Which processes should be automated? An evaluation matrix for mid-sized businesses

You should first automate processes that occur frequently, follow clear rules, take noticeable time, are prone to errors, and work with digital data. The more of these five criteria a process meets, the faster automation pays off. Processes that require a lot of judgment, negotiation, or empathy belong at the end of the list.

Step 1: Collect candidates

In each area, ask: Which task is annoying because it repeats constantly? Where is data being typed in, copied, or transferred between programs? Typical candidates are invoice intake, quote creation, appointment coordination, order confirmations, CRM data maintenance, and forwarding inquiries.

Step 2: Evaluate using five criteria

Rate each candidate per criterion from 1 (low) to 3 (high):

• Frequency: How often does the process run per week?
• Regularity: Does it follow fixed if-then rules, or is discretion required each time?
• Time required:How long does a cycle take, including rework?
• Error risk:How often do typing or transfer errors occur, and what do they cost?
• Data situation:Are the input data available digitally, and do the systems have interfaces?

The total results in a benefit value. Then subtract the estimated implementation effort (also 1–3). Processes with high benefit and low effort are your first projects.

An example

Suppose a company evaluates three candidates. The Invoice receipt – daily, clear rules, a lot of manual retyping, digital PDFs – achieves high values with manageable effort. Quotation creation is frequent, but heavily shaped by individual cases and lands in the middle. Annual planning is rare and based on discretion – not a candidate for automation. The example is illustrative; your values depend on your processes.

Where AI shifts the boundaries

Traditional automation required structured data. With AI, it is also possible to process unstructured inputs: understand and classify emails, extract invoices and delivery notes, and handle calls. This brings processes into scope that were previously ruled out because of “too much free text.” When rule-based automation is appropriate and when an AI agent is the better fit is explained in AI agents vs. RPA.

Typical mistakes when selecting

• The most complicated process first: The biggest problem is rarely the best pilot project.
• Automating chaos: If a process is already unclear manually, simplify it first, then automate it.
• Without involving the people concerned:Those who do the process every day know the exceptions.
• Without measurement: Without a baseline, no one can say whether the project was worth it.

If you prefer to do the assessment together: Our process automation starts with exactly this analysis. The Consulting before an automation project is eligible for funding: The BAFA consulting grant subsidizes consulting costs up to €3,500 by 50% or 80% (new federal states, Lüneburg and Trier regions) – for applications submitted by 31/12/2026. The INQA coaching funds 80% of a longer, supported change process nationwide. Which path is suitable is shown by the Funding Check.

The matrix as a template

• Columns: Process · Frequency · Regularity · Time required · Error risk · Data availability · Total benefit · Effort · Priority
• Values:each 1 (low) to 3 (high)
• Priority A: Benefit ≥ 12 and effort 1–2
• Priority B: Benefit ≥ 10 or effort 3 with very high benefit
• Priority C:everything else – later or not at all

The thresholds are a suggestion; adapt them to your situation. What matters is that all candidates are evaluated according to the same rules.

Typical candidates by area

Sales and service

Capture and assign inquiries, prepare quotes, confirm appointments, provide status updates by phone or chat.

Administration and accounting

Invoice receipt, document filing, dunning, data transfer between programs.

Purchasing and inventory

Orders, matching order confirmations, reorder levels, supplier communication.

Human resources

Incoming applications, onboarding checklists, vacation and absence notifications.

Who should be involved in the assessment

At least one person who does the process themselves every day, and one who can decide. The former knows the exceptions where automations fail; the latter ensures that the result is implemented. For processes involving personal data, data protection responsibility also belongs at the table.

From practice

Two examples from our own work: For WINGS, the distance learning provider of Wismar University of Applied Sciences, manual processing of application documents was replaced by a digital application assistant with automatic data transfer. For the integration support service of Caritas , documentation, scheduling, and the management of paper and Excel were converted to digital workflows – together with the people who do the work every day.

Three examples from the assessment

High-priority candidate: inquiries from the website form

Several incoming requests per day, clear fields, manual transfer into the CRM, occasional typos, digital data: high benefit with low effort. A typical first project.

Medium-priority candidate: preparing quotations

Frequent and time-consuming, but with many special cases. A good approach is partial automation: the draft is created automatically, while the review remains with a person.

Low-priority candidate: annual employee reviews

Rare, individual, discretionary. Here, at most, support with scheduling is worthwhile, not automation of the content.

Data protection and security in automations

Automations move data between systems – often personal data as well. That is why a few basic rules belong to every project:

• Processing location: Operate workflows and AI steps on servers in Germany or the EU wherever possible.
• Contracts: For every service involved, a data processing agreement.
• Minimal rights: Each connection receives only the access it needs.
• Logs: Every run is traceably logged.
• Documentation: What the workflow does is documented – including for the record of processing activities.

Frequently Asked Questions

Which processes are best suited for automation?

Processes that occur frequently, follow clear rules, take up a lot of time, and work with digital data – for example invoice receipt, data transfer between systems, or scheduling coordination.

Which processes should not be automated?

Rare workflows, tasks that require a lot of discretion, negotiation or empathy, and processes that are already unclear when done manually. The latter should be simplified first.

How many processes should you start with?

With one. A well-chosen first process quickly shows results, creates acceptance within the team, and provides experience for the next steps.

How do I recognize whether a process is ready for automation?

If it is already carried out the same way by different people, the input data is available digitally, and the rules can be written down. If a process cannot be described, it is not ready yet.

As of October 2026.