An editorial guide from the Akaen team, based on the departmental analyses we prepare for industrial and service companies.
The question they ask us the most is not what can be automated. Today, almost anything can be automated. The useful question is what is worth automating first: which process pays back sooner, avoids getting stuck halfway through and puts the company in a better position for the next project.
This guide summarises the criteria we apply in a process audit and the twelve most common candidates in companies with 10 to 250 employees. This is a list of processes: the technology is chosen afterwards, once the problem is clear.
Four criteria to choose the first process to automate
A good first process meets all four criteria. With three, it is still a candidate. With two, look for another.
1. Volume and repetition
A process repeated hundreds of times a month with similar steps pays off sooner than a complex one repeated ten times. Automation pays for itself through savings on each repetition, so start by counting: how often each month, how many minutes each time and how many people are involved. If nobody knows the numbers, that is the first week’s task.
2. Clear rules with limited exceptions
The standard path should account for 80% or 90% of cases. If every transaction is different, organise the process first; automating it as it stands only automates the chaos. Exceptions are manageable if they can be clearly defined: a price that differs from the price list, a missing product reference or a quantity that does not meet the required order multiple. What cannot be defined cannot be delegated.
3. Accessible data
Reasonably clean customer and item master data, plus a way into the destination system: API, import or, at worst, the screen interface. Without this, document reading works but data goes nowhere. A closed ERP does not rule out the process, but it changes the effort required, which must be understood beforehand.
4. A visible problem and an owner
Visible delays, errors reaching customers and month-end overtime. The first project needs an owner within the company who wants the problem solved and takes responsibility when the first issue appears. A clear need creates that commitment; without it, the project is abandoned during the first difficult week.
Twelve candidate processes, ordered by what usually comes first
These are the most common candidates when we apply the four criteria in industrial and service companies, in the order they usually appear in an impact–effort matrix:
- Order entry arriving by email, PDF, Excel or portal and typed into the ERP. This is the most common candidate and the easiest to explain. We cover it in detail in order automation.
- Supplier invoices reconciled with delivery notes and purchase orders before posting. A warning sign: administration spends the last days of the month matching documents that arrived without references.
- Delivery notes, labels and shipping documentation generated from the order and batch instead of being filled in by hand at the loading dock.
- Material and quality certificates read, checked against the specification and linked to the batch. A warning sign: quality teams file PDFs that nobody opens again until the audit.
- Work reports and consumption captured on the factory floor instead of being typed up at the end of the shift, so the actual cost of each order is known in time.
- Requests for quotations from suppliers generated from the bill of materials, with responses compared in one table using consistent units.
- Sales quotations prepared from the drawing, data sheet or customer request, with missing confirmations flagged before sending.
- Closing reports and dashboards that are currently assembled manually by combining exports from three systems.
- Repeated internal questions about procedures, technical data sheets or agreements, answered by an assistant that shows the source.
- Employee and customer onboarding documentation: the same file, rebuilt from scratch every time.
- Complaints and non-conformities with case records and probable causes prepared from earlier cases.
- Following up on unanswered quotations and cross-selling opportunities within the customer base, currently dependent on each salesperson’s memory.
How to rank them: impact, effort and prerequisites
For each candidate, assess three things: what it saves (hours, errors, lead time), what it costs (building, integration, maintenance) and what it needs first (clean master data, an API or a procedure change). A simplified example:
| Process | Impact | Effort | Prerequisite |
|---|---|---|---|
| Order entry | High: hours and errors reaching customers | Medium: reading, validation and integration with ERP | Up-to-date customer and item master data |
| Supplier invoices | Medium-high: month-end close | Medium | Purchase orders registered in the ERP |
| Closing reports | Medium: management and administration hours | Low | Access to data sources |
| Internal knowledge assistant | Medium: small amounts of time across the whole team | Low-medium | Documentation in an orderly repository |
The top-left candidate—high impact, low effort and no outstanding prerequisites—is the first project. A candidate with a prerequisite is not discarded: that prerequisite becomes a dated preparatory task. This is an impact-and-effort matrix with prerequisites, delivered by the process audit.
What tends to fail in the first project
- Dirty master data. AI reads the document correctly but has no reliable data to validate it against. Cleaning master data belongs in the first project, rather than becoming an endless prerequisite that postpones it.
- Starting with the exception. Designing for unusual cases before the normal case works. The correct order is the happy path first, the exception queue afterwards.
- Following trends. A website chatbot when the real problem is in administration. Technology follows the process, rather than leading it.
- No owner. Without an owner responsible for the process, automation is abandoned at the first incident.
- No baseline. Without measuring minutes per unit and errors per month before starting, there is no way to demonstrate the return, and the second project will not be approved.
What to Expect from the First Pilot
A well-designed pilot is deliberately small: one process, one department and, if customers or suppliers are involved, those with the highest volume. It runs for weeks rather than months and tracks three figures: time per unit, the percentage of cases needing human intervention and errors reaching customers or suppliers.
The pilot ends when the happy path runs unattended and every exception has an owner and deadline. Then expand to more customers, more formats and the next process in the matrix. If the three measures do not improve after a few weeks, the honest response is to stop and return to the matrix.
Frequently Asked Questions
Do we need to clean the data before starting?
Only enough for the chosen process, not the whole company. If the first project is order entry, you need customer and item master data for the pilot’s customers. Other data is cleaned when its process is addressed.
RPA, integration or AI?
They are tools for different parts of the same process. Integration connects systems that already have structured data. AI interprets documents and language that are not. RPA mimics clicks when there is no other way in. Almost every first project combines two of the three.
What volume is needed to make it worthwhile?
There is no universal threshold. As a rule of thumb, if a process takes more than one person-day a week or causes errors that reach customers, it is worth calculating the impact using real figures.
Can we start if the ERP has no API?
Yes. Almost all ERPs support file imports. If even that is unavailable, a robot can enter validated data through the screen. It is less elegant and must be identified in the first week because it changes the project effort.
To first see how this could work in your company, using no internal data and with no obligation, request the free first look: use cases by department based on your website and sector, within 72 hours. When you are ready to decide using your own data, the process audit turns this list into your own matrix.