Look for repetitive work, not clever work
When AI comes up, many business owners picture a smart assistant that can answer anything. It is an appealing idea, but it is hard to measure and expensive to start. A better starting point is a list of routine tasks that eat up your team's hours without needing difficult decisions.
Some common examples in trading and service companies:
- Entering supplier invoices into the purchasing system.
- Matching bank transactions to invoices that have been paid.
- Answering order-status questions on WhatsApp.
- Summarising daily reports from several branches.
- Sorting e-mails and complaints to the right team.
These tasks share three traits. The input is a document or text, the rules can be explained, and mistakes are easy to check. That is where today's AI is most reliable.
AI works best on top of a tidy system
An AI model can read an invoice in seconds, but the result still has to land in the right place: the right purchase order, the right supplier, the right account. If purchasing data is scattered across spreadsheets, group chats, and notebooks, AI only speeds up the mess.
That is why our order of work is nearly always the same. Tidy up the core data, connect the apps already in use, and only then add AI where it saves the most time. We cover the first two steps in our article on connected systems.
AI does not fix a messy process. It makes a tidy process run faster.
People keep the final say
In the systems we design, AI prepares a draft and a member of staff approves it. An invoice read by AI appears as a filled-in form that only needs checking. A reply to a customer appears as a suggestion that can be edited before sending. Above a set amount, approvals still go up the usual chain.
Every step is logged: who approved, when, and what was changed from the AI suggestion. The log helps with audits and shows how often the AI gets it right.
Measure before and after
Before you start, record three numbers for the chosen process: average time per document, errors per month, and how long customers or managers wait. Run the new system alongside the old way for a few weeks, then compare the same numbers.
Those numbers answer the question that matters most to an owner: is the cost of the system worth the hours it gives back to the team?
A realistic first step
- Pick the one process that is repeated most often and complained about most.
- Collect 50 to 100 sample documents or conversations from it.
- Set the rules: when AI may act directly and when it must wait for approval.
- Run it alongside the old way, measure the results, then move on to the next process.
Decide early where the data is processed and stored, and whether the model provider may use it for training. Answer these questions before the first document is uploaded.
If one task in your office feels like the most boring of all, it is a good place to start. Tell us about it, and we will help you judge whether AI is really the answer.