AI Payment Analytics: Spot Volume Drops Before Revenue

Adyen’s shares dropped 15% on Feb 12, 2026 even though the company reported 21% net revenue growth (constant currency) for the second half of 2025. The problem wasn’t revenue. It was payment volume.

That’s a useful wake-up call for any business that touches payments—especially in Singapore, where cross-border eCommerce, omnichannel retail, and travel-linked spending make volume swings show up fast. Revenue can look healthy while demand is quietly softening underneath. When the market finally notices, it’s already late.

In this instalment of the AI Business Tools Singapore series, I’m using the Adyen story as a cautionary tale: if you’re not applying AI to your payment and commerce signals, you’re probably reacting to financial risk after it hits—rather than seeing it early and steering around it.

Why payment volume matters more than “good revenue”

Payment volume is the truth serum of commerce. It measures customer behaviour directly—how often people buy, how much they spend, and how that changes week to week.

In Adyen’s update, processed transaction volumes rose 19% to €745 billion in H2 2025, but that missed forecasts of €771 billion. Analysts noted that higher fees per transaction helped offset the shortfall, but sentiment still sank.

Here’s the business lesson: revenue can be “managed” by pricing, fees, and mix. Volume is harder to disguise. If volume is soft, it often signals:

• Demand is slowing in certain regions or categories
• Checkout or payment acceptance is underperforming
• Fraud and false declines are suppressing conversions
• Competitive offers are pulling spend away
• Customer experience issues are increasing cart abandonment

For Singapore teams, this matters because many companies are exposed to regional volatility (FX, shipping constraints, platform policy changes, tourism cycles). Volume is usually the earliest indicator that something’s shifting.

A simple way to think about it: volume is the leading indicator

Most finance dashboards treat payments like an accounting output. That’s backwards.

Payments are behavioural data. If you monitor them like product analytics, you can detect downturns earlier—and fix operational causes that revenue reporting can’t reveal.

What Adyen’s results say about the payments sector in 2026

The Adyen story also fits a broader pattern: payments companies and merchants are operating in a market where growth still exists, but it’s uneven—and expectations are unforgiving.

From the Reuters details carried by CNA:

• Adyen forecast 20–22% revenue growth for 2026
• It expects core profit margin to exceed 55% by 2028 (vs 53% in 2025)
• Unified commerce remained strong: in-store terminal transactions rose 26% YoY to €173 billion in H2
• Starbucks expansion across Europe will roll out payment solutions in 943 stores

So why the sell-off? Because when investors are nervous about the sector, they overweight signals that suggest demand isn’t keeping up. Volume miss + cautious guidance becomes a story about “the ceiling on growth.”

For operating businesses, you don’t need to care about the share price to care about the underlying lesson:

If you wait for monthly revenue reporting to tell you there’s a problem, you’re already behind your customers.

Unified commerce is growing, but it’s also more complex

Adyen highlighted strong in-store growth. That tracks with what many retailers see: customers mix online research with offline buying, and loyalty programmes blur the line further.

But unified commerce also creates more places for volume to leak:

• A payment method works online but fails in-store (or the reverse)
• Fraud rules differ by channel, creating inconsistent approval rates
• Refunds and exchanges distort net volume if not tracked cleanly
• Terminal or gateway outages hit specific locations and skew results

This is where AI business tools in commerce earn their keep—by linking operational signals to financial outcomes.

AI insights that could flag a volume downturn early

AI doesn’t magically “predict the market.” What it does well is detect patterns humans miss when the data is messy, fast-moving, and spread across systems.

Adyen’s finance chief mentioned that payment volumes are an advantage for training AI because it’s not only about having data, but having it structured and usable in real time.

That’s exactly the point. Most businesses in Singapore already have the data—they just don’t have it organised into a model that supports decisions.

