That change is real in one important sense: the prerequisite technologies are now available locally. Salesforce expanded its Indonesia-based Hyperforce offering in July 2025 to include Agentforce, Data Cloud, Tableau Cloud, Tableau Next, and Marketing Cloud Next. For regulated businesses, particularly insurers, banks, telecommunications providers, and public-sector organizations, the local availability removes a practical obstacle that kept some AI projects in demonstration mode.
But “AI FOMO is fading” is a conclusion drawn largely from one vendor’s customer conversations. Salesforce’s own January 2025 survey of 150 Indonesian small and medium-sized business leaders points in the opposite direction: 41% said they feared falling behind without AI, while 77% said they were already using or experimenting with it. The more defensible reading is that FOMO has not disappeared. It has become harder to fund.
Salesforce’s Mandiri InHealth example needs a fuller ledger
Carvouni’s strongest example was Mandiri InHealth, where he said an Agentforce deployment cut the time needed to process letters of guarantee by about 46% over its first three months and improved chat response times by around 30%. If those figures hold in sustained production, that is the kind of narrow, high-volume workflow where AI can produce a credible business case: there is a discrete process, an established baseline, repeatable inputs, and a measurable service outcome.
Yet the figures remain vendor-reported performance claims, not independently audited results. Neither Salesforce nor Mandiri InHealth has publicly detailed the original processing time, the post-deployment time, the number of cases included, the rate of incorrect or escalated decisions, the human-review model, or the cost of licenses, integration, model usage, implementation, and ongoing support.
Those omissions matter more than the percentage reduction. A 46% improvement in a slow, manual workflow may create substantial capacity, but it does not automatically reduce staffing costs or increase profit. An insurer may instead redirect those recovered minutes into exception handling, customer follow-up, fraud review, or greater claim volume. All can be valuable outcomes; none is the same thing as a demonstrated P&L gain.
The same issue applies to Carvouni’s Philippines insurance example, in which a letter-of-authorization process allegedly fell from about 20 minutes to 30 seconds for 20,000 to 40,000 monthly requests. The magnitude is compelling, but it came through the Salesforce executive rather than a named customer’s public case study. Until the company identifies the insurer and publishes production-quality metrics, it should be treated as an illustration of what Salesforce says is possible, not a regional benchmark that other IT leaders can budget against.
For CIOs and operations teams, the practical lesson is simple: insist on a benefits ledger before approving an AI scale-up. It should separate time saved from cash savings, track quality and rework, count exceptions and human escalations, and include the recurring cost of the AI platform and the data infrastructure feeding it. A pilot that produces faster responses but requires constant data cleanup or specialist intervention has not escaped “pilot purgatory”; it has simply made the pilot more expensive.
Worker adoption is higher than daily production use
The PwC Global Workforce Hopes and Fears Survey 2025 gives the Jakarta Globe report useful context, but it measures worker experience rather than enterprise financial performance. PwC surveyed 49,843 workers globally, including 812 in Indonesia, between July and August 2025. It found that 69% of Indonesian respondents had used AI in their work during the prior 12 months, compared with 54% globally.
Daily use was far lower: 16% of Indonesian respondents said they used generative AI daily, only slightly above the 14% global figure. That gap between “used AI this year” and “uses it daily” is the more revealing number. It suggests that experimentation has spread broadly, while routine integration into work remains limited.
PwC also found that 96% of Indonesia’s daily generative-AI users reported improved productivity, against 75% of infrequent users. That is a meaningful correlation, but it should not be recast as proof that daily use caused the improvement or that employers captured the value. Workers who already have better tools, stronger management support, higher digital skills, or more adaptable job designs may be both more likely to use AI every day and more likely to report gains.
The survey’s other results underscore the risk of management oversimplifying the numbers. Daily users in Indonesia were more likely than infrequent users to report better job security and salary improvement. Those are self-reported perceptions, not payroll data or workforce-reduction figures. They support the case for training and work redesign; they do not establish that an enterprise AI rollout has paid for itself.
This is where executive rhetoric about “raw intelligence” meets the unglamorous reality of implementation. Large language models can draft, classify, summarize, retrieve, and route information. They do not arrive with clean customer records, authoritative policy documents, sensible permissions, a defined escalation path, or accountability for decisions that affect a customer’s insurance coverage. Those elements are the actual project.
Local data residency does not settle every data-flow question
Salesforce is right to stress Indonesia’s Hyperforce availability as an advantage for customers that need local data residency. Agentforce and Data Cloud became available on Hyperforce in Indonesia on July 16, 2025, while Salesforce had launched its Indonesian entity and the original Hyperforce availability in August 2023. The 2025 expansion is the material development for AI deployments; it gave local customers access to more of the products that connect customer data to AI workflows.
