AI Agents Are Reshaping Analytics—What SG Firms Do

Thomson Reuters fell nearly 18% in a day after investors reacted to a product update from Anthropic—not an earnings miss, not a scandal, not a macro shock. Just new AI plug-ins.

That’s the most useful signal in the whole news cycle: markets are repricing “knowledge work” as if automation is about to hit revenue models, not just headcount. If you run a business in Singapore, you don’t need to trade these stocks to care. You need to understand what changed—and how to build an AI adoption plan that protects your margins and makes your team faster.

This post is part of the AI Business Tools Singapore series, focused on practical ways to adopt AI for marketing, operations, and customer engagement. The goal here isn’t hype. It’s clarity: what these AI “coworker” tools mean, what gets disrupted first, and what Singapore companies can do this quarter.

One-liner worth repeating: When AI can do the work, customers stop paying for the workflow.

Why one AI product update spooked investors

Answer first: Investors sold off analytics, legal-tech, and software names because Anthropic’s Claude “coworker” plug-ins looked like a credible path to automating tasks those companies charge for.

According to the Reuters report republished by CNA, Anthropic launched plug-ins for its Claude Cowork agent that automate work across legal, sales, marketing, and data analysis. The market reaction was broad:

• Thomson Reuters: down ~18% in a day (concerns around legal research/Westlaw revenue exposure)
• RELX: down ~14% (largest drop since 1988, per the article)
• Wolters Kluwer: down ~13%
• LegalZoom: down ~19.7%
• FactSet: down ~10.5%
• Morningstar: down ~9%
• Other UK names (Experian, Sage, LSEG, Pearson): down 6%–12%

What’s the underlying fear? The article captures it well: the old “visibility premium” (predictable per-seat software pricing, stable renewals, long-term contracts) gets shaky when AI lets teams do more with fewer users—or bypass tools entirely.

The real shift: from “software features” to “outcomes”

For years, enterprise software sold features: dashboards, templates, alerts, workflow routing.

AI agents sell outcomes: “prepare a first draft of the contract summary,” “produce the weekly pipeline report,” “extract themes from 500 customer tickets,” “recommend who to follow up with and write the email.”

That outcome focus changes buying behavior:

• If an agent can complete the task inside chat, the UI and workflow layers become less valuable.
• If the agent can call multiple tools, customers start asking why they pay three vendors for what feels like one job.
• If AI reduces the number of humans in the loop, per-user pricing loses its “automatic growth” effect.

This is why investors reacted so violently: they’re trying to price an uncertain transition from seat-based economics to value-based economics.

What this means for Singapore businesses (not investors)

Answer first: For Singapore companies, the opportunity is immediate: treat AI agents as a productivity layer across functions—but design governance and measurement before scaling.

Singapore is unusually exposed to this shift in a good way. We have a dense concentration of:

• regional HQ functions (finance, legal, HR, marketing ops)
• professional services (consultancies, agencies, accounting)
• data-driven sectors (fintech, logistics, retail platforms)

These are exactly the workflows the article highlights.

Three places AI agents pay off fastest in SG firms

1. Marketing ops and performance reporting
   Weekly and monthly reporting is repetitive and costly: pulling data, normalising metrics, explaining variances. AI agents can draft commentary, flag anomalies, and generate segmented insights for different stakeholders.

2. Sales enablement and pipeline hygiene
   Most CRMs are full of stale notes and inconsistent stages. Agents can summarise calls, update fields, propose next steps, and draft follow-ups—without asking reps to do admin at 11pm.

3. Customer support analytics
   Ticket tagging, trend detection, root-cause summaries, and “top issues by product/version” can be agent-driven. This reduces escalation load and gives product teams cleaner signals.

Here’s the stance I’ll take: If your team is still doing manual cross-tool copy/paste reporting in 2026, you’re paying a “tax” your competitors are already avoiding.

The disruption risk: who gets squeezed first

Answer first: Companies that sell access to information, standardised workflows, or junior-level analysis get squeezed first; companies that own proprietary data and enforce compliance-heavy processes hold up longer.

The market selloff concentrated around legal analytics and data services for a reason. AI can replicate a lot of “first-pass” work:

• summarising and extracting clauses
• drafting standard client emails and proposals
• building basic charts and narratives
• doing initial research across public sources

A practical way to think about your own business risk

Split your revenue or internal work into three buckets:

1. Commodity tasks (high automation risk)
   Repeatable, rules-based, lots of examples online. If it’s trainable in a playbook, an agent will do it.

