Many companies feel pressure to start using AI. Some employees are already using ChatGPT or Claude on personal accounts, while management has yet to set a direction. The result is AI use with no rules, results nobody can measure, and company data ending up in places it should not be.
AI adoption in a company does not have to start with a big project. The safest route is to pick one process, run a small pilot, put a policy in place, train the team, and measure the results. This article walks through each step with examples that reflect how businesses in Indonesia actually work.
Choosing the first process for AI adoption in your company
The most common mistake is starting with an idea that is too broad, such as "an AI assistant for every department". Projects like this are hard to judge as a success or failure, and they usually stall halfway. A narrow process that the team handles often is a much better starting point.
A good first process usually ticks most of these boxes:
- It is done repeatedly, every day or every week.
- It is based on text or documents, such as emails, forms, invoices, or reports.
- The people who normally do the work can easily tell whether the output is right.
- Small mistakes do not immediately hurt customers or finances.
- There are examples of good work to use as a reference.
Take a distributor that receives purchase orders by email every day and retypes them into its system. Or a notary office where admin staff summarise client documents before the notary reviews them. A services company might start with draft replies to customer emails. All three share the same pattern: routine work, lots of text, and a person who still checks the result.
Running a pilot you can measure
A pilot is a limited trial with a clear timeframe, a defined group of people, and agreed measures. A few weeks is usually enough to see whether AI really helps.
Choose three to five people who do the process every day. They know best where AI helps and where it creates extra work. Before you begin, record the starting point, for example how long one document takes and how often typing errors occur. Without that baseline you will have nothing to compare against at the end.
During the pilot, every AI output is checked before it is used. Ask participants to note the cases where the AI was wrong, unsure, or of no help. Notes like these are often worth more than a general feeling that the AI is "quite useful".
Writing an AI usage policy
Prepare the policy alongside the pilot instead of waiting for a problem to force it. It does not need to be long. One or two pages are enough if the content is clear.
At a minimum, the policy should state which AI tools are allowed and whether staff must use company accounts. It should list the types of data that must never be entered into AI tools, such as customer personal data, national ID numbers (NIK), payroll data, and confidential client documents. It should also say that AI output must be reviewed before it goes to anyone outside the company, and name who to contact with questions or incidents.
If the process involves personal data, make sure the policy is consistent with your company's privacy policy and the principles of Indonesia's Personal Data Protection Law (UU PDP, Law No. 27 of 2022). Bring in your legal or compliance lead from the start.
Training the team to use AI well
Training is often treated as a single demo session. In practice, people need to practise on their own work. The most effective sessions tend to be small workshops for each team, built around that team's real documents and cases.
Cover how to give clear instructions, how to check AI output, common failure modes such as the model making up information, and what the usage policy says. It also helps to have one person in each team whom colleagues can ask for help. This person does not need a technical background. What matters is that they experiment regularly and are happy to share what they learn.
Pay attention to staff concerns as well. Some people worry that their jobs will be replaced. Be honest about what the company is aiming for, for example cutting down on retyping so the team can spend more time with customers.
Measuring results and deciding what comes next
At the end of the pilot, compare the baseline with the situation after AI was introduced. Useful measures include time per document, number of errors, the share of AI output that could be used with few edits, and what participants thought. Also total up the AI usage cost during the pilot, so your estimate for a larger rollout is grounded in real numbers.
The results usually lead to one of three decisions. If they are good, extend the pilot to more people or connect AI directly to the systems the team uses every day. If they are mixed, improve the instructions or the data sources and run it again. If AI did not help, stop and choose a different process. Ending a pilot that did not work is a sound decision, because the cost at that point is still small.
Support for your AI rollout
Once a pilot works, the next step is usually connecting AI to the applications and data you already use. Averoa helps companies choose the right process, build AI features with guardrails and evaluation, and keep costs measurable through our AI integration service.
