AI Literacy at Work: A Practical Plan for SG Firms

A team can spend months “training on AI” and still freeze the first time you ask them to improve a process with it. The problem usually isn’t effort. It’s where the learning happens.

That’s why Senior Minister of State for Manpower Koh Poh Koon’s point (shared during a visit to Grab on Feb 5, 2026) matters for every Singapore business: work and study requirements need to be better integrated to build AI literacy. In plain terms, the old model—study first, work later—doesn’t match how fast AI tools are changing day-to-day work.

This post is part of our AI Business Tools Singapore series, focused on practical adoption for marketing, operations, and customer engagement. I’ll translate the policy signal into an actionable plan: how to build AI literacy in your company without turning it into a “course completion” exercise that never touches real workflows.

Why AI literacy has to move into real workflows

AI literacy isn’t “knowing what AI is.” In a business setting, AI literacy is the ability to use AI tools safely and productively to complete real tasks—writing a better customer reply, summarising a call, forecasting inventory, drafting a campaign brief, or translating a menu.

Dr Koh highlighted two realities that business owners feel every week:

1. Business models are churning faster because technology and AI shorten the time from idea to execution.
2. Worker anxiety is rising when AI is framed as replacement rather than complementarity.

Here’s my stance: most companies worsen the anxiety by rolling out training that’s abstract (“AI fundamentals”) while leaving the actual work unchanged. Employees then assume AI is either a threat or a toy.

The fix is simple but not easy: design AI literacy around the work people already do, then adjust “study requirements” (training, assessments, career pathways) to match those tasks.

What the Grab example gets right

The Straits Times piece described GrabAcademy running an AI upskilling workshop for 30 driver-partners using tools like Gemini, ChatGPT, and ElevenLabs for practical tasks such as translating phrases into multiple languages. Grab also aims to train 10,000 drivers and merchant partners by 2028, has trained 300+ merchant partners, and has hired 50+ people into an AI Centre of Excellence created in May 2025.

You don’t need to be Grab to learn from this:

• Hands-on beats theory. Translation, scripting, and voice tools map to real needs.
• AI literacy can be role-based. Drivers and merchants aren’t being asked to code.
• Career mobility is part of the story. Training supports transitions, not just productivity.

For SMEs, the takeaway is not “run big programmes.” It’s: make AI practice unavoidable and useful—inside the workflow.

The business case: AI literacy is operational risk management

For Singapore companies, AI literacy is now a mix of growth strategy and risk control.

Without AI literacy, AI adoption fails in predictable ways:

• Marketing uses AI to produce more content, but brand quality drops and approvals slow down.
• Ops teams try “automation,” but no one maintains prompts, workflows, or exception handling.
• Customer service pilots a chatbot, but escalation rules are unclear and complaints rise.
• Managers ban tools after one incident, and shadow AI usage spreads anyway.

With AI literacy, the same tools become measurable improvements:

• Faster first drafts and better personalisation for campaigns.
• Cleaner handoffs between teams (sales → delivery → support).
• Better internal knowledge access (policy answers, SOPs, product specs).
• Consistent governance: what’s allowed, what’s not, and why.

The point Dr Koh made about “complementarity” is the practical lens: the win is not replacing staff; it’s raising the output per person while reducing rework.

A work-study integration model Singapore companies can actually run

If you want “work and study” to integrate, you need a structure that doesn’t collapse under real deadlines. Here’s a model I’ve found works across office teams and frontline operations.

1) Define AI literacy by role (not by department)

Start by mapping roles to repeated tasks. Then attach a small number of AI capabilities.

Example role-based AI literacy map:

• Customer service reps: summarise cases, draft responses in your tone, translate, detect intent, create handover notes.
• Marketing execs: campaign briefs, ad variations, audience Q&A mining, landing page structure, simple creative QA checklists.
• Ops coordinators: SOP drafting, incident post-mortems, checklist generation, shift handover summaries.
• Sales: call summaries, objection handling practice, account research templates, proposal outlines.
• HR / L&D: job descriptions, interview question banks, onboarding guides, policy Q&A prompts.

Keep it tight: 5–7 repeatable tasks per role is enough to start.

2) Build a “task library” (your internal AI playbook)

Dr Koh suggested a playbook approach so companies can adopt what others have learned. Do it internally first.

Your AI business tools playbook should include:

• The task (“Turn a 10-minute call into CRM notes”)
• The input (“Paste transcript; remove NRIC and bank details”)
• The prompt template (approved, versioned)
• The output standard (“Must include next steps, owner, due date”)
• The check (“Human review required for promises/fees/policy”)

Treat prompts like SOPs. Version them. Assign owners. Retire bad ones.

