AI automation strategies: collections, AML, compliance and risk

Key points

• Start where the work repeats: matching, chasing, screening and keying done hundreds of times a month, and choose work whose output can be checked.
• Fix cash application first: matching payments to invoices with partial or missing remittance clears unapplied cash and gives a truer DSO.
• Score payment risk, not just age, and let the system run reminders and escalation while people handle the difficult conversations.
• In AML, pre-fill onboarding files and risk-rank alert queues; AI can draft a suspicious activity narrative, but the decision to report stays with the MLRO.
• Map each regulatory change to the policies and controls it touches, and keep one version-controlled policy library behind any AI assistant.
• Build early-warning signals from payment behaviour, filing delays and director changes, and model scenarios rather than a single base case.
• Treat AI like any other model: document, validate, monitor for drift, log what it saw, and check high-risk uses under the EU AI Act (Annex III).