But local hosting should not be reduced to a blanket assurance that all information remains in Indonesia under all circumstances. Salesforce’s own Hyperforce documentation says customer data for products on Hyperforce is stored at rest in the selected country, while also noting that data can leave the country for processing or support and that some service components or integrations may operate elsewhere. Product availability also varies by service.
That distinction is particularly important for enterprises connecting Agentforce to outside systems, third-party models, analytics platforms, messaging channels, or document repositories. A data-residency decision should therefore be tested at the level of the complete workflow: where data is stored, where it is processed, which subcontractors are involved, how logs are handled, whether prompts or responses are retained, and what information crosses a border during support or integration.
Indonesia’s policy environment adds urgency. The Ministry of Communication and Digital Affairs completed cross-ministry discussions on draft presidential regulations covering AI ethics and the National AI Roadmap for 2026–2029 on May 5, 2026. As of late July, government officials were still describing the rules as awaiting presidential signature. Businesses cannot assume that a framework under final review is already a settled compliance answer.
For Windows administrators and enterprise architects, this points to a familiar principle: the AI interface may be cloud-hosted and conversational, but the governance work remains conventional IT. Identity controls, least privilege, data classification, audit logs, retention rules, endpoint security, integration review, and incident response determine whether an AI service can be operated safely. A chatbot’s apparent simplicity often hides a wider data-access footprint than a traditional line-of-business application.
Indonesia’s talent numbers are inconsistent—and the gap is still large
The story also repeats an Indonesian government estimate that the country needs about 12 million digital talents by 2030 against roughly 3 million available now. That formulation reflects remarks from Deputy Communications and Digital Affairs Minister Nezar Patria in July 2025.
More recent government-linked reporting has used a very different current-supply figure: 9.3 million digital talents, with demand still projected to reach 12 million by 2030. The two figures may be measuring different populations—perhaps available specialist talent versus a broader digitally skilled workforce—but the government has not supplied a common definition that makes them directly comparable.
The discrepancy should not be ignored because it changes the size of the claimed shortage from roughly nine million people to roughly 2.7 million. Either way, the issue for AI projects is less about the headline total than the availability of people who can do the specific work: data engineering, platform administration, cybersecurity, workflow design, Indonesian-language evaluation, model-risk management, and change management.
Carvouni’s assertion that AI will augment workers while displacing some roles is reasonable as a general expectation, but it is not an implementation plan. The immediate operational risk is that companies procure AI licenses faster than they redesign work or train the people supervising automated outputs. That produces shadow usage, inconsistent quality controls, and a support burden transferred to already stretched IT teams.
The bottom-line phase will favor narrow deployments
Salesforce’s message to start with a business outcome rather than technology is sound, even if the company benefits when that outcome leads customers to its platform. The emerging discipline should be to begin with a process that has known volume, known cycle time, measurable error costs, and a clear owner—not with a general mandate to “use AI.”
The most credible Indonesian AI deployments will likely be constrained systems such as claims-document routing, customer-service summarization, knowledge retrieval with reviewed source material, campaign operations, and internal service workflows. Each can be measured against a baseline and monitored for failure. Broad autonomous-agent projects connected to fragmented enterprise data will have a much higher bar, especially in insurance and other regulated sectors.
FOMO has not vanished from Indonesia’s AI market. Salesforce’s own SMB data shows it remains a live pressure point. What is fading is the willingness of senior management to accept “we have an AI pilot” as an outcome. The next funding decision will depend on whether vendors and internal project teams can show the full cost, the error rate, the governance controls, and a repeatable financial result—not simply a faster demo.
References
- Primary source: Jakarta Globe
Published: August 8, 2026 at 2:54 PM UTC
AI FOMO Is Fading as Indonesian Businesses Focus on the Bottom Line
Indonesian businesses are moving beyond AI experimentation as executives demand measurable gains in productivity, revenue and costs.jakartaglobe.id - Related coverage: pwc.com
Adopsi AI mendongkrak produktivitas, terutama bagi Gen Z Indonesia, namun kesenjangan keterampilan antar generasi masih menjadi tantangan
Para pekerja mulai angkat suara soal perdebatan besar tentang bagaimana AI memengaruhi produktivitas, pertumbuhan bisnis, dan dunia kerja. Hasil PwC’s Global Workforce Hopes and Fears Survey 2025 menunjukkan bahwa pengaruh AI makin terasa, dan rasa optimis terhadap manfaatnya jauh lebih kuat...www.pwc.com