How to Choose Processes Suitable for Automation: Prioritisation in a Company

Companies that decide to automate usually have no trouble putting together a list of candidates. Invoicing, onboarding, order approvals, reporting – there's no shortage of lists. The problem appears a step later: how do you pick, out of ten equally "logical" candidates, the one worth tackling first? This is where most bad decisions happen – companies automate the process that's most visible, or the one someone complained loudest about, rather than the one with the genuinely highest ratio of benefit to effort.

This article doesn't cover where to start with automation in the first place – we've already looked at that in Automating business processes: where to start and what to prioritise, which we recommend as a preceding step if you haven't yet mapped out which processes in your company are even candidates. Here we focus purely on what comes next: how to compare processes against each other objectively – which criteria to score, how to build a scoring matrix, and what to do when you end up with a tie.

Without a structured comparison, the choice of which processes to automate is almost always driven by who shouts loudest, or by what's "visible". The department that complains the most gets attention before the department that quietly loses dozens of hours to repetitive work but never raises it.

The second common problem is comparing things that aren't comparable. A process that happens once a month and takes three hours can't be judged on the same scale as a process that happens a hundred times a day and takes three minutes – both might be "time-consuming", but their real impact on the business is completely different. Without a consistent set of criteria you end up comparing apples with oranges, and the decision is ultimately made on gut feeling rather than data.

The basis of automation prioritisation is always the same principle: the ratio of impact to effort. To make that ratio measurable, it needs to be broken down into concrete, scoreable criteria.

Business impact

• Frequency – how often the process runs (daily, weekly, monthly, seasonally).
• Time burden – how many person-hours the process consumes over a given period, not just "how long one run takes".
• Error rate – how often the process fails or needs correcting, and what the consequences of an error are (an internal nuisance versus impact on a customer or supplier).
• Number of people and departments involved – a process that crosses three departments usually has more improvement potential than an isolated task done by one person.

Automation effort

• Data structure – does the process take in a structured document (e.g. an invoice in a consistent format), or free text and various attachments?
• Number of exceptions – how many variants and special cases the process has. A process with dozens of exceptions is harder to automate than one with clear rules.
• Dependency on other systems – does automating it require linking several systems via an API, or does it run within a single, self-contained tool?

Risk and sensitivity

• Type of data processed – personal data, financial data or contractual obligations require more caution and more thorough testing than internal reporting.
• Consequences of an error after automation – what happens if the automated step fails without human oversight. For some processes that's an inconvenience; for others it's financial or legal damage.

1. List the candidates. Use the list of processes you identified in the earlier prioritisation step.
2. Choose 5 to 8 criteria. More criteria won't improve the matrix – they'll just slow it down and dilute the differences between processes. Combine impact criteria (frequency, time burden, error rate) with effort criteria (data structure, number of exceptions, dependencies).
3. Set a consistent scale. A 1–5 scoring scale is most common, where 1 means low impact or low effort and 5 means high. It's important that the same person or team applies the scale across all processes – otherwise subjective bias creeps into the scoring.
4. Assign weights to the criteria. Not all criteria carry the same importance. If compliance is critical for the business, risk might carry more weight than time savings.
5. Calculate the score. The sum or weighted average of the points for each criterion produces the final ranking of processes.

It's worth visualising the resulting scores – even a simple bar chart helps the team quickly see which processes stand out from the rest.

This is an illustrative example of the approach, not real data – actual scores will differ for every company depending on which processes and which criteria weights are chosen.

In practice, two or three processes often end up with almost identical scores. In that case, the scoring matrix alone doesn't give a clear-cut answer, and you need additional criteria to decide:

• Data readiness right now. If one process already has digitised data and the other still needs cleaning and consolidating from various sources, the first is quicker to deliver regardless of an identical impact score.
• Process ownership. A process with a clear owner willing to help define the rules moves along more smoothly than a process with no responsible person, even if it looks equally attractive on paper.
• Independence from other changes. If one of the processes is going to change soon anyway (for example, because of a planned system replacement), it makes sense to wait for that change to happen and instead move forward with a process that's stable.
• Visibility of a quick win. In a tie, it can make sense to prioritise an RPA-type process, which is simpler to deliver and serves as a proof of concept for the rest of the organisation while the rollout of a more demanding process is still being prepared.

When scores are tied, the deciding factor isn't another round of scoring – it's which process is genuinely ready to go first.

If you're unsure which of two similarly scored processes has the better real-world prospects, a short internal workshop with the owners of both processes helps – it often surfaces information that never made it into the scoring table (an upcoming legislative change, say, or an internal project that will affect the process regardless).

A scoring matrix produces a ranking, not an automatic decision. Before you start implementing the process with the highest score, it's worth checking two things: whether the process has a clearly defined owner who can provide rules and feedback, and whether metrics exist that let you compare results after deployment. Establishing a baseline – how much time the process takes today, what its error rate is – needs to happen before automation, otherwise you'll have nothing to compare against.

If one of the candidates is a process related to customer communication or documents, it can help to look at how specific technologies work in practice – for example how an AI agent processes invoices and accounting documents, or when it's worth reaching for a tool like Make or Zapier instead of a custom-built solution. For companies also considering deploying AI agents, it helps to know in advance how the return on such a project is measured – that is, which metrics to track, not what number to expect upfront.

Choosing which processes to automate works best when it combines a structured scoring matrix with a realistic view of what's currently achievable within the company. If you're unsure how to set the criteria or weights for your specific case, or you'd like an outside perspective on which solution – from a simple system integration through AI and automation to custom development – fits the process you've chosen, you can discuss it in a no-obligation consultation via our contact form.