AI Automation Is Squeezing IT—What SG Firms Do Now

A 6% one-day drop in Indian IT stocks isn’t just a market mood swing—it’s a signal that AI automation is starting to hit the economics of traditional IT services. When analysts warn that 9%–12% of industry revenue could disappear over the next four years due to AI-led disruption, they’re really pointing at a simple truth: a lot of “hours-based” work is getting faster, cheaper, or avoided entirely.

For Singapore businesses, this isn’t a story about Indian vendors. It’s a preview of what’s coming for how software gets built, maintained, tested, supported, and improved—and how quickly your competitors will be able to ship changes. In this instalment of the AI Business Tools Singapore series, I’ll translate the headlines into practical decisions you can make in 2026 to protect budgets, increase speed, and create new growth.

The core shift: AI reduces time-to-deliver for application work, and that changes pricing, vendor strategy, and the internal skills you need. If you get ahead of it, you’ll spend less on routine work and more on outcomes.

Why Anthropic-style automation threatens IT services margins

Answer first: AI makes common application tasks dramatically faster, so the industry that charges for people and time faces immediate pressure on revenue and margins.

In the Reuters/CNA report, analysts flagged “high-margin application services” as the most exposed. Application services typically include:

• Enhancements and change requests
• Bug fixes and incident resolution
• Testing and QA
• Integrations between systems
• Ongoing maintenance and minor feature delivery

These are exactly the areas where modern AI assistants and agent-style automation can compress delivery timelines. The article cites that application services account for roughly 40%–70% of revenues for IT firms, with some large providers (e.g., TCS, Tech Mahindra, LTIMindtree) reportedly closer to ~55%–60% exposure, while another (HCL Tech) is nearer ~40%.

When a vendor’s revenue depends on high volumes of repeatable work, speed is a double-edged sword:

• If AI halves the time, clients will demand lower run-rate costs.
• If AI improves quality, clients will expect fewer incidents and less support spend.
• If AI enables more in-house capability, clients may buy fewer external hours.

That’s why Jefferies warned there’s “more pain ahead,” and why JPMorgan pushed back that the selloff could be overdone—because replacement won’t happen overnight for mission-critical systems. Both can be true.

The real issue isn’t “AI replacing IT”—it’s deflation

Most companies won’t rip out core banking, ERP, or critical logistics systems just because a new model ships. But pricing power still weakens when delivery becomes less labour-intensive.

Deflation looks like:

• The same backlog delivered in fewer sprints
• A fixed budget renegotiated downward at renewal
• More work bundled into “managed services” with tougher KPIs

For Singapore buyers, this changes how you negotiate and how you structure contracts.

What this means for Singapore businesses buying IT services

Answer first: If you buy IT services, you should assume faster delivery, lower unit costs, and new operational risks—and update procurement, governance, and KPIs accordingly.

Singapore’s digital transformation agenda has matured. Many firms already run cloud platforms, data stacks, and omnichannel customer journeys. The next phase is about speed and operating leverage: getting more done without growing headcount or vendor spend at the same rate.

AI accelerates that—if you set it up properly.

1) Procurement: shift from effort-based to outcome-based pricing

If your contract still pays primarily for time (e.g., day rates, “capacity teams,” or ticket volumes), you’re effectively paying for inefficiency.

Here’s a better stance:

• For enhancements: price by features shipped, not hours spent
• For support: price by service levels and incident reduction, not ticket handling
• For QA: price by coverage and escaped defects, not number of test cases

A practical approach I’ve found works: keep a small “capacity” component for discovery and uncertainty, but put the bulk of spend under measurable outcomes.

2) Governance: require an AI delivery playbook from vendors

Don’t accept vague statements like “we use AI internally.” Ask for specifics:

• Which SDLC stages use AI (requirements, coding, testing, release notes)?
• What’s the human review process and who is accountable?
• How do they prevent leakage of confidential code/data?
• What are their benchmarks (cycle time, defect rate, MTTR) before vs after AI?

If a vendor can’t answer, you’re not buying capability—you’re buying marketing.

