AI Automation Is Squeezing IT Services—SG’s Next Move

A 6% single-day drop in Indian IT stocks is a loud signal, not market drama. It reflects a growing belief that AI automation can shrink the very work that used to pay for big IT services teams—especially application services, where analysts say 40–70% of revenue sits for many firms.

The trigger, according to reporting on Feb 5, 2026, was renewed attention on automation pushes by firms like Anthropic (and also Palantir), plus analyst notes warning that AI could compress project timelines and structurally reduce high-margin application work. Whether you agree with the market reaction or not, the underlying message matters for Singapore: when AI reduces billable hours elsewhere, it changes pricing expectations everywhere.

This post is part of the AI Business Tools Singapore series, where we focus on practical adoption for marketing, operations, and customer engagement. Here, the lens is different but urgent: AI is not just a tool you buy—it’s a force that rewrites how services are priced, delivered, and justified.

Why analysts think application services revenue is at risk

Answer first: AI hits IT services revenues by reducing human effort per ticket—and the industry’s traditional model charges for that effort.

In the Reuters/CNA story, analysts point to application services as the pressure point. This includes maintenance, enhancements, testing, incident response, minor upgrades, and the steady stream of “keep the lights on” work that many enterprises outsource.

Here’s the mechanism that spooks investors:

• Automation compresses timelines. If an app change used to take 10 days of blended effort and now takes 4, the old pricing logic breaks.
• Labour intensity becomes a weakness. The more your economics depend on staffing, the more you feel “deflation” when software does the work.
• Clients renegotiate faster than vendors can re-skill. Even if IT firms build AI practices, procurement teams will still ask: “Why are we paying the same for fewer hours?”

Jefferies’ note (as cited) argues that deflation in legacy services may outweigh growth in new AI-related work in the next 1–2 years. Motilal Oswal’s estimate quoted in the story is also specific: 9%–12% of industry revenues could be eliminated over the next four years due to AI-led disruption.

That’s not a niche impact. That’s a restructuring of how work is sold.

“But enterprises won’t replace everything overnight” is true—and still not comforting

JPMorgan’s view in the article is the counterweight: it’s illogical to assume enterprises will replace every layer of mission-critical software because a few new tools launched.

I agree with the spirit of that pushback. Most organisations don’t swap core systems quickly, and they shouldn’t.

But the market fear isn’t “everything gets replaced tomorrow.” It’s this: even partial automation can erase the most profitable hours first—the repetitive tasks, testing cycles, L1/L2 support, documentation, and routine integration work that used to be stable revenue.

What this means for Singapore businesses buying IT services

Answer first: Singapore buyers will gain negotiating power, but only if they modernise how they procure and measure outcomes.

Singapore enterprises sit in an interesting position. We’re not the global factory for IT services, but we’re heavy consumers of them—directly and indirectly—through regional delivery models.

When global providers face pricing pressure, Singapore firms will see:

1. Lower tolerance for “time-and-materials” billing
   • Expect a shift toward fixed outcomes, managed services with tighter SLAs, and pricing tied to automated throughput.
2. More AI clauses in contracts
   • Vendors will pitch AI accelerators; buyers should ask who owns the prompts, runbooks, fine-tuned assets, and telemetry.
3. A new baseline for speed
   • If competitors ship changes weekly because they’ve automated testing and release workflows, quarterly release cycles start to look negligent.

There’s also a less obvious implication: your vendors will try to protect margin. Some will do it by upselling “AI transformation programs.” Some will do it by reducing senior oversight. That’s why procurement and IT leadership need to get sharper, not just cheaper.

The hidden cost: “cheaper delivery” can create more operational risk

AI can reduce effort, but it can also increase the pace of change—and with pace comes risk.

For Singapore firms in regulated or high-trust industries (finance, healthcare, critical infrastructure), the question isn’t “Can AI do it?” The question is:

• Can we prove what changed?
• Can we trace who approved it?
• Can we roll back safely?
• Can we audit AI-assisted decisions and outputs?

If your current operating model relies on manual checks as safety rails, automating delivery without upgrading governance is how you get unpleasant surprises.

How Singaporean enterprises should respond: build an “AI-ready ops” layer

Answer first: The winning move is not buying more AI tools—it’s redesigning workflows so AI produces measurable outputs with strong controls.

If you’re leading ops, IT, or a business function with heavy workflow volume, treat this as a 2026 priority: AI readiness is operational readiness.