1. Judgment tasks (medium risk)
   Require domain context, trade-offs, accountability. Agents assist, humans decide.

2. Assurance tasks (lower risk, but governance-heavy)
   Anything requiring audit trails, regulatory defensibility, and strict controls.

For Singapore firms, assurance tasks matter because of PDPA, sector rules (MAS guidelines in financial services), and client contractual obligations. That doesn’t block adoption—it changes how you implement.

How to adopt AI agents without creating new problems

Answer first: Start with 2–3 workflows, instrument them with clear metrics, keep humans in the loop, and set up data boundaries from day one.

I’ve found that most companies get AI adoption backwards. They buy tools first, then go searching for use cases. Better results come from starting with a workflow that’s already painful and measurable.

Step 1: Pick workflows with measurable before/after

Good starter workflows (measurable within 4–6 weeks):

• marketing performance recap (time to publish, accuracy, stakeholder satisfaction)
• sales call summary + follow-up drafting (response time, meeting set rate, rep admin time)
• customer ticket categorisation + weekly insights (time to classify, escalations reduced)

Define success metrics upfront:

• hours saved per week (by role)
• cycle time reduction (e.g., report turnaround)
• quality measures (error rate, rework rate)
• business impact (conversion rate, churn signals surfaced)

Step 2: Build “human checkpoints” where risk is real

Agents are great at drafting. They’re not accountable.

Use checkpoints for:

• legal advice / contractual commitments
• financial figures used externally
• HR decisions
• customer claims (refund eligibility, policy promises)

A simple operating rule works well:

• AI drafts → human approves → system sends

Step 3: Set data boundaries (especially important in Singapore)

Minimum viable guardrails for most SMEs and mid-market teams:

• Define what data can enter prompts (no NRIC, no sensitive medical data, no bank account numbers)
• Use role-based access for internal knowledge bases
• Keep an audit log for regulated workflows
• Standardise an “approved prompt library” for recurring tasks

If you operate in regulated sectors, treat this as an internal control, not an IT preference.

Step 4: Decide what you’re actually buying: model, agent, or workflow

This is where many teams overspend.

• A model answers questions and drafts text.
• An agent executes multi-step tasks across tools.
• A workflow is a production-ready process with inputs, approvals, logging, and reporting.

The selloff described in the article is fundamentally about workflow displacement. So your competitive advantage comes from owning the workflow design, not from picking the “coolest chatbot.”

What investor reactions reveal about pricing and competition

Answer first: Investor panic is a hint that pricing power is moving from “access to software” to “proof of outcomes,” and Singapore firms should negotiate and implement accordingly.

The Reuters/CNA piece notes fears that AI tools let businesses do more with fewer staff, undermining per-user pricing. That’s not just a stock-market narrative—it changes how you should buy business software.

How to buy smarter in 2026

When renewing analytics, CRM add-ons, legal research tools, or marketing platforms, push for:

• usage-based or outcome-based pricing tied to delivered value
• clear exportability of your data (avoid tool lock-in)
• integration commitments (APIs, logging, permissions)
• roadmap clarity: what will be agent-native vs “AI labels” on old features

If a vendor can’t tell you how their product coexists with AI agents—or insists everything must run through their UI—you’ve learned something important.

FAQ: common questions Singapore teams ask about AI agents

“Will AI agents replace my analytics team?”

They’ll replace a chunk of the work, not the function. Expect fewer hours on manual reporting and more time on measurement design, experimentation, and decision support.

“Where should we start: marketing, ops, or customer support?”

Start where you have high volume + clear metrics + low regulatory risk. For many SG SMEs, that’s marketing ops or customer support insights.

“What’s the biggest adoption mistake?”

Letting everyone use AI however they want. You get inconsistent outputs, data leakage risk, and no measurable ROI.

What to do next (this week, not “someday”)

The market reaction to Anthropic’s plug-ins is a loud reminder: AI automation is now a competitive expectation, not a side project. Whether or not investors overreacted, the direction is clear—tools that automate knowledge work are arriving faster than most planning cycles.

If you’re building your stack for the rest of 2026, take a simple next step:

1. List your top 10 recurring knowledge-work tasks (reports, summaries, proposals, ticket analysis).
2. Choose the top 2 with the clearest metrics.
3. Pilot an AI agent workflow with approvals and logging.
4. Measure hours saved and quality changes—then scale.

You don’t need to “wait for certainty.” The companies that win in Singapore will be the ones that treat AI like process engineering: design, test, measure, iterate.

Source article: https://www.channelnewsasia.com/business/anthropics-new-ai-tools-deepen-selloff-in-data-analytics-and-software-stocks-investors-say-5906991