3) Make learning measurable: “proof of work,” not attendance

Course completion is a weak signal. Replace it with proof of work.

Examples of proof-of-work assessments:

• Customer service: 5 real tickets redrafted with AI + quality review score
• Marketing: 1 campaign brief generated + revised + approved by manager
• Ops: 1 SOP rewritten + tested in a real shift + exceptions documented

This integrates “study requirements” into workplace output—exactly the direction policymakers are pointing to.

4) Set rules that are clear enough to follow on a busy day

If governance is vague, people ignore it.

A practical AI usage policy for Singapore SMEs should answer:

• What data is prohibited? (NRIC, bank details, medical info, confidential client pricing)
• Which tools are approved? (named tools and approved accounts)
• What requires human approval? (public-facing claims, refunds, legal terms)
• How do we store outputs? (where summaries/SOPs live, retention rules)

Keep it to one page. Add examples.

Where to start: 3 high-ROI AI literacy projects (marketing, ops, service)

If you’re trying to build momentum in Q1–Q2 2026, pick projects that (1) touch daily work, (2) can be measured weekly, and (3) don’t require complex integrations.

Project A: Marketing content that doesn’t sound like AI

Answer first: standardise your brand voice and QA checks before you scale AI content.

A workable rollout:

1. Create a brand “voice sheet” (words you use, words you avoid, tone examples)
2. Build 3 prompt templates: ad copy, email, landing page sections
3. Add a QA checklist: claims, pricing, compliance, tone, CTA, audience fit
4. Track: time-to-first-draft, revision cycles, conversion rate by variant

This avoids the common trap: pushing volume and creating more editing work than before.

Project B: Ops playbooks that actually get used

Answer first: use AI to reduce the friction of documenting and updating SOPs.

Try this:

• Record a 15-minute walkthrough of a process (e.g., outlet closing)
• Transcribe and have AI draft the SOP
• Get the shift lead to test it once
• Update with exceptions (“If X happens, do Y”)

Track: incidents, training time for new hires, and handover errors.

Project C: Customer support summaries and faster escalations

Answer first: summaries are the safest, quickest AI win for service teams.

Use AI to produce:

• Case summary (customer issue, timeline, actions taken)
• Next-best action suggestions (internal only)
• Escalation note (what’s needed from L2/L3)

Track: average handling time, first response time, repeat contact rate.

The hidden requirement: managers need AI literacy first

If managers can’t judge AI output, teams will either overtrust it or refuse to use it.

A manager’s AI literacy checklist should include:

• Can you spot when an output is confidently wrong?
• Can you ask for evidence, assumptions, and edge cases?
• Can you coach prompt iteration without micromanaging?
• Can you define “done” with quality standards?

This is where “work-study integration” becomes culture: leaders model the behaviour.

People also ask: what counts as “basic AI literacy” for non-technical staff?

Basic AI literacy in the workplace means a person can:

1. Choose the right task (drafting, summarising, translating, brainstorming, structuring)
2. Give usable inputs (context, constraints, examples, format requirements)
3. Verify outputs (check facts, policy alignment, tone, numbers)
4. Handle data responsibly (know what not to paste into tools)
5. Improve the workflow (save prompt templates, document exceptions)

Coding is optional. Judgment isn’t.

A practical 30-day plan for Singapore SMEs

If you want something you can run without a giant budget, here’s a tight sequence.

Week 1: Pick one workflow per team

• Marketing: 1 campaign asset type
• Ops: 1 SOP
• Service: 1 case type

Week 2: Build templates + policy

• Create 3–5 prompt templates
• Publish a one-page AI usage policy

Week 3: Run proof-of-work assessments

• Each person submits 2–3 real outputs
• Manager reviews using a scoring rubric

Week 4: Standardise and scale

• Add templates to your playbook
• Assign owners and version control
• Decide what to automate next (only after the workflow is stable)

If you do only this, you’ll have real AI literacy—not a slide deck.

What this means for AI Business Tools Singapore (and your next move)

Singapore’s direction of travel is clear: AI literacy will increasingly be built through integrated work-and-study pathways, not isolated training. The Grab example shows what hands-on adoption looks like, and Dr Koh’s comments signal that the workforce strategy will push further in that direction.

For businesses, the opportunity is immediate: treat AI literacy as a company operating capability. Put it into daily tasks, measure it through proof of work, and document it as a playbook your people can actually use.

If you had to choose one process to “bring AI into work” this month—marketing production, ops SOPs, or customer support—which one would show results in 14 days, not 14 months?