AI ROI for Business: Lessons from IQVIA’s Push

A six-day market rout doesn’t happen because investors get bored. In early February 2026, a selloff erased about US$830 billion from software and services stocks as new AI tools (including from Anthropic) raised an uncomfortable question: If AI can do more of what “knowledge businesses” do, who gets displaced?

That anxiety showed up clearly in a Reuters-reported earnings call for IQVIA, a healthcare data and clinical research giant. Analysts pressed the company on whether fast-moving AI would erode its moat. IQVIA’s CEO Ari Bousbib didn’t hedge. He argued that AI doesn’t replace IQVIA’s core—it makes it more valuable, largely because the company’s advantage is proprietary healthcare data assets that general-purpose AI models can’t simply copy.

This matters for the AI Business Tools Singapore series because most Singapore companies are facing the same debate, just at a different scale. You don’t need a global earnings call to ask: Will AI improve margins, speed, customer experience—and can we prove it? The reality? Most AI projects fail in Singapore for the same reason they fail elsewhere: they’re built around tools, not business outcomes.

Below is the practical takeaway from IQVIA’s moment in the spotlight: how to think about AI ROI, why “AI will replace us” is often the wrong fear, and what Singapore teams can do this quarter to deploy AI business tools in marketing, operations, and customer engagement—with numbers that hold up in a boardroom.

What IQVIA’s AI debate really signals for businesses

Answer first: The IQVIA story is a reminder that AI changes the value of your assets and workflow, not just your headcount.

Analysts worried that new AI capabilities could replace parts of IQVIA’s services. IQVIA’s rebuttal was blunt: a model isn’t the business. The business is the combination of:

• Unique data (IQVIA claims the “largest proprietary healthcare information assets in the world”),
• Domain workflows (clinical research operations, regulatory-grade processes, quality systems), and
• Trust (clients pay for reliability, auditability, and outcomes).

In Singapore, the same structure applies—even if you’re running a 30-person e-commerce brand or a 300-person professional services firm.

A useful Singapore framing: “AI is cheap; certainty is expensive”

If there’s one line I’d steal for local decision-makers, it’s this: AI output is getting cheaper; business certainty is not.

General-purpose AI can produce drafts, summaries, forecasts, and “good enough” answers. But the expensive part is ensuring those outputs match your policies, product rules, customer promises, and regulatory obligations.

That’s why companies with better internal “assets” (clean data, documented SOPs, strong QA) often see faster AI ROI than companies that rush to buy a chatbot.

The ROI question analysts asked—and Singapore leaders should ask too

Answer first: AI ROI becomes clear when you measure it against one of three levers: revenue lift, cost reduction, or risk reduction.

IQVIA forecast 2026 adjusted earnings of US$12.55–US$12.85 per share, below expectations (LSEG consensus cited as US$12.95), partially due to nearly US$80 million in higher interest expenses. Notice what happened: the market didn’t just ask about AI features—it asked about financial impact.

Singapore SMEs and mid-market firms should do the same. If your AI initiative can’t map to a lever, it’s a science fair project.

A simple AI ROI scorecard (steal this)

Pick one primary metric and one secondary metric:

1. Revenue lift
   • Primary: conversion rate, average order value, qualified leads/week
   • Secondary: time-to-first-response, proposal turnaround time
2. Cost reduction
   • Primary: hours saved per role per week, tickets handled per agent
   • Secondary: rework rate, average handling time
3. Risk reduction
   • Primary: incident rate, compliance exceptions, chargebacks
   • Secondary: audit time, policy violations caught

Then define a baseline and a target. If you don’t have a baseline, you can’t claim ROI.

The moat myth: AI won’t replace you—unless you’re “just output”

Answer first: If your business sells undifferentiated knowledge output, AI will pressure pricing. If you sell verified outcomes tied to proprietary context, AI will expand your capacity.

The analyst fear around IQVIA mirrors fears across consulting, agencies, and B2B services: “What if a client uses AI instead of us?”

Here’s the stance I take: AI replaces deliverables, not accountability. Clients still pay for someone to own results, manage risk, and operate consistently.

What counts as “proprietary context” in Singapore companies?

