AI Underwriting: Fix the Judgment Deficit, Not Data

A typical enterprise credit deal still eats up 16+ hours of manual work—collecting documents, spreading financials, scanning news, formatting memos. That’s not “rigour.” It’s clerical drag.

The more interesting point from a recent e27 story about Kevin Lee (ex-PayPal risk leader) isn’t that AI can speed underwriting up. It’s where the speed comes from: automation that protects underwriter judgment instead of burying it under more dashboards.

This matters to Singapore SMEs even if you’re not a bank. Underwriting and marketing have the same hidden bottleneck: teams spend most of their week preparing data instead of deciding what to do with it. When AI is used as a prep engine—cleaning, structuring, drafting, and citing—humans finally get their thinking time back. That’s the real ROI.

The hidden flaw in credit underwriting: a judgment deficit

Credit underwriting doesn’t mainly fail because of missing data. It fails because skilled people have too little time and energy to interpret the data that matters.

Kevin Lee’s “judgment deficit” framing is blunt—and accurate. Most credit teams aren’t short of information. They’re short of attention.

Here’s what the legacy workflow often looks like:

• 2 hours collecting documents and data from different places
• 6 hours spreading financials (PDFs → Excel → corrections)
• 8 hours researching, writing, and formatting a credit memo

By the time analysts reach the exposures that actually move portfolio risk, they’re tired. And tired analysts don’t make sharper calls—they make safer calls, slower calls, or copy-paste calls.

“The top 20% of cases drive 80% of portfolio risk.” When judgment time is rationed, the riskiest decisions get the least cognitive capacity.

Why “more data” can make risk worse

More data increases work unless you also change the workflow.

Many organisations responded to complexity by adding more sources (adverse media, ESG, industry reports, alternative data). The intention was good. The outcome was predictable: more tabs, more screenshots, more manual stitching.

That creates a dangerous illusion: teams feel thorough because they touched many sources. But the actual decision quality depends on whether the underwriter had time to:

• challenge assumptions
• pressure-test covenants
• model downside scenarios
• spot inconsistencies in management narratives

You can’t scale that by demanding longer nights.

What TrustPlus AI gets right: automate prep, keep humans accountable

The most useful AI in risk management isn’t the one that “decides.” It’s the one that prepares the decision and makes it auditable.

The e27 article describes TrustPlus AI (deployed in a first phase at PayPal) as compressing prep time from 16 hours to under 2, by automating:

• financial spreading across formats, languages, and country standards (reported 95%+ accuracy)
• adverse media review across hundreds of sources
• industry and business model analysis
• credit memo drafting aligned to an institution’s internal templates
• audit trails that log sources and human overrides

The inversion is the point: AI handles prep; humans spend ~80% of time on judgment.

That’s the core theme in our “AI dalam Insurans dan Pengurusan Risiko” series: AI should raise decision quality by improving risk signals, governance, and process discipline—not by removing human responsibility.

“Process trust before outcome trust” is the only sensible approach

Financial services (and insurance) don’t get to “move fast and break things.” One wrong call can cascade into losses, compliance breaches, or reputational damage.

A practical AI underwriting approach needs:

• Complete audit trails (what sources were used, when, and how)
• Explainability with citations (page numbers, URLs, document references)
• Human approval gates (AI drafts, humans sign off)
• Security posture aligned to enterprise expectations (e.g., SOC 2; privacy compliance)

This is also where many “AI pilots” die. They optimise for speed demos, not for trust design.

Why this matters to Singapore SMEs (yes, even for marketing)

SMEs don’t need AI to generate more content or more reports. SMEs need AI to reduce preparation time so decisions happen faster—and better.

Underwriting teams waste hours on data prep. SMEs do the same in different forms:

• exporting leads from forms, ads, and chat
• cleaning spreadsheets
• manually tagging enquiries
• summarising sales calls
• writing “first drafts” for proposals, email sequences, or campaign briefs

When prep consumes the week, leadership decisions get pushed to late nights. And that’s when you default to familiar channels (“just boost a post”) instead of making clear calls on:

• which segment to prioritise
• which offer is actually converting
• which funnel step is leaking revenue

The lesson from AI underwriting: automation is most valuable when it restores time for human judgment.

