Automate what’s really worth it.

Step 1Take stock, without the gloss

It all starts with a list, and an honest one. For two weeks, note down which tasks come back again and again. Six details per task are enough: what exactly is done, how often, how long it takes, who does it, which programs are involved and how often there are excep­tions.

Ask the people who actually do the work, not just their managers. The view from above almost always under­es­tim­ates the effort, because small manual steps stay invisible for as long as they work.

Step 2Assess every candidate soberly

Not everything that repeats is suitable for a machine. Five questions separate good can­did­ates from bad ones:

1. Volume: how much time does the task cost over a whole year?
2. Rules: can you describe in clear if-then sentences what needs to be done?
3. Stability: does the process stay the same for months, or does it change all the time?
4. Data: are the inputs digital and con­sist­ent, or do they arrive as photos, faxes and free text?
5. Con­sequences of errors: what happens if something goes wrong, and who would notice?

A strong candidate has high volume, clear rules, a stable process and digital inputs. Serious con­sequences of errors don’t rule a task out, but they call for a human check at the right point.

A simple cal­cu­la­tion helps with the first question. Enter your own figures:

The figure is an order of magnitude, not a promise. But it quickly shows whether a closer look is worth­while. A task that costs twenty hours a year rarely justifies a project. One that costs three hundred almost always does.

Step 3Prioritise: benefit against effort

Now your can­did­ates sit side by side. Place each one on a simple grid: effort of imple­ment­a­tion to the right, benefit upwards. Four areas emerge, and each calls for a different response.

Start with one or two ini­ti­at­ives from the top left. Not because they have the biggest leverage, but because they take effect in weeks rather than months. Your team exper­i­ences auto­ma­tion as relief rather than a threat, and you learn how your systems work together. You’ll need both for the big ini­ti­at­ives at the top right.

Step 4Simplify first, then automate

Before a single line of code is written, the process itself needs scrutiny. Which steps have become super­flu­ous? Which approval only exists because something went wrong years ago? Often a third of the effort can be cut without any tech­no­logy at all.

Next, stand­ard­ise the inputs: a form instead of free-form e-mails, fixed fields instead of running text, one template instead of ten variants. And decide what happens to excep­tions and who handles them. Only then do you choose the tool, and it should be the simplest one that solves the task reliably:

• Existing features first. Many programs you already pay for can do more than you use: rules in your mailbox, workflows in your account­ing software, inter­faces in your inventory system.
• Connect rather than build. Integ­ra­tion platforms link programs through their inter­faces, often without any custom devel­op­ment.
• Software robots only as a last resort. They click through screens like a person would. That helps when there is no interface, but it breaks easily as soon as a screen changes.
• Custom devel­op­ment when it pays off. For example, when the process is your com­pet­it­ive advantage or no ready-made solution fits.

Arti­fi­cial intel­li­gence is strong where inputs are unstruc­tured: reading invoices, sorting e-mails, preparing drafts. But it makes mistakes that look plausible. So plan for a human check wherever an error would cost money or trust.

Step 5Operation, measurement, ownership

Going live is not the end of the job. An auto­ma­tion is like a new colleague: it works reliably as long as someone looks after it. Three things make sure it will still do so in two years’ time.

Measure before and after. Before launch, record how long a task takes and how often errors occur. Only then can you prove later what the solution delivers, and decide what comes next.

Name an owner. Someone has to know how the auto­ma­tion works, approve changes and be available when something goes wrong. Without that person, the solution is orphaned at the first change of staff.

Make sure failures get noticed. An auto­ma­tion that fails silently is worse than none at all. A short e-mail alert when a task gets stuck, and a look at the figures once a quarter, are often enough.

FinallyStart small, measure properly

The most suc­cess­ful auto­ma­tions are rarely spec­tac­u­lar. They begin with an honest list, choose a man­age­able starting point and grow with exper­i­ence. Here are the five steps once more:

1. Record recurring tasks honestly for two weeks.
2. Assess each candidate by volume, rules, stability, data and con­sequences of errors.
3. Rank them by benefit and effort, and start with quick wins.
4. Simplify the process first, then choose the simplest tool that fits.
5. Measure before and after, name an owner, make failures visible.

Work this way and you gain more than time. You gain a team that exper­i­ences change as relief, and with it the best possible starting point for the next step.