Is your process ready for AI?

When is a process suitable for AI?

This scan tells you in about three minutes whether one specific process is a good candidate for AI, for rule-based automation, or for neither yet. A process suits AI when it happens often, takes noticeable time and revolves around information you have to read or interpret: sorting emails, extracting data from PDFs, answering customer questions or summarising documents. That is where AI is strong, because it copes with varying wording and layouts where fixed rules break down.

At least as important is what happens when it goes wrong. AI makes mistakes, and a wrong answer can sound very convincing. A process in which an employee can check the result quickly, or in which a mistake does little harm, is therefore a better starting point than one in which every decision goes straight through without a check.

The third condition is data. AI needs access to the information and somewhere to put the result. If the information is spread across mailboxes and Excel files, or sits in a system without an integration, most of the time goes into collecting and returning data rather than into the clever part.

How the AI scan works

Pick one concrete process, such as "processing incoming invoices" or "answering quote requests", and answer ten multiple-choice questions. The more specific the process, the more useful the result. A whole department or "our administration" is too broad to assess meaningfully.

The questions fall into four areas: volume and time saved, suitability of the input, risk and control, and data and integrations. For each area you get a score from 0 to 100 with a concrete next step. The scan also looks at whether the work really needs interpretation, or whether fixed rules will do.

The main advice is one of four outcomes: AI is a good first step, start with rule-based automation, get your data and integrations in order first, or do not automate (yet). That last outcome is valuable too: it stops you investing in something that will not pay for itself.

AI or rule-based automation?

Not everything that can be automated needs AI. If the input arrives as fixed fields from a form or export, and you can write the decision down as "if this, then that", a workflow with fixed rules is cheaper, faster and more predictable. You know exactly what it does and why.

AI becomes interesting as soon as text, documents with a varying layout or judgement calls are involved. A common approach is a combination: fixed rules do the predictable work, AI extracts information from unstructured input or makes a proposal, and an employee reviews the doubtful cases.

Common mistakes in a first AI project

The most common mistake is starting too big. A project that has to take over an entire process in one go often runs aground on exceptions nobody foresaw. Start with one step, have a person check the output, and only expand once you know how often the result is right.

A second mistake is not measuring. Record beforehand how much time the process takes now and how often errors occur. Without that baseline you cannot say afterwards whether the solution delivers anything. Also keep a set of real examples, including difficult cases, to test every new version against.

Finally, privacy is often looked at too late. If personal data or confidential information is involved, find out beforehand where the data is processed, what has been agreed with the vendor and whether you need a data processing agreement.