Anomaly Detection Isn’t Enough

AI-native ERP platforms increasingly promote anomaly detection as a major competitive advantage.

And appropriately so.

Modern systems can now identify:

• unusual transactions

• suspicious spending

• abnormal variances

• irregular payment activity

• operational outliers

That capability is valuable, particularly as finance functions become more automated and transaction volumes increase.

But for many SMEs, there is another problem that receives far less attention: similarity.

Not what suddenly changes —but what quietly never changes at all.

In founder-led businesses, stale assumptions often create more operational distortion than obvious anomalies.

I recently worked with an online retailer operating through Shopify. At first glance, the financial data appeared relatively stable. Nothing stood out dramatically. Gross margins were within expected ranges. Inventory reporting looked consistent. No major operational red flags appeared.

But underneath the surface, a more subtle issue had developed.

SKU-level COGS values had effectively been copied forward year after year across product variants and seasonal cycles. The same cost assumptions remained embedded in the system despite changing sourcing conditions, freight costs, pricing structures, and inventory realities.

No anomaly-detection system flagged the issue because nothing spiked.

The problem was the opposite: the data looked suspiciously unchanged.

The operational consequences were significant:

• distorted margin reporting

• inaccurate product profitability analysis

• flawed pricing decisions

• misleading inventory economics

• unreliable forecasting assumptions

This is extremely common in SMEs.

Not because founders or operators are careless.

Because businesses move quickly.

Teams are lean. Reporting processes evolve incrementally. Data maintenance gets delegated operationally. Copy-paste becomes a practical survival mechanism during periods of growth.

Over time, assumptions become institutionalized quietly inside systems:

• static vendor costs

• outdated labor assumptions

• stale inventory valuation logic

• legacy pricing structures

• unchanged forecasting drivers

Eventually, management teams begin making decisions based on data that appears stable but no longer reflects operational reality.

This creates a critical challenge for AI-native finance tooling.

Most anomaly detection systems are designed to identify volatility.

But many SME finance problems stem from hidden inertia.

That is where similarity detection becomes strategically important.

In practical terms, similarity detection asks different questions:

• What operational assumptions have not changed despite changing conditions?

• Which values appear mechanically repeated?

• Where does historical data remain static beyond reasonable probability?

• Which reporting trends look artificially smooth?

For growing businesses, those questions can be just as important as traditional anomaly alerts.

The broader issue is that finance systems increasingly shape operational decisions:

• pricing

• inventory purchasing

• margin analysis

• capital allocation

• forecasting

• staffing

• procurement

When stale assumptions become embedded inside those systems, businesses gradually lose decision quality without immediately realizing it.

This is especially relevant as finance functions become more AI-assisted.

Automation improves speed dramatically. But faster systems still require trustworthy underlying assumptions.

AI can automate workflows. It can surface patterns. It can accelerate visibility.

But it cannot independently determine whether the underlying operational logic still reflects commercial reality.

That still requires experienced financial judgment.

In many ways, this is where the future CFO role becomes more valuable rather than less.

As systems automate transactional processes, finance leadership increasingly shifts toward:

• validating assumptions

• interpreting operational patterns

• designing financial controls

• improving decision infrastructure

• connecting financial outputs to real-world commercial behavior

The strongest finance systems will ultimately combine:

• automation

• anomaly detection

• similarity detection

• operational context

• human judgment

Because sophisticated finance is not simply about detecting what changes unexpectedly.

It is also about recognizing when the business has changed — but the assumptions inside the system have not.