Which process should you automate first?
If you know of several manual processes, you have a choice to make. The best first candidate is rarely the one with the greatest potential, but the one that shows how automation works in your own company.
A good first candidate for automation is a process that comes up frequently, has a clear start and a clear end, can be measured before and after, where an error doesn’t cause major damage and whose information is technically accessible. The first project should do more than save time. It should show how automation works in your own company and lay the groundwork for the next steps.
This article assumes there are several candidates. Whether a single process is worth it at all is covered in the article When is process automation worth it? The two questions belong together but are answered in different places.
Step 1
Collect candidates without judging them straight away
Most companies know their most annoying processes, but rarely all of them. A list of five to ten processes helps, each described in one sentence: how it starts and how it ends.
Where to look can be said fairly precisely. Unnecessary manual work almost always arises in the same places:
- Someone reads emails or enquiries and transfers the details into a system.
- Someone opens PDFs and types up values.
- Someone copies information from one program into another.
- Someone gathers information for a quote that already exists somewhere.
- Someone reminds others that a case needs to be taken further.
The list is best put together with the people who do this work every day. They know where time is lost, and often also why a step is the way it is today.
Step 2
Put rough figures on it
For the comparison, a rough idea of frequency and time per case is enough. The point is not the amount in euros, but the order: which processes take up a lot of time, which take little? How to calculate the cost-benefit of a single process more precisely is covered in the article on whether automation is worth it.
Wanner and colleagues (2019) proposed prioritising automation candidates on the basis of measurable indicators rather than estimates. They use process mining for this, i.e. analyses of data from large IT systems. For a mid-sized company that is usually too much effort, but the basic idea still applies: where figures are missing, it is often the loudest process that wins, not the most important one. For the two or three most promising candidates, a short measurement is therefore worth it.
Step 3
Compare against six criteria
Economic leverage is one criterion, but not the only one. For a first project, what counts are properties that decide success and trust.
Economic leverage
How much time does the process take up today? This is the result from step 2. Frequent processes are the obvious candidates. Van der Aalst, Bichler and Heinzl (2018) describe how automation traditionally concentrates on the most frequent types of case, because the effort doesn’t pay off for rare cases, especially when several systems have to be connected.
Clear boundaries
Does the process have a clear start, a clear end and few people involved? A process that runs through three departments is hard to manage as a first project. A good fit is a sub-process whose result can be described in one sentence: after the project, process X works in way Y.
Measurability
Can the current state be captured and measured again after implementation? Only then can you say later whether anything has improved. Without a baseline, all you have at the end is the impression that it is “better now”.
Risk of errors
What happens if the solution makes a mistake? A misfiled document is annoying; a wrong price in a quote can be expensive. A good first project is a process where errors are noticed and can be corrected before they cause damage.
Technical accessibility
Does the information arrive digitally, and can the systems involved be addressed, for example via an interface, an export or an existing mailbox? A process whose data is on paper or in a closed legacy system is a difficult start, even if the benefit would be large.
Follow-on potential
Can what is built in the first project be used for other processes? Once you can read documents reliably, you can do the same for the next type of document. A one-off that isn’t needed anywhere else is a less valuable start.
A simple comparison grid
The criteria can be laid out side by side in a table, with a rough score from 1 (unfavourable for a first project) to 3 (favourable). The following example is made up and only shows how the grid works.
| Candidate (example) | Leverage | Boundaries | Measurable | Risk of errors | Technology | Follow-on potential |
|---|---|---|---|---|---|---|
| Transferring supplier order confirmations into the system | 2 | 3 | 3 | 2 | 2 | 3 |
| Creating quotes | 3 | 1 | 2 | 1 | 1 | 3 |
| Compiling the monthly report from several Excel lists | 1 | 3 | 3 | 3 | 3 | 1 |
The grid doesn’t produce a decision by adding up the scores; it makes differences visible. In the example, the quoting process would be the most interesting economically, but difficult as a first project: poorly bounded, with a high risk of errors and technically demanding. The monthly report would be quick to implement, but brings little and hardly helps with the next step. The order confirmations combine a noticeable benefit with clear boundaries and measurability. Here, they would be the sensible place to start.
A score of 1 for risk of errors or boundaries is almost always a reason to rule a process out as a first project, even if the other values are good.
Context
Why the biggest lever is rarely the first step
The processes with the greatest economic potential are often the ones that involve many people, systems and decisions. Quotes are a good example: prices, costing, technical details, special terms and approvals all interlock. That makes them valuable and, at the same time, risky as a starting point.
On top of that comes an effect that Lisanne Bainbridge described for industrial process control as early as 1983: what remains with people after automation is often the difficult cases, and they need more experience for those, not less. A first process with many exceptions therefore doesn’t just shift work; it makes the remaining work more demanding.
That doesn’t take the quoting process off the table. It is often suitable as a second or third project, and even then usually in part first, for example: capturing enquiries in a structured way and organising quote preparation.
Scoping the first process properly
Once the candidate is chosen, the scope decides whether it succeeds. Three rules help:
- Start with the standard case. The frequent, uniform cases first. Exceptions still go to a person and are not solved in the first step.
- Define the end. Before starting, it is clear what will work differently after the project and how you will recognise it.
- Clarify responsibility. Someone in the company is responsible for the process, knows it and decides on rules and exceptions. Without this person, even technically good solutions fail.
Typical mistakes when choosing
- Choosing the most annoying process, not the most suitable one. Annoyance is a good pointer, but not a criterion.
- A process without an owner. If nobody can decide how the process should run, nobody can automate it either.
- Shortly before a system change. A solution for a system that will be replaced in a year is rarely a good investment.
- Automating without simplifying first. If you rebuild today’s process one to one, you automate its detours as well.
- Starting without a baseline. Then it can’t be shown after the project what it achieved, and the decision on the second step becomes harder.
The next step
For the two or three most promising candidates, it is worth describing the process properly once, for example with the Process Profile. After that, a short measurement with the “Measure First” tracking sheet helps base the assessment on real figures.
How the principle “automate the standard case, leave the exception with people” can look in a specific case is shown in the practice article on customer enquiries via messenger.
Sources
Describes how automation traditionally concentrates on the most frequent types of case, rare cases are left out because of the cost, and the “long tail” stays with people.
Proposes prioritising automation candidates on the basis of measurable indicators rather than estimates. The authors use process mining for this, i.e. data from large IT systems.
Classic work on human–automation interaction: what is not automated is often the difficult tasks, and people need more practice for them, not less. Written in the context of industrial process control.
Would you like to look at a specific process together? Arrange an initial call