Part 3 of our "CASP Compliance Toolkit" series. The BWRA sets the firm's risk appetite; customer risk scoring applies it to each individual. It is the model that decides who gets enhanced due diligence, whose transactions are watched closely, and who sails through — and in crypto it has a second half that traditional scoring lacks.
A customer risk score is a single number (or band) summarising how much AML risk a customer represents, used to drive the intensity of due diligence and monitoring. In traditional finance it is built from off-chain factors alone. In crypto, the customer's on-chain behaviour is observable in a way a bank customer's cash habits never are — and a score that ignores it is leaving the most predictive data on the table.
The Off-Chain Half
The familiar inputs, drawn from KYC and the customer relationship:
- Identity & status — individual vs entity, PEP status, adverse media, sanctions screening result
- Geography — jurisdiction of residence and nationality, exposure to high-risk countries
- Profile — declared occupation, source of wealth, expected activity level and purpose of the account
- Product use — which services the customer uses, and whether that fits their declared profile
The On-Chain Half
The crypto-specific inputs, drawn from the customer's actual on-chain behaviour — the data that makes a crypto score sharper than a bank's:
- Counterparty exposure — the risk profile of the addresses the customer transacts with: exchanges, mixers, sanctioned clusters, darknet, high-risk services
- Source-of-funds quality — how clean the provenance of incoming funds is, by hop distance to known-illicit origins
- Behavioural patterns — use of privacy tools, cross-chain hopping, structuring-like activity, dormancy-then-burst
- Wallet history — the age and history of the addresses involved, and whether they connect to prior risk
- Consistency — whether observed on-chain activity matches the declared profile, or contradicts it
The on-chain half is the half that updates itself
Off-chain factors are mostly static — a customer's declared occupation doesn't change daily. On-chain factors move constantly, which is what makes a crypto risk score genuinely dynamic: a customer who scored low at onboarding climbs as their counterparty exposure deteriorates, before any off-chain factor changes. A scoring model that only refreshes when KYC is renewed misses exactly the risk that develops mid-relationship. Wire the on-chain factors as live inputs, not a onboarding snapshot.
Making It Explainable
Whatever the model, each factor's contribution has to be documented and defensible — an examiner will ask why a customer is rated where they are, and "the model said so" is not an answer. The weights, the thresholds, and the reason a given customer sits in a given band all have to be reconstructable. This is as true for the on-chain inputs as the off-chain ones: "elevated because 30% of inbound is within two hops of a high-risk exchange" is explainable; an opaque score is not.
How BA helps. BA supplies the on-chain half of the score as structured, explainable inputs — counterparty risk, source-of-funds quality by hop distance, behavioural flags, and wallet history across 80+ chains against a 1B+ label graph — that feed a customer risk model and update as behaviour changes. The off-chain KYC half stays yours; the on-chain half becomes measurable rather than assumed. For where the score drives monitoring, see Ongoing Customer Wallet Monitoring.
Next in the series — the close: Independent AML Audit, where everything in this toolkit gets tested by someone whose job is to find what's missing.
Feed real on-chain behaviour into your customer risk scores
Screen wallets, monitor entities, and generate compliance reports with 1B+ labeled addresses and 305+ data sources.
See Screening Solutions