent suits processes that need judgment or interpretation, while an integration or workflow automation is enough for fixed, predictable steps.

How many processes do I need to automate before it's noticeable day to day?

Often the effect of the first process is already noticeable, as long as you chose it well based on frequency, cost of errors and wait time. Whether you notice it depends less on the number of processes and more on whether you tackled the right process first.

Curious where things get stuck for you?

Walk me through one process. In half an hour I go through your business and point out where work gets stuck, what it costs now and what an integration would save.

No newsletter spam. One email a month at most, and unsubscribing is just a reply.

Walk me through one process