What to monitor (and what AI should be trained to explain)

If you want early warning on payment softness, build monitoring around leading indicators:

1. Authorisation rate by payment method (Visa/Mastercard/PayNow/GrabPay/BNPL)
2. False decline signals (customers retrying, switching cards, or abandoning)
3. Drop-off by checkout step (shipping → payment → 3DS → confirmation)
4. Fraud rate vs approval rate trade-off (tight rules can “protect” you into lower sales)
5. Refund velocity and dispute rate (often rises before churn becomes visible)
6. New vs returning customer volume (returning can mask new-customer softness)
7. Geo and time-of-day anomalies (especially for regional expansion)

AI should not only alert. It should answer:

• What changed? (which segment, channel, payment method)
• When did it start? (pinpoint the first deviation)
• What’s driving it? (likely contributors, ranked)
• What should we do next? (recommended experiments)

A practical example (Singapore omnichannel retail)

Say your revenue is up because you increased average order value via bundles. Great. But your payment volume is flat.

An AI anomaly model might reveal:

• PayNow success rates dropped at peak hours due to timeout issues
• In-store terminals at 8 outlets show higher “reversal” events
• A fraud rule update increased 3DS challenges for Malaysia-issued cards, raising abandonment

Without AI, these look like isolated operational annoyances. With AI, they become one coherent story: volume leakage in high-traffic segments.

Agentic AI in commerce: useful, but only if you set guardrails

Adyen said it’s in “deep discussion” with retailers about agentic AI in commerce—systems that can take actions, not just provide insights.

I’m bullish on this, but I’m also strict about where teams get burned: automation without governance creates expensive mistakes in payments.

What agentic AI can safely do first

Start with low-risk actions that improve volume without touching pricing or compliance-sensitive decisions:

• Auto-route transactions to the best-acquiring path based on approval rate and cost
• Recommend fraud rule changes with human approval
• Trigger incident workflows when approval rates drop (Slack/Jira/ServiceNow)
• Personalise checkout payment method ordering based on user history
• Detect and suppress bot-driven checkout attempts to protect approval rates

What should stay human-led (at least initially)

• High-impact pricing changes tied to payment costs
• Changes to dispute handling policies
• Risk appetite decisions (fraud vs conversion) without sign-off
• Regulatory and data residency decisions (relevant for Singapore + regional operations)

A good rule: let AI propose and simulate; let humans approve.

A 30-day AI plan to stabilise payment volume (Singapore-friendly)

If you’re running eCommerce, retail, subscriptions, marketplaces, or cross-border services, here’s a realistic one-month plan I’ve seen work.

Week 1: Build a “volume truth” dashboard

Define a single view of:

• Gross payment volume (GPV)
• Approval rate (by method, channel, issuer country)
• Checkout conversion rate
• Fraud and chargebacks
• Refunds and disputes

The key is consistency: one set of definitions, one cadence.

Week 2: Add anomaly detection

Use AI (or even simpler statistical detection at first) to alert on:

• Approval rate deviations
• Spikes in retries
• Sudden 3DS challenge increases
• Outlet-level terminal issues
• Geo-specific drop-offs

Make alerts actionable: every alert should include “probable cause” fields.

Week 3: Run 2–3 controlled experiments

Examples:

• Reorder payment methods at checkout for mobile users
• Adjust fraud thresholds for a single segment with high false declines
• Offer one-click payments to returning customers

Measure impact in volume, not just revenue.

Week 4: Operationalise the loop

Turn insights into repeatable workflows:

• Weekly payments performance review (Ops + Finance + Product)
• A/B testing pipeline for checkout changes
• Playbooks for approval-rate dips

This is where AI business tools stop being “reports” and become a management system.

What Singapore businesses should take from Adyen’s 15% drop

Adyen didn’t get punished because it wasn’t growing. It got punished because volume growth missed expectations, and the outlook didn’t calm nerves.

For operators, the takeaway is more practical than financial:

If your organisation tracks revenue but can’t explain volume, you’re flying on instruments that lag reality.

The payments layer is now one of the richest sources of real-time customer behaviour. Treat it that way. Build AI around it. And make sure the insights reach the teams that can fix issues—product, risk, operations, and customer experience—not only finance.

If you’re following the AI Business Tools Singapore series, this post sits in the “operations meets growth” lane: the companies that win in 2026 will be the ones that connect AI to day-to-day decisions, not quarterly narratives.

Where are you most exposed right now—approval rates, fraud friction, or checkout drop-off—and do you have the data to prove it within one business day?