How to use AI in your company: start with the one job that eats the most time

Many business owners ask the same question: "How do I use AI in my company?" It usually comes after trying ChatGPT, being impressed, and then not knowing where to start at the office. Our most common answer: don't start with AI. Start with the one job that takes up the most of your team's time every week.

Next stepHow we work: start with one team, first system running in 2–6 weeksSee how it works

Why not start with an "AI project"

Projects that begin with "we need to use AI" almost always end as a presentation, not a tool anyone uses. The reason is simple: nobody knows what is being fixed, so nobody can measure whether it worked.

Projects that work usually start with a very specific complaint:

• "Admin staff are doing the daily sales recap until nine at night."
• "Salesmen write their visit reports by hand, and nobody reads them."
• "Every time the exchange rate rises, the price list takes three days to recalculate."

Complaints like these have the three things AI needs: the work repeats, the data already exists, and the result can be measured in hours or rupiah.

Step 1: list the jobs that repeat every week

Ask each department head to write down the five jobs that take the most of their team's time. Not the hardest jobs, the repetitive ones. In a distribution company the list is almost always similar:

• sales recaps and daily reports from Excel
• salesman visit reports
• working out orders to the factory
• recalculating the price list
• reminding shops that pay late
• answering stock and price questions from shops on WhatsApp

Also write down roughly how many hours a week each one takes, and who does it. Rough numbers are enough.

Step 2: pick one job with three questions

From that list, pick one job, not three. Use these three questions:

1. Does the data already exist? Sales, stock and receivables data usually sit in the accounting system or ERP, even if messy. Work whose data lives only in people's heads is harder to start with.
2. Can the result be measured? "Admin hours dropped from 15 to 2 a week" can be measured. "The team is more productive" can't.
3. Do the people doing it want help? Start with the job your team hates most, not one that makes them feel threatened.

The job that passes all three questions is your starting point.

Step 3: tidy the data enough, not perfectly

A common mistake is waiting for perfect data first. Data is never perfect. You only need the data for that one job, in a consistent form: the same name for the same shop, the same item code in every branch, correct dates.

For shop analysis, for example, 12 to 18 months of order history per shop is enough. We explain how to read it in five signs your customer shops could actually buy more.

Step 4: put AI inside the tools people use every day

AI is most useful not as a chatbot you ask now and then, but as part of daily work tools. For example:

• Salesmen record a voice summary of each visit on their phone, and AI turns it into a clean report that can be searched.
• Every night, an AI Agent reads the order data of every shop and prepares the list of shops to visit tomorrow, with the items worth offering. We explain how this works in AI for sales analysis per shop.
• When a factory raises prices, the app recalculates every price tier in minutes. The manual steps are in how to recalculate your price list when the exchange rate rises.

The key is that AI's output shows up where people already work, such as the sales app or WhatsApp, not in a new system they have to open specially.

A real example. At one distributor we work with, sales visit reports used to be handwritten and never analysed. Once they were replaced with voice recordings summarised by AI, every salesman was using it within a week, because it was faster than writing.

Step 5: measure the result within 30 days

Before you start, write down the starting numbers: how many hours, how many people, how many mistakes. After 30 days, compare. If the numbers dropped, continue. If not, find out why before adding another job.

The most useful measures are usually:

• hours per week spent on that job
• how fast the result is available, for example tomorrow morning instead of next week
• how many mistakes are found
• for sales work: extra rupiah from the shops that were recommended

Step 6: connect the next job

Once one job is running, the second is much easier, because the data is already tidy and the team already trusts it. This is where AI's biggest value appears: when the sales, stock, pricing and receivables apps read the same data, a change in one place is visible everywhere else. That is the core of transforming your company with AI, and it happens step by step, one job at a time.

Is ChatGPT alone enough?

For personal work like writing emails or summarising documents, ChatGPT or Claude already help a lot. For company work, there are three limits:

• A general AI can't read your sales, stock and receivables data directly.
• Its results aren't stored in your systems, so everyone repeats the same work.
• Company data pasted into a public chat needs rules: who may enter which data.

That's why, after early experiments, most companies need AI connected to their own data. That can be an extra module in the ERP you already have, or an app built for you. We compare both in ERP or a custom app for distributors.

In short

• Start with one job that repeats, has its data already, and can be measured.
• Tidy the data for that job only.
• Put AI's output in tools your team already uses every day.
• Measure within 30 days, then add the second job.

If you're not sure which job is worth starting with, tell us about it in a free consultation. We'll help you choose one starting point using your own company's data.