The Gap Between The Moats

Most banks run a handful of vendors across the decision pipeline: one for fraud scoring, another for KYC, a third for compliance screening and a fourth for credit decisioning. Each vendor claims that data is their competitive advantage. Each operates on their own slice of the customer. None of them sees each other’s slice.

It’s not because vendors refuse to cooperate. The bank receives each vendor’s output—the fraud score, the identity match and the KYC result. But outputs are summaries. The fraud vendor’s score doesn’t tell you which device signals or velocity patterns drove it. The identity vendor’s match doesn’t expose the raw biometric confidence or the document anomalies it weighed. Each vendor’s real data—the features and signals underneath the score—is what they call their moat, and almost none of it crosses vendor boundaries.

So the bank is left making decisions based on a set of summaries, each computed against different data at a different moment in time, none reflecting what the other vendors knew.

Who Pays For The Gaps

When a vendor’s model makes a bad call because it lacked data that was sitting inside another vendor’s system, the vendor doesn’t eat the loss. The bank does.

One of the bank strategists at that table told me about a case at their institution. No single vendor’s data looked alarming on its own; the transaction amounts were normal, the account wasn’t flagged and the compliance checks passed. But if you combined the signals across vendors—a new payee pattern from the payment system, a device change from the fraud vendor, a recent profile update from KYC—it looked like a textbook account takeover. No one vendor had enough to raise a flag. Together, the data would have stopped the transaction. The bank ate the loss.

The Metric Few Banks Measure

Banks evaluate vendors on model accuracy, coverage and latency. A fraud vendor can report strong precision and recall against confirmed fraud and chargebacks, and those numbers are real.

But when a vendor misses something, they analyze the miss against their own data. They retrain and improve their features all within their own silo. What they can’t discover is whether the miss would have been caught if they’d had another vendor’s data. That case gets filed as “hard to detect” rather than “missing data from another system.” Nobody asks the cross-vendor question because no vendor has the ability to test it.

Why Sharing Doesn’t Fix It

The obvious response is to just get the vendors to share. But sharing what? Vendors will expose scores, reason codes and maybe some feature importance. The raw signals and proprietary features underneath—the part that would actually help another vendor’s model—are exactly what they consider their moat.

Even if you could get full access to every vendor’s raw signals, there’s a second problem. Each vendor’s signals were captured at a different moment. No system at the bank is reconciling them into a consistent view of the customer before the decision fires. The orchestration layer treats each vendor’s output as independent.

So there are two gaps. Most banks can’t get the raw data, and even if they could, they have no infrastructure to unify it in real time. Adding more vendors doesn’t fix this. Each new vendor adds another partial view, another integration and another seam where information is lost or stale by the time the decision runs.

Who Should Own The Picture

After the roundtable, one of the bank strategists pulled me aside: “We spend all this time evaluating each vendor’s accuracy, and nobody’s ever asked me what accuracy we get when we put them all together.”​

That leads to the question banks should be asking: not which vendor has the best model, but who owns the complete picture of the customer. Right now, no one does. The bank outsourced it to vendors who, by design, only see their own data.

The fix isn’t adding another vendor; it’s leadership in banking taking ownership of the unified customer view. That requires three shifts:

1. Renegotiate contracts to secure access to raw signals—not just scores—for any data that influences cross-vendor decisions.

2. Build a bank-owned context layer that ingests and reconciles those signals into a single, real-time customer state that models can query before decisions are made.

3. Measure performance at the system level—evaluating how well decisions improve when signals are combined—alongside traditional per-vendor metrics like precision and recall.

​The biggest obstacle is organizational. Risk, fraud and compliance each own their vendor relationship; no one has a mandate over the seams. Institutions making progress name a single owner under the CRO or CDO, accountable for the unified customer-state layer with authority to set data-access requirements across vendors. Without that ownership, the gaps keep getting filed as "hard to detect," and the losses stay on the bank's books.

Every vendor at that table told me data was their moat, but the bank was the one paying for the distance between them.​​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?