The fraud officer I sat with in Yogyakarta last month had spent eleven years catching counterfeit invoices, suspicious wire transfers, and informal collusion patterns at a regional bank’s branch network. She is exceptionally good at her job. She has also spent the last three weeks trying to learn how to detect a deepfake — and she does not know whether her bank will pay for the training, whether the training even exists in her language at the level she needs, or whether the job will still be hers by the time she finishes.

That conversation, repeated across enough branches and back offices, is the part of the AI upskilling story we are not reading honestly.

We talk about AI talent and AI upskilling as if they are the same wave hitting the same shore. They are not. The corporate upskilling programmes I see across ASEAN — generative AI literacy modules, prompt engineering courses, model evaluation training — are reaching the same demographic in every market: urban, English-speaking, mid-career, mostly male, mostly already inside one of the region’s twenty largest institutions. They are not reaching the eleven-year veteran in Yogyakarta. They are not reaching the back-office compliance officer at a regional multifinance company who has just been told to monitor AI-driven credit decisions she has no training to read.

This is the second governance debt — and it is compounding alongside the first.

The shape of the divide

The actual divide cuts across at least four dimensions, and they reinforce one another.

Geographic. Most AI training in Indonesia happens in Jakarta and a handful of secondary cities. The branch officer in Surabaya might catch the wave. The same role in Manado will not.

Linguistic. The strongest AI literacy materials are still written in English, with second-best versions in Bahasa Indonesia for general audiences — and almost nothing in Bahasa at the technical depth that risk and compliance work actually requires.

Gender. The pipelines into AI roles across ASEAN financial services skew male more sharply than the underlying workforce does. The mid-career women who staff much of the back office — fraud detection, customer due diligence, claims, member services — are simultaneously the most exposed to displacement and the least likely to be inside a corporate upskilling cohort.

Role. The credit officers, branch managers, and customer service staff are being treated as if their jobs will be unchanged by AI. The credible forecast is the opposite: their jobs change first, fastest, and most. They are also the layer least visible to the head-office programmes designed to upskill people who already look like the people designing the programmes.

Also Read: The future of marketing isn’t about AI, it’s about judgment

Why traditional upskilling is missing them

The corporate AI upskilling model assumes three things that do not hold for most of the workforce.

It assumes the learner already has digital fluency at the level a generative AI tutorial requires. For a large slice of the regional financial workforce, that baseline is uneven.

It assumes the learner has discretionary time. The branch officer running a six-day workweek with overtime cannot complete a six-hour learning module — and her manager is not measured on whether she does.

It assumes the learner will apply the new skill in their current job. For the workforce most exposed to displacement, that is not the right framing. They need either a new role inside the institution or a transition plan out of it. The programme that does not address that question reads as condescension.

What is starting to work

A few quieter responses are visible if you look for them.

Peer-led learning channels. The most active AI literacy communities I see in Indonesian financial services right now are running inside WhatsApp and Telegram groups organised by mid-career practitioners themselves — sharing tutorials, screenshots, and case discussions in Bahasa, often at a pace that no formal corporate programme can match. The credential is informal, but the practical literacy is real.

Vernacular content. A small but growing number of practitioners are publishing tutorials and case discussions in Bahasa Indonesia on YouTube and TikTok, often in fifteen-minute formats that match how working professionals actually learn. The audience is large. The production cost is low.

Internal apprenticeship over external certification. The institutions making the most progress are the ones that have paired senior practitioners with frontline staff inside cross-functional projects. The certificate is a side effect of the work, not the work itself.

Also Read: When startups fail, should VCs go to jail?

What the stakeholders should be doing

Institutions should stop measuring upskilling by completion rates of vendor-delivered courses and start measuring it by retention and internal mobility of frontline staff. The metric drives the programme. Change the metric.

Regulators should require disclosure of who is being upskilled, not just how many. If a bank reports that ninety per cent of its head office has completed AI training while sixteen per cent of its branches have, that asymmetry should be visible to the supervisor.

Government and civil society should invest in vernacular AI literacy at scale, particularly for the back-office workforce that will be most affected. The cost of this investment, relative to the cost of the dislocation it would prevent, is small.

The macro stakes

In every wave of automation, the people who adapt first compound advantages, and the people who adapt last absorb the dislocation. AI will not be different. What is different about this wave is the speed and the visibility.

ASEAN has roughly thirty-six months before the second-order effects of the current upskilling pattern become irreversible — before the branch closures, the role re-bundlings, and the displacement decisions are made on the basis of who has and has not learned the new tools. We are not running out of time to teach. We are running out of time to teach the right people.

The first governance debt was about who governs the AI. The second governance debt — quieter, slower, more politically charged — is about who gets to work alongside it. The institutions that ignore the second one will pay for both.

The post The fraud officer in Yogyakarta won’t catch the AI wave, and most ASEAN institutions don’t know it yet appeared first on e27.