Step 1: Identify work that will deflate first

The work most likely to shrink (and therefore change vendor economics and internal headcount planning) is typically:

• Regression testing, test case generation, test data preparation
• Routine change requests and “small enhancements”
• L1/L2 support, knowledge base upkeep, ticket classification
• Documentation, release notes, environment setup scripts
• Basic integration mapping and code scaffolding

A practical exercise: pull 90 days of tickets and classify them into:

• Repeatable (same pattern, different parameters)
• Judgement-heavy (requires domain decisions)
• Risk-heavy (requires strict approvals/audit)

Repeatable work is where AI business tools deliver ROI fastest.

Step 2: Shift from “hours saved” to “cycle time and error rate”

Most companies get measurement wrong. They celebrate “we saved 200 hours,” then fail to show faster delivery or fewer incidents.

Use metrics that actually change business outcomes:

• Change lead time (idea → production)
• Deployment frequency (and rollback frequency)
• Post-release incident rate
• Mean time to resolve (MTTR)
• Cost per resolved ticket

When you measure these, it becomes easier to choose which AI automation is worth it—and which is just noisy.

Step 3: Put guardrails where AI is most likely to mislead you

AI is excellent at plausible output. That’s also the problem.

For enterprise use, I’ve found the simplest controls work best:

• Human approval for high-risk changes (permissions, payment flows, customer data)
• Golden test suites that must pass regardless of AI-generated code
• Prompt + output logging for auditability (especially for customer-facing responses)
• Source-of-truth retrieval (RAG) so AI answers from your policies, not its memory

If you can’t explain why a decision was made, you can’t defend it to auditors, customers, or your own board.

A practical playbook: using AI business tools without blowing up governance

Answer first: Start with narrow use cases, integrate into existing workflows, and renegotiate vendor relationships around outcomes.

Here’s a straightforward plan many Singapore SMEs and mid-market teams can execute in 30–60 days.

1) Pick one workflow with volume and pain

Good candidates:

• Customer support triage and first response drafting
• Invoice processing and exception handling
• Sales proposal drafting with product policy checks
• IT service desk automation (routing, KB suggestions)

Bad candidates (at the start): anything that directly moves money or changes permissions without a human in the loop.

2) Define “done” in operational terms

Examples of solid “done” definitions:

• “Reduce first-response time from 6 hours to 1 hour, while holding CSAT steady.”
• “Cut month-end reconciliation from 5 days to 3 days, with documented audit trail.”
• “Increase release frequency from monthly to weekly without raising incident rate.”

This is how you avoid AI pilots that never graduate.

3) Decide the operating model: build, buy, or partner

If the CNA story is right about service-line deflation, then this becomes a strategic choice:

• Buy (SaaS tools): fastest for common workflows; best for SMEs.
• Build (internal): best when data and processes are unique or sensitive.
• Partner (IT vendor): useful, but rewrite the incentives—pay for outcomes, not headcount.

A blunt stance: if a vendor still wants to bill you for the same hours after automation is introduced, you’re funding their margin protection.

4) Renegotiate contracts around automation reality

Add clauses that reflect the new world:

• Productivity assumptions (expected cycle-time reduction)
• Shared ownership of AI assets (prompts, runbooks, test generators)
• Security and data-handling terms for AI usage
• Reporting requirements (what was automated, what was reviewed)

This is where Singapore buyers can be disciplined and win.

What happens next: fewer “projects,” more “always-on improvement”

Answer first: AI pushes IT and ops toward continuous delivery—so organisational bottlenecks (approvals, unclear ownership, data silos) become the real constraint.

Even if some market panic is “plenty of panic over a little flutter” (as Kotak Institutional Equities put it), the direction is clear: automation raises expectations.

That doesn’t mean IT services disappear. It means the value shifts:

• from manual execution → to workflow design and control
• from writing code → to verifying, integrating, and governing changes
• from staffing capacity → to shipping outcomes reliably

Singapore enterprises that adapt will move faster with the same (or smaller) teams. Enterprises that don’t will still pay—just not always to their vendors. They’ll pay in slower launches, higher incident rates, and talent churn.

If you’re building your stack of AI business tools in Singapore this year, treat this moment as a forcing function: modernise procurement, modernise ops metrics, and modernise governance—together.

One question worth bringing to your next leadership meeting: if your competitors cut delivery time in half using AI automation, which part of your organisation becomes the bottleneck first—technology, process, or decision-making?

Source article: https://www.channelnewsasia.com/business/anthropics-ai-push-raises-analyst-concerns-over-it-services-revenues-5909371