MIXEDAutomating business processes with AI: where to start
"We want to automate" is a good intention and a terrible starting point. Projects succeed when you start from a specific process, measure it and accept changing it. Here's the method we use with our clients, and the three mistakes we see most often.
Classic automation or AI? It depends on the input
Many processes don't need artificial intelligence. If the rules are fixed — "when an order arrives, create the job and send the confirmation" — classic automation or a well-configured management system is enough. AI is needed when the input is unstructured: emails written in different ways, supplier PDFs in different formats, photos, natural-language requests. Or when there are so many rules nobody can write them down.
| Process | Right tool |
|---|---|
| Attendance from clock-ins → payroll report | Rule-based management system (no AI) |
| Deadlines → automatic alerts | Rule-based management system |
| Incoming emails → classification and draft reply | AI (language) |
| Supplier PDF invoices → rows in the management system | AI (documents) |
| Historical orders → purchasing forecast | AI (predictive models) |
The 5-step method
1. List the processes that "eat hours"
Ask every department: "what do you do every day that you could do with your eyes closed?". Copying data, answering the same questions, checking deadlines, preparing the same report. Next to each one write the hours per week. That's your list of candidates.
2. Pick just one, using four criteria
- Hours: enough to justify the project (roughly 5 hours a week or more).
- Recognisable rules: whoever does it can explain how they decide.
- Data already digital: emails, PDFs, management system. If it's on paper, digitise first.
- An owner: one person who does it today and will validate the result.
3. Measure before
Hours spent, errors, response times: without a "before", you'll never know whether it worked. Two weeks of notes are enough.
4. One-month pilot, with a human in the loop
The automation proposes, the person confirms. It's slower than "fully automatic" but it eliminates the risk of silent errors and gets the users to accept the tool. Once accuracy is stable, you reduce the checks.
5. Measure after, then expand
Compare with the "before". If hours have gone down, move on to the second process on the list. If not, understand why before expanding.
What we learned replacing Excel sheets. In several companies with construction sites, attendance, deadlines, vehicles and costs lived in separate Excel files: one for the office, one for each site manager, updated by hand and often out of sync. The first process we replaced with V-Site was almost always the same: attendance. Workers clock in with a QR code, hours arrive ready for payroll, nobody retypes anything.
From there, deadlines and costs followed naturally, because the data was finally in one place. The benefit clients report is hours recovered every week by admin, site managers and workers. No artificial intelligence in this first step: just a well-chosen process and data put in order. AI comes later, on data that is now reliable.
The three most common mistakes
- Automating a process that doesn't work. If the flow is confused today, automation just makes it confused faster. Simplify first, then automate.
- Starting from the tool instead of the problem. "We bought a licence for X, what do we do with it?" is the premise of many abandoned projects.
- Leaving out the people who do the work. The person who runs the process today knows the exceptions no document mentions. If they aren't involved, the automation will get exactly those wrong.
Frequently asked questions
Do you need AI to automate a process?
Not always: with fixed rules a well-configured management system is enough. AI is needed for unstructured input or too many rules.
Which process should you automate first?
The one with the most repetitive hours, recognisable rules, digital data and a clear owner.
How long does it take to see results?
A 3-6 week pilot on a well-chosen process is enough to measure the hours saved.
Already know which process eats most of your hours?
Tell us about it in 30 minutes: we'll tell you whether you need AI or just a management system, and what it costs.