Introduction
Almost every company in Indonesia is now talking about artificial intelligence (AI). Yet there is a wide gap between "we have tried AI" and "AI is delivering real business impact". Suara.com's report from BytePlus Indonesia AI Day 2026 cites PwC research: 96% of Indonesian businesses have adopted AI technology, but only 12% report actual, measurable business impact.
The "Unlocking Indonesia's AI Potential 2026" study by Amazon Web Services (AWS) and Strand Partners, presented at AWS Summit Jakarta 2026, paints a similar picture. AI adoption among Indonesian companies rose to 40%, up from 25% the previous year, but 56% of adopters are still in the early exploration stage. Only 12% have fully integrated AI into their business processes.
In other words, many AI projects stall halfway. Below are the five most common mistakes businesses make when starting an AI project, along with practical ways to avoid each one.
| Indicator (AWS & Strand Partners study, 2026) | Figure |
|---|---|
| Companies that have adopted AI | 40% (25% last year) |
| Still in early exploration | 56% |
| AI deployed in some business functions | 18% |
| Scaling AI across multiple functions | 14% |
| Fully integrated into business processes | 12% |
| Reporting higher productivity | 75% (previously 68%) |
1. Starting With the Technology Instead of a Business Problem
The first mistake is asking "which AI can we use?" instead of "which problem costs our business the most?". Teams end up subscribing to a language model or building a chatbot without knowing what success looks like. Anthony Amni, Country Manager of AWS Indonesia, names business strategy readiness and measuring return on investment among the main obstacles organisations face.
How to avoid it:
- Write down one specific problem in a single sentence, such as "customer service response times are too long".
- Set the metrics before the project starts: processing time, cost per transaction, conversion rate.
- Measure the starting point (baseline) so pilot results can be compared against it.
Transjakarta offers a good example. The problem was clear: thousands of user questions on social media during peak hours. With an AI chatbot, customer inquiry handling time fell by 93.82%. That figure can be reported precisely because the problem and its metric were defined up front.
2. Chasing Ambitious Use Cases Before the Data Is Ready
Many businesses want to jump straight to an AI assistant that "knows everything" about the company, while their data is scattered across spreadsheets, email and applications that do not talk to each other. AI is only as good as the data it can reach.
The Volantis Technology case presented at BytePlus AI Day shows a more sensible order: AI was first used for data consolidation and automation, and only later grew into products such as the "Sophia" AI assistant and an expansion into the B2C segment.
How to avoid it:
- Inventory your data sources: where they live, who owns them, how complete they are.
- Pick a first use case whose data is already available and reasonably clean.
- Budget data cleaning and integration as an official part of the project, not a side task.
3. No Project Owner, So the Pilot Ends at the Demo
The fact that more than half of adopters are still exploring points to a familiar pattern: a pilot runs, the demo looks promising, and then the project stalls. The usual cause is that nobody on the business side is accountable for moving the pilot into daily operations. The AWS study also flags AI governance as one of the obstacles.
How to avoid it:
- Appoint a project owner from the business unit that feels the problem, not only from IT.
- Define "pilot passed" criteria and a decision deadline: continue, adjust or stop.
- Plan from day one who will use, maintain and oversee the system after go-live.
4. Budgeting Only for the Build
An AI project is not finished when the application ships. There are running costs: API or model compute usage, data storage, monitoring answer quality, updating prompts or models, and the staff time to manage it all. Businesses that budget only for development are often surprised when the monthly bill grows with the number of users, and then shut down a system that was actually useful. This ties directly to the difficulty of measuring ROI that AWS highlights.
How to avoid it:
- Calculate the cost per transaction or per conversation, then project it for 6–12 months of volume.
- Separate the build budget (one-off) from the operating budget (monthly).
- Compare those running costs with the savings or revenue measured in point one.
5. Not Preparing the Team
Technology can be bought, but the ability to use it has to be built. The AWS study places talent development alongside strategy, ROI and governance as a key obstacle. A skills gap makes employees hesitant to use new tools, or the opposite: using them without understanding their limits.
How to avoid it:
- Map the skills you need: daily users, data stewards and decision-makers.
- Run short training sessions built on real cases from each person's job, not generic theory.
- Involve future users from the pilot stage so they help shape the solution.
Conclusion
The two studies point in the same direction: AI is widely tried in Indonesia, and 75% of adopters report higher productivity, yet only a small share have truly integrated it or seen measurable impact. The difference rarely comes down to which AI model was chosen. What decides it are the fundamentals: a clear problem, ready data, an accountable owner, a realistic operating budget and a trained team. Start small, measure the results, then scale.
