Module 3Academy

Finding processes that are worth AI

Process mapping, working out the return and choosing the one process that becomes a system first.

Finding processes that are worth AI is a training day in which the team maps its own work processes, works out the return per process and chooses one or two candidates for a blueprint. The team learns which work suits AI and which work is better left with people. The result is a ranked list with a number per process.

After this module the team can
  • 01The team maps the main work processes, with input, steps, owner and output
  • 02Per process the team calculates weekly frequency, minutes per time and annual hours
  • 03The team scores each process on five criteria: volume, regularity, structure of the input, checkability and access
  • 04The team chooses one or two processes for a blueprint, and records why the rest come later or drop out
Agenda
  1. 01

    The process on paper

    We draw one process per group, from first arrival to the result. Where it stalls and who waits for it comes with it.

  2. 02

    Working out the return

    Frequency times minutes times number of weeks, converted to hours per year. We use a number the participants supply, and set a low and a high scenario beside it.

  3. 03

    The five criteria

    Volume, regularity, structure of the input, checkability of the result and access to the systems. Each process gets a score, and the scores are discussed, not only added up.

  4. 04

    The choice

    The team chooses, and records the choice in writing with the reason. The processes that drop out get a sentence too: why not, and what would have to change.

Why the start is the process

Most AI projects that stall began with a tool or an idea, not with a process. Someone saw a demo, the company bought a licence, and afterwards somebody had to work out what it was for. That is the wrong order. The work where AI earns something is often dull, recurring and has been done the same way for years.

This module reverses the order. We start with the work that comes back every day, and only then look at which tool fits. That gives a shorter list and a more honest picture of what it delivers.

Process mapping without a project plan

Mapping a process does not have to take months. We use a sheet of paper or a board with five columns: what comes in, what we do, who does it, what comes out, and where it waits. It is about reality, not the ideal process from the handbook. That is why we ask the people who do the work every day.

The most important finding in process mapping is usually the waiting time. An email that sits on someone’s desk for two days is not arithmetic but delay. That delay belongs in the return, because a system that prepares faster but leaves the desk unchanged delivers less than it seems.

Working out the return without making the numbers pretty

The formula is simple: frequency per week times minutes per time times 52, divided by 60, gives hours per year. The number the team enters is the real work. That is why we always show two scenarios: the cautious one, with the minutes someone says it costs, and the high one, with the time it takes including corrections and consultation. The gap between those two is often the whole discussion.

There is a second step: what is an hour of this work worth, and what does the system cost to deliver it? That need not be exact. It only has to be honest enough to make a choice. A process that costs two hours a year does not justify a build; one that costs three hundred hours usually does.

Five criteria that decide whether a process fits

Volume: does it come back often enough to earn back the build? Regularity: is the work predictable, or does it depend on exceptions? Structure of the input: does it arrive as structured data, such as an invoice PDF or an order, or as free text that a person must read? Checkability: can someone quickly see whether the result is right? Access: can we reach the systems where the work happens within a reasonable time?

A process that scores on all five is a candidate for the first build. A process that scores on volume but not on checkability needs a control step first. A process that rarely occurs does not need AI; a good work instruction is often enough. We show that with practical examples such as purchase invoices or order handling.

Which process first, and what to wait for

The first process has to be small enough to work within weeks and large enough to convince the team. That is the paradox where most choices get stuck. The training ends with one or two candidates for a blueprint, and with a list of what comes later. The most important part of that list is the reason: a process that drops out because access is missing is a different kind of decision from one that drops out because it is no longer needed.

Frequently asked questions

  • Do we need a process model, such as BPMN?

    No. A sheet of paper with five columns is enough for this step. A formal model may be needed later, for example for a large build, but it does not belong in the first day.

  • What if the participants’ numbers are wrong?

    That is often the first result of the day. We put the cautious and the high scenario side by side, and only then choose. An estimate that proves too low afterwards is not a problem; an estimate nobody checks is.

  • Is this the first step of a build day?

    Often, yes. A build day starts with exactly this work: which bottlenecks, how often, and what can go live on the day. See the build day at /workshops/.

Which team needs this module?

For up to 12 participants €3,950 excl. VAT. In the conversation we say whether the module fits, or whether a build day or the adoption programme is better.