Skip to content
caadoo
de

What to automate first: count for one week

Most companies pick their first automation by whatever sounds best in the management meeting. The better answer comes from a tally sheet your team keeps for one week. This piece covers how to count, how to choose, and how to fit the brief for the first build into a single paragraph.

By T. Krause7 min read

Somebody at the leadership table has been to a trade fair. They saw a demo where an AI read a tender document and wrote the response in four minutes, and now the question on the agenda is why your company isn't doing that yet.

It's a fair question with a bad starting point. The project that looks best in a demo tends to be the one with the most exceptions, the most departments involved and the least obvious way to measure whether it worked. Six months in, nobody can say if it paid off, and "AI" has become a word people roll their eyes at in the canteen.

Your first automation should come from somewhere less glamorous. It should come from a tally sheet.

Photo: Unsplash

Count where the hours go for one week

People are bad at knowing where their own time goes. The tasks that annoy them most are rarely the ones that cost the most hours, and the ones that cost the most hours have usually been around so long that nobody sees them anymore. Managers are worse at it than their teams, because they're guessing about someone else's day. I include myself in that.

So count. Pick the departments where you suspect the waste lives and give everyone a sheet for one working week. A shared spreadsheet works, and so does a printout taped next to the monitor. The people doing the work fill it in. One row per task, five columns.

How often. A stroke every time the task comes up. No estimates. "A few times a day" turns into 38 strokes by Friday, or into 5, and either result tells you something.

How long. Minutes per round, roughly timed. Six minutes to type up a service report sounds like nothing until you multiply it.

How many people. Some tasks sit with one person. Others are done by three colleagues in parallel, each with their own spreadsheet and their own way of doing it. That costs more than it looks, because every variant has to be maintained and nobody knows the other two exist.

How error-prone. A separate stroke for every time something had to be fixed afterward. A missing line on an invoice, a wrong delivery address, an order entered twice.

How annoying. A score from 1 to 5. It's the softest column and it still belongs on the sheet, because it tells you where the team will welcome an automation and where they'll quietly work around it.

By Friday you have a more honest picture than any workshop produces. Expect the top of the list to hold something nobody would ever have pitched as a project.

Score by frequency, time and the cost of mistakes

Now the arithmetic. Frequency times minutes gives you the weekly time per person. Multiply by the number of people doing it and you have the real load. Then add the errors. For each one, estimate what the fix cost: the phone call, the credit note, the second delivery, the customer who is now slightly less patient. Most people leave that number out. It's often the biggest one.

Sort by the total. Use the annoyance score only to break a tie.

What doesn't go into the score is how impressive the task sounds. That's the whole point of counting.

Take a 45-person machine builder. After every service call, the technician's handwritten report goes to the back office, where someone types hours, parts and travel time into the ERP so the invoice can go out. The tally shows 60 reports a week at six minutes each, so six hours. On top of that come the invoices that went out without a replaced part on them, which nobody notices until the quarterly stock count. Further down the sheet is the tender response the managing director saw at the fair. The company answers two tenders a month, and each takes half a day.

The tender is the more interesting problem. The service report is the more expensive one, and it goes first.

The first automation also has a second job beyond saving hours. It has to prove, inside your company, that this kind of thing works. That's why boring wins.

Frequent means you learn fast. A task that happens a dozen times a day shows you within a week whether the solution holds up and where it breaks. A task that happens once a quarter shows you in a year, and by then the process has changed.

Boring means nobody fights it. Nobody defends retyping service reports. Nobody's standing in the company depends on it. The people doing it today will be glad to lose it, and they'll help you test because they know exactly where the strange cases are.

Measurable means you can answer the obvious question. At some point someone asks whether it was worth it. You have the before number from the tally. Four weeks after go-live, count the same task again. The difference is your answer, and it's the best argument you'll have for the second automation.

What to leave for later

A high score doesn't automatically earn a spot at the front of the queue. Three kinds of process should wait, even when the numbers look good.

Rare processes. Year-end closing, the annual stock count, the autumn price update. Lots of effort in one go, too rare to learn from. By the time the automation runs for real the second time, half its assumptions are out of date.

Political processes. Anything where two departments disagree about who owns it, or where one person's influence rests on being the only one who knows how it works. Discount approvals often land here. Automating a conflict doesn't resolve it. It writes the conflict into software, and then everyone argues about the software.

Processes made of exceptions. If every answer to "how does this work here?" starts with "it depends", you don't have a process yet. Goodwill complaints, customers with special terms, orders that go through the owner personally. Write the rules down first. That alone often fixes half the problem, and sometimes it turns out the rules never existed.

These come later, once you know how your company reacts to automation and who on the team likes to test things. You'll know both after the first boring one.

Write the brief in one paragraph

Before anyone builds anything, write down what you want. One paragraph. If it runs longer, the task is too big or not yet understood.

The paragraph answers five questions. What comes in, and from where? What should exist at the end, and where does it live? Who looks at the result before it takes effect? What happens to the cases that don't fit? How will you know in four weeks that it works?

For the machine builder, it reads like this: "Technicians photograph their signed service report and send it to service@ from their phone. The automation reads hours, parts with article numbers, and travel time, and creates a draft invoice in the ERP against the right service order. The back office checks each draft and releases it. Reports it can't read, or that don't match an open order, go to the back office queue untouched, with a note saying why. Success means that after four weeks the back office spends its time checking drafts, not typing them."

That's not a specification document. It's the minimum clarity someone needs to start, and the minimum you need to check afterward whether they delivered. That holds whether the builder is your own IT person or someone from outside, such as an external Chief AI Officer. Pay attention to the sentence about the cases that don't fit. Leave it out and you'll get an automation that does something silent and wrong the first time a technician sends a blurry photo.

The first automation doesn't need to impress anyone. It needs to run, visibly, with a number from before and a number from after. The tender project will get its turn once the boring ones have paid for it. Count for a week before you decide anything, because the tally sheet knows your company better than the trade fair does.

Topics

  • Automation
  • Process analysis
  • Mittelstand
  • AI in business
  • Prioritization
All articles

Keep reading

  • 9 min read

    How to automate quote follow-up, and when to pick up the phone

    An open quote sits until someone remembers it. The CRM can track status, set reminders, create tasks and prepare follow-up emails, and a language model can draft the email from the last conversation. Calls and negotiation stay with sales. Here is a sample cadence to adapt and the setup in five steps.

  • 11 min read

    How to create a quote with AI: a prompt, a worked example and a checklist

    A language model can turn a customer request into a draft quote if you give it your item list, your price rules and your standard terms. A person still sets prices, discounts and delivery dates. Below is a prompt you can adapt, a worked example with the numbers checked, the data protection questions to settle first and a checklist for before you send.

End of call

You can hang up now.

Or book 15 minutes. I'll ask you the questions I'd ask on the phone, and tell you at the end whether it fits. If it doesn't, we've both lost a quarter of an hour instead of a month.

Or call. Cold, if you like. My AI assistant picks up and I call you back. +49 5136 9414999

Analytics

With your consent I use Google Analytics to see which pages get read. Without it, no analytics code loads. Your answer is stored locally in your browser.

Privacy notice