Practical guide · Discovery

Which process should you automate first? Not the one that screams loudest.

Almost everyone starts with the process that annoys them most today. But annoyance isn’t impact, and the best candidates are often invisible. Here’s the four-step funnel to find the right one, with the tools to do it yourself.

Raffaele Zarrelli·Founder, Yempik·May 30, 2026·11 min read
In summary
  • The right process isn’t the one that annoys you most: it’s the one that’s high-impact and recurring.
  • The best candidates are often invisible; people run them almost unconsciously.
  • A 4-step funnel: surface them, filter with signals, find the constraint, score them.
  • At equal impact, start with the bottleneck: it’s the only point that moves the output.
The problem

The right process isn’t the one that screams loudest

When you’re deciding where to start with automation, instinct pulls you toward the most annoying process: the one everyone complains about. But perceived annoyance doesn’t match real impact, and the noisiest process is almost never the one that, once fixed, actually moves the numbers.

You need a method that starts from many processes and lands on a single one, the right one. It’s a four-step funnel: surface the processes, filter them with signals, find the constraint, and finally score the candidates. Let’s walk through them one by one, with a tool for each.

Step 1

Surface the processes (even the invisible ones)

The first mistake is asking people “what processes do you have.” They answer with the manual, the official flow, not the real one. The most automatable processes are exactly the ones no one mentions, because whoever runs them does so almost unconsciously.

The best candidate is often the one no one sees: hopping between four tabs to close a single task doesn’t feel like a problem, because you do it on autopilot.

To surface them you don’t ask, you watch where the time goes. Two sources: people (what they actually do in a day, not what they think they do) and the data in your business systems, so-called process mining, which shows the real flow instead of the imagined one.[2] Here’s where to look.

The tab-hopping

A person opens 4 platforms to close a single task. They do it on autopilot; they never flag it to you.

Copy-paste between tools

Data moved by hand from one system to another: the sign that two tools don’t talk to each other.

Queues and waiting

Work stuck waiting for an approval, a file, or a reply.

Repeated requests by email

The same question that comes back every week, handled from scratch each time.

The shadow spreadsheets

The personal spreadsheet that holds up a piece of the process outside the official systems.

The 2–30 minute micro-tasks

Data entry, routing, and routine approvals that repeat many times a day.

Step 2

Filter with the five signals

From the long list in step 1, keep only the processes that show these signals. The more signals light up, the better the candidate.

  • High volume and repetitiveness: it happens many times, always similar.
  • High variability: two people run it in different ways.
  • A lot of person-time spent by hand, on low-value tasks.
  • Frequent errors and rework.
  • Queues and waiting: work piles up before or after this step.

At this point you’ve gone from a long list down to a handful of serious candidates. Now the step up in quality: among these, which one matters most?

Step 3

Find the constraint, not the noise

A chain is only as strong as its weakest link. In a process, total output is set by the slowest step, the bottleneck. Improving a step that isn’t the constraint doesn’t raise the overall result: it’s wasted work.[3]

Request intake
100/day
Lead qualification
100/day
Custom quote
40/day◆ bottleneck
Send and follow-up
100/day
Real output of the whole process= 40/day, as much as the weakest link can handle
Putting an agent on the 100/day steps changes nothing: the system stays stuck at 40. It pays to standardize (and then automate) only the link that chokes the flow.
How to spot the constraint
  • Work piles up right before this step.
  • It’s the point that needs constant nudging to keep moving.
  • When it slows down, everything downstream slows down too.
Step 4

Score it: the scorecard

Two or three candidates remain. To choose without deciding by gut feel, score them on the four dimensions that matter.[1] Fill it in for each one and compare the scores: the first process to automate is the one that wins.

Tool · scorecard

Score the candidate

Rate a process on the four dimensions that matter. Repeat for each candidate and compare the scores.

1. Economic impact

How much it weighs on revenue, costs, or risk

2. Volume and frequency

How many times it repeats

3. Degree of standardization

How repeatable and uniform it already is across everyone

4. Is it a bottleneck?

How much it chokes the rest of the flow

Heads-up

The trap: high impact, zero standard

There’s a case where the scorecard warns you instead of giving you the green light: the process that’s perfect on impact but still poorly standardized. It’s the trap almost everyone falls into, because automating a messy process doesn’t improve it, it amplifies the mess.

Here the project isn’t “buy the automation,” it’s first making the process repeatable and measurable. How to do it, step by step, we explain in how to standardize a process before automating it. It’s the natural next step, once you’ve picked the candidate.

If the winning candidate isn’t standard yet, you’ve still won: you know where to start. It’s just that the first job isn’t the AI, it’s the standard.

Want a hand finding your first process?

On a call we do the discovery together: we map the real processes and pinpoint the right candidate to start from.

Book a call
FAQ

The questions we get asked most

Which process is worth starting with for automation?

Not the most annoying one, but the one that’s high-impact, recurring, already fairly standardized, and that chokes the rest of the flow (the bottleneck). To find it you need a funnel: surface all the processes, filter them with signals, identify the constraint, and finally score the candidates with a scorecard.

How do I discover the processes people don’t mention?

The most automatable processes are often the ones people run almost unconsciously, like hopping between several platforms to close a single task. They don’t surface by asking “what processes do you have”: they surface by watching where the time goes, the copy-paste between tools, the queues, the repeated emails, and the shadow spreadsheets. Data from your business systems (process mining) helps you see the real flow instead of the stated one.

What is the bottleneck and why does it matter in the choice?

It’s the slowest step in a process: the one that limits the output of the whole chain. Improving or automating a step that isn’t the constraint doesn’t raise the overall result. That’s why, at equal impact, the first process to automate is often the one that contains or coincides with the bottleneck.

And if the best process isn’t standardized yet?

Then you don’t automate it right away: first you standardize it. A high-impact but poorly repeatable process is exactly where almost everyone fails, because automation amplifies the mess. Make it repeatable and measurable first, then automate it.

Transparency note

I wrote this article myself. The method and the opinions are the result of my own work and Yempik’s real projects. For the writing I got help from Claude Opus 4.8 on editing, clarity, and layout. The substance is mine; the tool is disclosed.

Transparency

Sources

  1. [1]Centric Consulting: a scorecard to identify the processes best suited to automation. centricconsulting.com
  2. [2]UiPath: task mining pinpoints repetitive micro-tasks (2–30 minutes) as ideal candidates. www.uipath.com
  3. [3]Lean Production: the Theory of Constraints and the five steps to manage the constraint. www.leanproduction.com