Organizations across Asia have become adept at deploying AI. Whether anyone uses it well is another question, and for Jun Tay, finance technology specialist at DBS Bank in Singapore and an ACCA member, the gap between the two is where adoption is won or lost.
“AI adoption most often breaks down at the onset of change: the point where operating reality, data context and human behavior meet,” she says. Capable platforms are not in short supply, she observes, but adoption becomes meaningful only when people understand the problem being solved, trust the knowledge behind the tool and know how to apply its output in their day-to-day work.
That is why she considers adoption rarely a technology problem. What makes AI different from previous waves of workplace technology, in her view, is that it exposes organizational gaps faster—because it changes not only the system people use, but how they ask questions, interpret answers and make decisions. The shift asked of employees is considerable: from searching manually to prompting effectively, from relying on static documents to validating generated responses, and from treating technology as a separate system to embedding it into the operating rhythm. “That transition requires confidence, ownership and repeated practice,” she says.
See also: Is prompt engineering dead? One expert describes what HR should focus on instead
What a pilot revealed
Tay draws on a generative AI chatbot pilot she supported for a finance function. The technology itself was already available and could be scaled by onboarding the relevant knowledge base. She found the real work lay in making that knowledge usable.
Reviewing the pilot’s queries, she observed that how closely a user’s prompt matched the language of the underlying documents was a key determinant of answer quality, and she converted knowledge files into more reliably readable formats to serve a wider audience. The pilot ultimately performed at 89.8%, clearing the recommended 80% threshold, but that was not the lesson she took away. “Adoption depended on knowledge, design, access, governance and user readiness, not just the AI engine,” she says.
The pilot also surfaced what she describes as a translation gap—an early warning sign that a workforce is not yet ready for the tools being rolled out. A user might ask about “milestones” while the underlying document describes an “operating cycle.” The answer suffers, but not because the technology has failed. “That is not a technology failure; it is a translation gap between business language, documentation structure and AI retrieval behavior,” she explains.
Other warning signs follow a pattern: employees with access to a tool who revert to old ways of working because they are unsure when to trust it; low-quality prompts; inconsistent interpretation of outputs; over-reliance without validation; or outright avoidance because users cannot see how the tool fits their responsibilities. In finance, she notes, the stakes are higher because outputs often inform reporting, control reviews and stakeholder decisions.
In her view, the cause is underinvestment in change management. Organizations concentrate on the rollout while spending less time on use-case design, prompt guidance, documentation quality, training and governance. Her response is to treat AI adoption as an operating model change: clarifying what the tool should and should not be used for, who owns the knowledge, how quality is reviewed and how users build confidence over time.
She is careful, though, not to frame imperfect starts as failure. “Most products and tools are not released in a perfect or ideal state. From a progress standpoint, starting, failing safely and learning quickly are far better than avoidance or remaining with the status quo,” she says.
Accountability stays human
If one principle anchors Tay’s view of responsible adoption, it is accountability. “When AI fails, creates risks or leads to a poor decision, humans remain accountable. You cannot sue an AI in the courtroom,” she says. “AI may support judgement, but it does not replace responsibility.”
This, she argues, is where domain professionals earn their place in AI adoption. Technologists build the platform and enable the capability; domain experts determine whether the output is meaningful, the interpretation sound and the recommendation appropriate in context. In the pilot, her role was not to tune a model technically but to assess whether responses made sense to enterprise users asking real questions—gathering user champions, reviewing both acceptable and weaker responses, and identifying where document structure and prompt wording needed better alignment. “That is the contribution of a domain expert: knowing what ‘good’ looks like in practice,” she says.
The organizations that succeed at scale, she believes, will combine technology expertise with strong domain ownership—generating not just answers, but trusted, decision-useful ones.
The post ‘You cannot sue an AI’: Why adoption comes down to people, not platforms appeared first on HR Executive.
This article was originally published on HR Executive. Click below to read the complete article.