Automate before you talk about AI

AI is on the agenda of many 2026 business plans. Plenty of CEOs want to "do something with AI" this year without knowing where to begin. My advice may sound unfashionable: automate first, and talk about AI afterwards.

This is not a matter of principle. I founded Oliverlist in 2016 without outside investors, and today we manage more than 100 people day to day. Our real jump in productivity did not come from AI. Between 2019 and 2023, automation tripled our revenue per person. That was before generative AI.

What automation did at Oliverlist

Oliverlist books sales meetings on behalf of its clients. The work is a long chain of repetitive steps: importing contact lists, preparing outreach, tracking replies, logging each meeting, invoicing.

From 2020, we automated that chain end to end, from contact import to invoice. Each step reads the output of the step before. Information moves from one system to the next without anyone typing it in again.

The result shows in a single ratio: revenue per person tripled between 2019 and 2023. Put simply, each person on the team carried three times as much business as in 2019.

Our first call to an AI model inside the product came in May 2023. The tripling came from automation, not from generative AI: simple rules, applied every day, without exception.

Predictable work comes first

A task is predictable when you can write down its rule. "Ten days after the due date, send the first payment reminder." "Sixty days before a trade licence or an employee visa expires, open a renewal task." "When a quotation is approved, create the customer in the accounting system."

For this kind of work, automation beats AI on three counts.

• It is consistent. The same input always produces the same output. An AI model can answer two identical requests differently.
• It is traceable. When something goes wrong, you know which rule to fix.
• It is cheap to run. Once in place, a rule can run thousands of times at negligible cost.

There is a fourth benefit, less visible. Automating forces you to map the process: who does what, in which system, with which data. That clarity pays off later, AI included.

A simple test: if you can write the rule on one sheet of paper, automate it. If the only honest answer is "it depends", that is where AI starts to earn its place.

Where AI takes over

Rules break down as soon as a task requires reading, understanding or writing. That is where AI brings something new. I group its uses into three families.

• Sorting. To a keyword filter, "not now, call me after the summer" and "please take me off your list" look alike. To you, one is an opportunity and the other is not. Reading an email, a WhatsApp message or an attachment, whatever the language, then putting it in the right box: that is sorting, and AI does it well.
• Drafting. A first reply, meeting minutes, a call script tailored to one buyer: AI writes, a person reviews.
• Preparing a decision. Gathering the facts, comparing them, setting out two or three options with the reasoning. The decision stays yours; the hours spent preparing it shrink.

At Oliverlist, AI entered the product in May 2023. By 2025 it was in production on three tasks of this kind: sorting prospect replies, writing call scripts and coaching the team. None of them can be written as a rule.

All three rely on the chain we automated from 2020. A sorted reply is only useful if it is already attached to the right contact and the right client.

Order matters. Dropped onto a messy process, AI speeds up the mess. Placed on top of an automated chain, it takes over part of what was left to people: reading, sorting, writing.

Where to start in your company

Here is the order I would follow in a growing business, whether it is a single trading company or a family group with several entities.

1. List the recurring tasks. Yours and your teams': who does what, how often, how long it takes. A week of observation is often enough.
2. Put each task in one of three columns. A rule: automate it. A judgement on a text or a document: AI prepares, someone approves. A real decision: it stays with you or a manager.
3. Automate the first column. Connect your systems so nothing is keyed in twice, including the spreadsheets and WhatsApp groups where orders and approvals often live. It is rarely spectacular. Yet that is where Oliverlist made its big jump.
4. Add AI to the second column. One task at a time, with human approval at first. Track the errors, fix them, then lighten the approval once the results are reliable.
5. Keep the third column. Setting a price, hiring, turning down a client: AI can prepare these decisions. It does not make them.

One trap to avoid: opening with a showcase AI project, such as a chatbot that answers everything, while your team still types the same data into three systems.

Key takeaways

• At Oliverlist, revenue per person tripled between 2019 and 2023 thanks to automation, before generative AI.
• Automate what follows a rule first: it is consistent, traceable and cheap to run.
• Keep AI for what rules cannot do: sorting, drafting, preparing a decision.
• Put AI on top of an automated chain, never on a messy process.
• The decision stays with you.