Workflow automation: the four-step method for SMEs

Workflow automation works when you take it in a fixed order: map the workflows, prioritize them by time and regularity, build one at a time, and measure afterwards whether it saves what it promised. Most people skip the first two steps and start picking tools. That is the one truly expensive mistake in the whole exercise.

The tool is rarely the problem. The problem is that nobody has written down what actually happens between an order landing and that order being invoiced, or who does what along the way.

Below is the method we use with our clients, including the prioritization and the measurement that most guides leave out.

What is workflow automation?

Workflow automation means software performs the fixed steps of a recurring routine without a person having to start each step. A trigger sets the chain in motion, data is fetched and moved between systems, and a person only gets involved when something falls outside the rule.

It helps to separate two words that tend to get used interchangeably. Digitization is moving a paper process onto a screen. Automation is removing the clicks afterwards. A PDF invoice in your inbox is digitization. An invoice that is read, coded and sent for approval on its own is automation.

The Danish numbers are worth knowing before you start. The use of artificial intelligence in companies with more than ten employees rose from 15% in 2023 to 28% in 2024, and 70% of those using it point to more efficient workflows as a positive effect, according to Statistics Denmark. Among those who are considering it but have not started, 80% say they lack knowledge and experience in-house. That is not a technology problem. It is a method problem.

How to approach workflow automation: four steps

The method is the same whether the workflow is quoting in a trades business or order handling in a web shop. Four steps, in order, and you do not skip any of them.

The third box is the one everyone wants to start with. The first box is the one that decides whether the build hits the mark.

Steps 1 and 2: map and prioritize your workflows

Mapping sounds like something that needs a consultant and a giant spreadsheet. It does not. Sit down with the person who actually does the routine and write down every step in order, along with the system each step happens in.

Three questions keep the exercise on track:

• What starts the workflow? An email, an order, a date, a message?
• Which systems are touched along the way, and where is data typed in by hand?
• Where does it stall, and who is waiting on whom?

You are done when you can read the list aloud to someone who does not know the task and they can follow along. It typically takes an hour per workflow. Expect five to ten workflows in a company with 10-50 employees.

Two things always turn up in the mapping, and both are worth their weight in gold. The first is the shadow calculation nobody has mentioned: the spreadsheet on someone's desk that an employee maintains next to the system, because the system cannot do that one thing. The second is the double entry, where the same details are typed in twice because two departments never agreed on one place.

Write both down as steps. They look small on their own, and they are often most of the time the workflow costs.

Then you prioritize. Give each workflow a score from 1 to 5 on four questions, and add the numbers up.

The workflow with the highest total is your first. Not the most annoying one, not the most visible one, and not the one the boss mentioned at the last meeting. The one with the highest total.

In a Danish SME this method almost always lands in the same place: the weekly report assembled by hand from three systems, or the invoice chain from delivered job to paid invoice. Both are covered in detail in our guides to automated reporting and invoice automation.

Measure a baseline before you build

This is the step almost nobody takes, and it costs them the proof later. Before anything gets built, you time the workflow for a week.

Note three things: how many times it ran, how long each run took, and how many times something went wrong. It takes five minutes a day for the person doing the task.

Without those numbers you have no baseline. And without a baseline, the discussion about whether the automation worked turns into a feeling instead of a calculation.

Step 3: build one workflow at a time

Now, and only now, the choice of tool makes sense. Once the map is on the table, the choice more or less makes itself.

Which type of automation fits the workflow?

There are really three families to choose between, and the map tells you which one.

In practice most workflows end up as a mix. Integrations handle the fixed steps, and a model handles the one step where an email has to be read or an attachment understood. The order is worth holding on to: use the simplest thing that solves the step, and save the model for the step that genuinely varies.

Three rules keep the first build on track.

Build only the one workflow. The temptation to take number two along while you are at it is strong. Resist it. One chain in production is worth more than three half-finished ones, and the first one gives you experience you cannot get from reading.

Handle the exception from day one. Every chain has a case that falls outside the rule. It must never stop the whole chain, and it must never disappear silently. It should land with a person, with a note explaining why.

Settle ownership before a line of code is written. Who holds the account with the platform, who has the source code, and what happens if you end the collaboration? With us the answer is written into the agreement: you own the source code, the documentation and the access. You can read how we work, and what a build costs in our walkthrough of the price of AI automation.

In terms of time, this is shorter than people expect. From mapping to a chain in production typically takes three weeks for a scoped workflow.

Step 4: measure whether the automation worked

Four weeks after go-live, you take the same three numbers as in the baseline: number of runs, time per run, number of errors. Subtract one set from the other. That is the whole measurement.

Key numbers

15 hours

freed up per week for a typical client

70%

of Danish AI users see more efficient workflows

dst.dk

from $1,500

starting price for a scoped build, one-time

Two things deserve attention beyond the time. The first is the error rate: a chain that runs fast and wrong is more expensive than the manual process it replaced. The second is what the freed-up time actually gets used for. Time saved that just goes into more meetings is not a gain.

If you want to run the numbers before you decide, use our ROI calculator with your own figures from the baseline.

If the gain is there, you repeat the method on number two on the priority list. If it is not, it is almost always because the mapping was too shallow, and then the fix is cheap: back to step 1 with the same workflow.

Which workflows should you not automate?

Don't, if the workflow changes every month. You will end up maintaining more than you save, and you lock in a process that is still evolving.

Don't, if it needs judgment in every single case. A price negotiation, a complaint with emotion in it or a job interview are not candidates. The parts around them are: booking, reminders, minutes, follow-up.

Don't, if it runs a few times a month and takes ten minutes. The math does not hold up, however irritating it is.

And don't, if nobody in the company agrees on how the workflow should look. Then you are automating a disagreement, and it only gets more expensive by running faster. Get aligned first, build afterwards.

The gap between industries is large, by the way. Four out of five companies in information and communication use AI, compared with just over 20% in construction, according to Danmarks Nationalbank. If you are in an industry where few have started, there is more to gain in the administrative workflows nobody has touched yet.

FAQ about workflow automation

What is a workflow?

A workflow is the fixed sequence of steps that takes a task from start to finish. It has a trigger, a number of steps in one or more systems, and a result. An order that turns into a packing slip, an invoice and a bookkeeping entry is one workflow with four links.

What is the difference between RPA and AI automation?

RPA mimics a user's clicks in a program and follows fixed rules. It is strong when the system has no integration, and brittle when the screen changes. AI automation can handle free text, documents and variation, but the output has to be checked. In practice most workflows use a mix, where integrations handle the fixed parts and a model handles whatever varies.

How long does it take to automate a workflow?

A scoped workflow typically takes three weeks from mapping to production, and the first week is mapping and prioritizing. Larger chains that touch four or five systems take longer, but they should still be split up so that something goes live along the way.

Can you get a grant for workflow automation?

Yes. SMV:Digital gives grants of up to 250,000 DKK for digitization and automation projects in small and mid-sized businesses. The scheme suits a scoped project with a defined result, which is one more reason to map and prioritize before you apply.

Where do I start if we have never automated anything?

Start by mapping the three workflows that take up the most time in your day, and score them on the four questions above. Pick the highest, measure a baseline for a week, and build only that one. That experience is worth more than a plan for every workflow at once.

Workflow automation is not a choice of tool. It is four steps in the right order, and the first is always to write down what actually happens today.