AI Metrics That Prevent ‘Good Results, Bad Stock’ Days

Adyen grew net revenue 21% in the second half of 2025. Then its shares still fell 15% in a day.

That’s not a paradox—it’s a warning label. Markets (and boards) don’t reward “growth” in the abstract. They reward growth that matches expectations on the right leading indicators. For a payments company like Adyen, that indicator was processed transaction volume: €745B vs a market expectation of €771B. Even with higher fees per transaction partly offsetting the miss, sentiment across the payments sector stayed heavy.

This matters for Singapore operators far beyond fintech. Whether you run an e-commerce brand, a retail chain, a B2B SaaS company, or a marketplace, you can absolutely post “good” headline results and still get punished—by investors, by lenders, or simply by your own cash flow. In this instalment of the AI Business Tools Singapore series, I’ll break down what the Adyen reaction teaches us about AI-driven financial forecasting, the metrics that really move outcomes, and the practical AI tooling you can put in place to avoid unpleasant surprises.

A clean one-liner to keep: If your leading indicators miss, your headline numbers won’t save you.

What Adyen’s drop really signals (and why it’s not about “panic”)

The direct answer: Adyen’s results show that volume is a market’s proxy for momentum, and momentum is what people price.

Adyen’s net revenue growth was strong, and guidance for 2026 stayed in the 20–22% growth range. It also reiterated an ambition to push core profit margin above 55% by 2028 (vs 53% in 2025). Yet the market focused on volume coming in below expectations.

Why? Three reasons you should care about:

1. Volumes are harder to “manage” than revenue. Revenue can be influenced by pricing, mix, and fees. Volume tells you whether customers are actually transacting.
2. Volumes are leading; revenue is lagging. If volume softness persists, revenue pressure usually follows.
3. Sector sentiment amplifies misses. Analysts noted that even decent numbers may not lift broader negativity in payments.

For Singapore businesses, the parallel is simple: if your demand signal softens (orders, conversions, usage, repeat purchase rate), reporting strong revenue via pricing or one-off enterprise deals doesn’t protect you for long.

The KPI mistake most teams make: tracking “what happened,” not “what’s next”

The direct answer: Most companies over-index on financial outputs and under-invest in predictive, operational KPIs.

Adyen’s story is basically a KPI hierarchy lesson:

• Output KPI: net revenue (+21%)
• Leading KPI: processed transaction volume (+19%, but below forecast)
• Operational drivers: in-store terminal transactions (+26%), customer expansion (e.g., Starbucks rollout), pricing/fee dynamics

Singapore SMEs and mid-market firms often do this in a simpler form:

• Output: monthly revenue, gross margin
• Leading: pipeline quality, conversion rate, repeat rate, basket size, churn
• Drivers: page speed, stockouts, support backlog, payment failures, fraud blocks, ad fatigue

A practical KPI hierarchy you can implement next week

If you want an “Adyen-proof” dashboard, structure it like this:

1. Demand & volume signals (leading)
   • Orders / transactions / active users
   • Conversion rate by channel
   • Repeat purchase rate / retention cohorts
2. Unit economics (stability)
   • Contribution margin per order
   • CAC payback period
   • Refunds, chargebacks, fraud loss rate
3. Financial outputs (reporting)
   • Revenue growth, gross margin, EBITDA
   • Cash conversion cycle

AI becomes useful when it connects layers 1 → 2 → 3 and tells you what will happen before month-end closes.

How AI business tools reduce “surprise misses” in volume, revenue, and margin

The direct answer: AI reduces surprises by detecting trend breaks early, forecasting with scenario ranges, and surfacing the drivers behind changes.

Adyen’s CFO told Reuters that their payment volumes are an advantage when training AI because it’s not just data—it’s structured data usable in real time. That’s the key point. Many companies have data, but it’s scattered across POS, web analytics, payment gateways, CRM, and spreadsheets.

Here are three AI applications that map directly to the Adyen-style problem.

1) Real-time anomaly detection for volume and conversion

If processed volume is your lifeblood, you need to know when it deviates from expectation today, not at month-end.

What to implement:

• An AI/ML anomaly detector that monitors transactions, approvals, conversion rate, and average order value by channel and geography.
• Alerting rules that trigger when metrics deviate from forecast bands (not just absolute thresholds).

Example (Singapore retail + e-commerce):

• If online conversion drops 12% week-on-week and payment failure rate rises, the model flags a payment routing issue or a PSP outage before revenue takes the hit.

