RPA or AI? An honest choice for your process

Software robots (RPA) and AI are often thrown together in pitches. For practice that is confusing, because they are fundamentally different solutions for very different problems.

The short difference

An RPA robot does exactly what you write down in step-by-step instructions. Click here, copy that field, paste it elsewhere, save. Fast, error-free, predictable. Fits a process where the rules are clear and almost never change.

AI makes decisions based on patterns it learns from data. It answers questions, classifies content, recognises exceptions. Indispensable when nuance or interpretation is needed, but never a hundred percent predictable.

In practice they work best together: AI decides what should happen, RPA carries it out.

When you pick RPA

Three traits of a good RPA process:

• Rule-based. "For an invoice from supplier X, always link cost centre Y." No grey area.
• Frequent enough. Ten times a month is borderline, daily or more is a no-brainer.
• Between systems that share no API. That old accounting package reachable only through its own interface is RPA paradise.

Concrete examples from projects I ran: invoices from ERP transferred to accounting software (six hours per week saved). Customer data synced between CRM and webshop (two hours per day). Weekly revenue reports compiled from three different dashboards (a day and a half per month).

When you pick AI

AI is the right choice when there is an element of interpretation or context in the work:

• Answering customer questions based on manuals and old tickets.
• Classifying incoming emails by department (sales, support, finance).
• Summarising long documents for management.
• Generating draft text for quotes or newsletters.
• Pulling sentiment from feedback (complaint or compliment?).

The win with AI is often not strict time savings per task, but things that would not have happened without AI: every customer email now gets a targeted first reply within seconds, including evenings and weekends.

Hybrid works best

In most of my projects RPA and AI run together. A typical example: a customer sends an email with a question. AI reads the email, classifies the intent ("invoice request"), pulls the right invoice from your own database via an RPA step, and drafts a reply. The employee checks and clicks send.

Result: three minutes of work instead of fifteen, and nothing gets stuck because someone is on holiday.

The question to start with

Not: "Where can I use AI?" But: "Which process frustrates my team the most, and are the rules clear enough to write out?"

If the answer is clear: RPA. If the answer is vague: rethink the process first, then consider AI.

Also read

• What AI really means for your SMB - practical AI applications without the hype
• How half a second of load time costs you 7% revenue - website speed and conversion, step by step

Want to test this against your own processes? A coffee chat of thirty minutes often gives direction surprisingly quickly. No obligation, no sales pitch.