Where to start
Artificial intelligence is on the agenda in many businesses, but a way in is often missing. Between chat tools, marketing promises and concerns about data, the question remains where a first step actually pays off.
The answer rarely lies in large projects. It lies in the processes that run the same way every week, take up a lot of time and are still done by hand. That is where automation delivers the fastest and most verifiable benefit.
How to recognise good first candidates
Not every process is equally suitable. Good candidates usually share these traits:
- They occur frequently, daily or weekly.
- They follow a recognisable pattern, even if the inputs look different.
- A mistake is annoying but not serious, or the result is checked anyway.
- The outcome is easy to verify.
- The data is already digital, for example as an email, PDF or photo.
The more of these points apply, the better the process works as a starting point.
Extracting data from documents
In many businesses, receipts, reports, delivery notes or records are typed up by hand. Someone reads a document and transfers the details into another system. This is exactly what AI now handles well: it recognises the relevant details and stores them as clean data in the right system.
People stay involved. They review the cases where the system is uncertain instead of capturing every document themselves. The work shifts from typing to checking.
Pre-sorting enquiries
A shared inbox, a contact form and enquiries arriving through several channels: often someone spends time every morning reading, sorting and forwarding messages. AI can classify incoming emails and forms, answer standard questions and pass the rest to the right person.
What matters is a clear rule on which replies may go out automatically and which are read first. At the start, it is advisable to prepare replies only as drafts and to release them gradually after a period of review.
Routine work with approval
Many processes consist of several steps: checking data, creating a record, sending a message, assigning a task. Such chains can be automated with so-called agents that work through the steps one after another.
A proven principle is the threshold: cases below a defined limit run automatically, anything above goes to a person for approval. The decision stays where it belongs, and the routine still gets done.
Working with your existing systems
Automation does not mean replacing your software. On the contrary, the greatest benefit usually comes when AI works with the systems already in use. It reads from the inbox, writes to the inventory and order management system or creates records in your existing software.
So check early which interfaces your current programs offer. That determines how seamlessly an automation can be fitted in.
What should not come first
Less suitable as a starting point are processes that occur rarely, where every case is different or where a mistake would have serious consequences. Decisions with legal or financial implications should also stay with people. AI can prepare such decisions, but it cannot be accountable for them.
Data handling belongs at the start, too. Clarify which data is processed, where that happens and whether personal data is involved.
How to proceed
- Collect recurring processes and roughly estimate how much of your team's time they take.
- Choose a process that occurs often and is easy to check.
- Define what may run automatically and what goes for approval.
- Start small, review the results and adjust the rules.
- Only then extend to the next process.
Step by step, this builds an automation your team trusts, because they have been involved from the very beginning.