3) Risk: AI can increase speed while creating new failure modes

Faster code generation is helpful, but it also increases:

• Change volume (more releases means more potential regressions)
• Dependency risk (generated code may pull in libraries or patterns you don’t standardise)
• Security exposure (secrets, insecure defaults, weak auth flows)

The fix isn’t to ban AI. The fix is to strengthen:

• Automated testing gates
• Secure coding standards
• Code review discipline
• Observability (logs, traces, alerts)

Where the biggest opportunity is: AI tools for marketing and operations

Answer first: The winners won’t be the companies that “use AI.” They’ll be the ones that rebuild workflows so marketing and operations run faster with fewer handoffs.

The CNA piece focuses on IT sector revenues, but the buyer-side upside is huge. For Singapore SMEs and mid-market firms, the fastest ROI usually comes from front-office and ops automation, not building fancy AI products.

Use-case map: quick wins Singapore teams can ship in 30–60 days

Start where work is repetitive, text-heavy, and measurable.

Marketing

• Draft and version ad copy across channels (Meta, Google, LinkedIn)
• Generate landing page sections and FAQs based on product notes
• Summarise campaign performance and propose next-week experiments
• Create sales enablement snippets (objection handling, pitch variations)

Customer operations

• Auto-triage inbound emails/chats into categories and priority
• Suggest replies with policy-compliant language
• Summarise cases for handoffs and escalations
• Build an internal “answer engine” for SOPs and product knowledge

Finance and admin

• Extract fields from invoices/POs
• Draft vendor follow-ups and reconciliation notes
• Summarise monthly spend anomalies

IT and product (internal)

• Generate test cases from user stories
• Draft release notes from commits/tickets
• Propose refactors and document APIs

A useful rule: if the output is reviewed by a human before it goes external, it’s easier to launch safely.

A practical 90-day plan to adapt (without chaos)

Answer first: Build a small AI operating system—data rules, tool choices, KPIs, and training—then scale what works.

If you try to “roll out AI” as a single big initiative, it becomes a mess of unapproved tools and inconsistent quality. A 90-day plan keeps it real.

Days 1–15: Pick 2 workflows and define success metrics

Choose one marketing workflow and one ops workflow. Define 3–5 metrics.

Examples:

• Marketing: content production time, cost per lead, landing page conversion rate
• Ops: first response time, resolution time, deflection rate, CSAT

Days 16–45: Standardise tools and set guardrails

Decide:

• Which AI tools are approved (and for what data sensitivity)
• What must never be pasted into public models
• Who can create automations and who reviews them

Write it down as a one-page policy. If it takes 20 pages, nobody will follow it.

Days 46–90: Automate, measure, and renegotiate vendor expectations

This is where you turn AI into leverage:

• Automate parts of the workflow (drafting, triage, summarisation)
• Train reviewers (what good looks like, what to reject)
• Measure improvements weekly
• Bring the data to vendor discussions: “We’re seeing 30% faster turnaround internally; how does your delivery model change?”

That last step matters. If AI is deflating effort-based work, your contracts should reflect it.

People also ask: Will AI replace outsourcing for Singapore firms?

Answer first: AI won’t eliminate outsourcing, but it will shrink low-complexity outsourced work and raise the bar for what you pay external teams to do.

Outsourcing still makes sense for:

• Complex system modernisation
• Multi-system integrations with heavy governance
• Regulated environments needing strong controls
• Niche expertise (security, data engineering, performance)

What becomes harder to justify is paying premium rates for repeatable tasks that AI-assisted teams can do quickly.

What should you expect from IT partners in 2026?

You should expect:

• Faster delivery with transparent benchmarks
• Stronger automation in QA and support
• Clear policies on AI usage and data handling
• More outcome-based commercials

If a partner insists nothing changes, that’s a red flag.

The stance I’d take if I ran a Singapore business unit

Answer first: Treat AI as a pricing and speed reset. Use it to cut cycle time, then reinvest part of the savings into better data and customer experience.

The Reuters/CNA story highlights fear in the IT market—compressed timelines, disruption to labour-intensive models, valuation risk. On the buyer side, it’s a chance to buy outcomes instead of headcount.

Here’s what works:

• Start with workflows tied to revenue (leads, conversions) and cost (support load)
• Put guardrails around data and approvals early
• Demand measurable AI-enabled productivity from vendors
• Renegotiate contracts as unit costs fall

If you do this well, your competitors will feel like they’re moving through molasses.

You don’t need a moonshot AI strategy. You need two real deployments, measured weekly, and expanded only when they perform.

If you’re building your 2026 plans now, what’s the one business workflow you’d most like to make 30% faster—without hiring another headcount?