The SME translation: use AI like a sous-chef, not the head chef

The article compares AI to a kitchen sous-chef: prep is automated so the chef can focus on taste and execution.

In SME digital marketing terms, AI should:

• prepare the “mise en place” (clean data, summarise insights, draft assets)
• keep a traceable workflow (what changed, why, by whom)
• leave the final call to you (pricing, positioning, targeting, budget)

If your AI stack only produces more outputs (more captions, more designs, more dashboards) but doesn’t reduce the work of deciding, it’s noise.

A practical framework: “Judgment-First Automation” for risk and growth

If you want AI to improve outcomes, design it around decision moments, not around data volume.

Here’s a framework I’ve found works across underwriting, insurance operations, and SME marketing.

1) Identify your highest-stakes decisions (the 80/20)

Underwriting has the “top 20% exposures.” SMEs have equivalents:

• which 2–3 channels drive qualified leads
• which industries (or buyer roles) close fastest
• which products have the healthiest margin after fulfilment

Write down five decisions you wish you made faster, with more confidence.

2) Map the preparation chain (where time disappears)

Before buying tools, document the prep steps:

• Where does data originate?
• Who cleans it?
• Who reconciles conflicting numbers?
• How many handoffs happen before a decision?

Most SMEs discover they’re not lacking tools—they’re lacking a workflow.

3) Automate prep with auditability

This is the underwriting lesson you should copy directly.

For SMEs, “auditability” can be lightweight but real:

• keep source links to campaign reports and CRM fields
• store call summaries with timestamps
• track prompt versions and what was edited by humans

If you can’t trace why you changed a budget, a targeting rule, or a lead scoring threshold, you’ll repeat mistakes.

4) Create human approval gates

AI shouldn’t publish, spend, or approve without a human gate—especially in regulated or reputationally sensitive categories (finance, insurance, healthcare).

Examples:

• AI drafts ad copy → marketing lead approves
• AI flags anomalies in conversion rates → growth owner decides actions
• AI drafts a proposal → sales lead adjusts scope and pricing

This keeps accountability clear and prevents “automation drift,” where the system slowly becomes the decision-maker by default.

Common questions SMEs ask (and the straight answers)

“Will AI replace underwriters, risk managers, or marketers?”

It’ll replace tasks, not ownership. The organisations that win will keep humans responsible for decisions, while AI handles prep and monitoring.

“Is speed actually safer?”

Speed isn’t automatically safer, but prepared speed is. If AI reduces clerical effort and improves evidence trails, you get faster cycles and better governance.

“How do we avoid AI hallucinations in risk work?”

Treat AI outputs like drafts, require citations, and keep human gates. If a system can’t point to sources, it shouldn’t be used for approval decisions.

Where AI in underwriting and insurance is heading in 2026

The next wave isn’t “bigger models.” It’s workflow-native AI that fits how risk teams actually operate.

Expect more platforms to compete on:

• integration into approval hierarchies and exception logic
• evidence trails that satisfy compliance and audit
• configuration without custom code (so you’re not paying consultants forever)
• multilingual and multi-format document understanding

This is especially relevant across ASEAN, where cross-border trade, multi-currency operations, and varied reporting standards are normal—not edge cases.

What to do next if you’re an SME leader

If you’re exploring AI for marketing or operations, copy the underwriting playbook:

1. Pick one workflow where prep eats time (lead qualification, proposal writing, reporting).
2. Measure baseline time spent per week (be honest).
3. Automate preparation first (collection, cleaning, summarising, drafting).
4. Add an approval gate and a simple audit trail.
5. Reinvest saved hours into higher-quality decisions: segmentation, offer testing, sales enablement.

Most companies get this wrong by using AI to produce more “stuff.” The better move is to use AI to protect judgment, because judgment is where profit (and risk control) actually lives.

If AI can help underwriters leave the office before 11 PM and still make better credit calls, it can help SMEs stop running marketing on fatigue too. The question worth asking isn’t “How much can AI generate?” It’s: Which decision do you want your team to have the time and clarity to make this quarter?