You don’t need IQVIA-scale data to build defensibility. In local terms, proprietary context can be:

• Historical sales + margin data by SKU, channel, and campaign
• Customer service transcripts tied to outcomes (refunds, churn, upsell)
• SOPs and QA checklists that encode “how we do things here”
• Local regulatory constraints (PDPA handling rules, sector compliance)
• Pricing rules, inventory constraints, supplier lead times

When you connect AI business tools to your context, you stop competing with generic AI.

Where AI business tools pay off fastest (marketing, ops, CX)

Answer first: The fastest ROI usually comes from high-volume, repeatable workflows with clear acceptance criteria.

In Singapore, I’ve found the winners tend to be unglamorous: reducing cycle time, improving response quality, and eliminating manual handoffs.

Marketing: move from “more content” to “better conversion control”

Most teams start with AI copywriting. That’s fine, but it rarely creates durable ROI.

Better bets:

• Lead qualification assistants that score inbound leads against your ICP and route them correctly
• Offer testing: generate 10 variants, but only ship the ones that pass brand and compliance checks
• Sales enablement: auto-generate first-draft proposals tied to your price book and case studies

What to measure:

• MQL-to-SQL conversion
• Cost per qualified lead
• Time from inbound to first human follow-up

Operations: AI that reduces rework beats AI that “sounds smart”

Operations ROI is often easiest to prove because it shows up as time saved.

High-ROI patterns:

• Document processing (invoices, POs, claims) with extraction + validation
• Internal knowledge search grounded in your SOPs, not the open internet
• Exception handling: AI flags anomalies, humans decide

What to measure:

• Processing time per document
• Error/rework rate
• SLA adherence

Customer engagement: the goal isn’t fewer agents; it’s fewer bad conversations

A customer-facing bot that answers confidently—but wrongly—is worse than no bot.

A safer, more profitable approach:

• Use AI for triage and summarisation, not final decisions
• Use approved answer libraries (grounded responses) for policy topics
• Escalate with context: AI hands off a clean summary + recommended next step

What to measure:

• First contact resolution
• Average handling time
• CSAT changes per issue type

A practical 90-day AI adoption plan for Singapore teams

Answer first: A 90-day plan works when you limit scope, lock your data, and ship measurable pilots—not “platform programmes.”

Here’s a pragmatic sequence you can run without a huge data science team.

Days 1–15: pick one workflow and define “done”

Choose a workflow with:

• High volume (daily/weekly)
• Clear quality bar (pass/fail)
• Clear owner (one accountable team)

Write acceptance criteria like you would for a vendor:

• “Summaries must include: customer intent, constraints, next action, and risk flags.”

Days 16–45: build guardrails before you build features

If you’re in Singapore, you’ll run into PDPA and reputational risk fast.

Minimum guardrails:

• Data classification: what can/can’t be used in prompts
• Redaction rules for NRIC, health data, financial details
• Logging: what was asked, what was answered, who approved
• A “human override” path

Days 46–90: pilot, measure, and decide whether to scale

Ship the pilot to a small group. Measure against the baseline.

If the metrics don’t move, don’t argue—change the workflow or stop the pilot. That discipline is what analysts were implicitly asking IQVIA for: proof, not vibes.

Common questions Singapore leaders ask about AI ROI (quick answers)

Answer first: Most ROI confusion comes from mixing up capability with deployment.

“Will AI replace our team?”

If your team’s work is primarily producing repeatable first drafts, some tasks will shrink. But in practice, teams reallocate time to higher-value work: QA, client outcomes, and edge cases.

“Should we buy an AI tool or build something?”

Start by buying for common workflows (support triage, meeting summaries, marketing ops). Build only when you need deep integration with proprietary context or strict auditability.

“How do we stop hallucinations?”

Don’t ask AI to invent. Ground it in approved sources, limit it to classification/extraction where possible, and require citations to internal documents for policy answers.

What to do next if you’re serious about AI business tools

The IQVIA episode is a good reminder that investor-grade AI stories are always the same: unique assets + measurable outcomes + controlled risk. Singapore companies can absolutely do this without enterprise budgets, but you need the discipline to measure and the restraint to keep scope tight.

If you’re planning your 2026 AI roadmap, start with one workflow in marketing, operations, or customer engagement where you can show results in 90 days. Make the ROI undeniable. Then scale.

AI is getting more powerful every quarter. The companies that win won’t be the ones that “use AI.” They’ll be the ones that can explain—clearly—why AI improves their margins, speed, and customer trust.