2) Forecasting that includes “driver-based” scenarios

Most forecasting fails because it extrapolates revenue without modelling drivers. The better approach is driver-based forecasting:

• Transactions = traffic × conversion × repeat rate
• Revenue = transactions × AOV
• Margin = revenue − (COGS + payment fees + logistics + refunds)

Then use AI to forecast each driver and run scenarios:

• Base case: stable conversion, slight AOV lift
• Downside: conversion -2pp due to ad fatigue
• Upside: repeat rate +5% from loyalty campaign

This is how you avoid being surprised by a “volume miss” that was visible in micro-signals two weeks earlier.

3) Margin protection through intelligent payments and fraud controls

Adyen noted higher fees per transaction partly offset softer volumes. That’s a reminder: pricing and routing decisions matter, but they’re not free—they can affect approval rates and customer experience.

AI helps you balance:

• Authorization rate vs. cost per transaction
• Fraud prevention vs. false declines
• Refund/chargeback risk vs. growth targets

For Singapore merchants selling cross-border, this is especially relevant: small shifts in approval rates or FX-related declines can quietly erase margin.

Agentic AI in commerce: useful, but only if your controls are grown-up

The direct answer: Agentic AI can automate actions (not just insights), but you must set guardrails so it doesn’t optimize the wrong thing.

Adyen mentioned “deep discussion” with retailers about using agentic AI in commerce, and it plans to hire around 600 people while expanding internal automation.

Here’s my stance: agentic AI is genuinely helpful in operations, but most teams introduce it backwards—automating decisions before they’ve agreed on the metric hierarchy.

Where agentic AI works well (safe first steps)

• Payment routing suggestions with human approval (optimize cost without hurting approvals)
• Dispute triage: draft evidence packs, classify reason codes
• Support automation: detect payment-related complaints and open internal incident tickets
• Inventory + promo coordination: pause campaigns when stockouts spike

Guardrails you should set

• Hard constraints: never drop approval rate below X% without review
• Budget constraints: caps on promo spend changes
• Audit trails: every automated action logged with “why”
• Separation of duties: model suggests, manager approves (at least initially)

This is what “structured data in real time” is really for: safe automation with measurable outcomes.

A Singapore-ready blueprint: your 30-day AI metrics sprint

The direct answer: You can get meaningful forecasting and early-warning signals in 30 days if you focus on a narrow set of metrics and data sources.

If you’re reading this as a Singapore founder, finance lead, or ops manager, you don’t need a moonshot. You need an early-warning system that covers demand, payments, and margin.

Week 1: Decide the “market-moving” metric for your business

Pick one primary leading indicator (like Adyen’s volume):

• E-commerce: paid orders or checkout completion
• Subscription: net revenue retention or churn
• Retail: same-store transaction count
• Marketplace: successful matches / fulfilled orders

Week 2: Unify the minimum data set

Connect:

• Sales/transactions (POS, e-commerce platform)
• Payment outcomes (approval, failures, chargebacks)
• Marketing source data (spend, clicks, channel)

The win condition is a single daily table like:

• date, channel, geography, transactions, approval_rate, AOV, refunds, chargebacks, spend

Week 3: Build forecasts with confidence bands

Don’t aim for perfect accuracy. Aim for:

• 7-day and 30-day forecasts
• scenario ranges (base/upside/downside)
• driver explanations (what changed?)

Week 4: Turn insights into decisions

• Define alert thresholds tied to forecast deviations
• Assign owners: who responds to which alert
• Set a weekly “forecast vs actual” review (30 minutes)

If you do only this, you’ll already be ahead of the majority of teams relying on monthly reporting.

The bigger lesson from Adyen: metrics are strategy (and AI makes them executable)

Adyen’s results underline a harsh truth: you don’t get to choose what stakeholders focus on; you only get to prepare for it. For them, processed volume was the narrative. For your business, it might be retention, conversion, pipeline velocity, or approval rates.

AI business tools in Singapore aren’t just about automation for its own sake. Their real value is tightening the loop between signal → forecast → action, so you’re not explaining surprises after the damage is done.

If you’re building your 2026 operating plan right now (and many Singapore teams are), this is a good moment to audit your dashboards. Are they explaining the past, or protecting the future?

What’s the one leading indicator in your business that—if it missed by 3–5%—would change your entire quarter?