Six deployments from live portfolio operations across the UAE and Saudi Arabia. Client names are withheld under confidentiality. Every metric carries a locked definition, cohort and attribution method, shared in diligence — and where a client ran its own test and control, the comparison is theirs, not ours.
A major UAE bank was running early-bucket collections the way the industry always has: a large dialling floor, a panel of agencies behind it, and no way to tell which of the two was actually producing the recovery. Contact rates were falling every quarter as borrowers stopped answering unknown numbers.
Rising cost per resolved account, provisions building on accounts that were reachable in principle, and a conduct exposure the bank could not evidence, because the record of what was said on a call lived in an agent’s note field.
Not with a pilot ClearGrid graded itself. The bank ran a champion-challenger proof of value: a slice of the book split between the AI cohort and its existing process, conversation rules co-designed with its own Financial Remediation, Compliance, Legal, Quality and Digital teams, and its own KPI thresholds set in advance with pass or fail defined before anything went live. ClearGrid reported the AI cohort. The bank compared the two itself, on its own dashboard.
Borrower engagement doubled. Recovery improved by 30%. Cost to collect was cut roughly in half. Every conversation, AI and human, arrived reviewable and scored.
A growing early-stage delinquency book sat with incumbent agencies. Contact was call-heavy and untargeted. Roughly 3% of answered conversations converted to a payment or promise.
Every month of low conversion pushed accounts deeper into delinquency, where recovery gets expensive and provisions grow. The agencies' answer was more dialling, billed at the same commission.
The lender moved a portfolio slice to ClearGrid Allocate. The engine built a live state for every account, then worked each one through AI voice, SMS and the self-service portal, escalating to human specialists only on policy triggers, inside the lender's own rules.
AED 3M recovered at roughly AED 26K in execution cost. Answered-conversation conversion rose from 3% to 15%. Accounts resolved twice as fast as the prior setup.
A high-volume BNPL book where average balances were too small for agency commissions to make sense, and where call-center pressure was damaging repeat-purchase behaviour among recoverable customers.
Uneconomic accounts were simply not worked, rolling straight to write-off. The accounts that were worked came back resolved but resentful, and lifetime value quietly bled out of the cohort.
Messaging journeys with salary-cycle timing, micro instalment plans and two-tap self-service. Automation carried 95% of the volume at software cost, which made every account economic to work. Specialists handled hardship flags only.
Recovery uplift in the 38–50% band against the prior agency benchmark, with borrower satisfaction holding at 4.8 out of 5 across the collections experience.
Postpaid and device-financing receivables at telecom scale: millions of small balances, where the debtor is also the subscriber. Hard collection pressure was recovering dirhams while pushing customers to port out.
The lifetime value at risk on a churned subscriber routinely exceeded the balance being chased. Recovery on one line of the P&L was quietly funded by losses on another.
Reminder-first journeys with self-service plans, churn-risk signals feeding the decisioning, and pressure applied only where the account state justified it. Instalment products received instalment treatment, timed to salary cycles.
Resolution uplift consistent with the aggregate 38–50% band, at software cost per account, with complaint rates falling against the prior operation. Definitions and cohorts available in diligence.
An auto-finance arrears book heading to repossession on a schedule: tracing, towing, storage and remarketing costs stacked on a depreciating asset, applied to borrowers who had missed two instalments, not abandoned the loan.
Repossession is the most expensive way to collect. Each avoidable case burned recovery margin and destroyed a customer relationship the lender had paid to originate.
Day-one arrears journeys: reminders and payment links in the first 48 hours, restructure and plan-adjustment offers checked against policy eligibility, skip tracing integrated before any field action, and human escalation on hardship.
The engine resolved the majority of early-arrears cases before field remedies were triggered, with resolution running twice as fast as the prior process and repossession reserved for the cases that actually warranted it.
Static strategies decay. A contact plan that worked at launch drifts as portfolios, seasons and borrower behaviour shift, and nobody notices until the quarter closes soft.
The gap between a tuned strategy and a stale one is invisible on any single account and expensive across a book.
Every outcome, payment, promise, broken promise and dispute, feeds back into treatment strategy continuously. Reachability windows sharpen, offers calibrate, channels re-weight per segment, automatically.
Across live deployments, the continuous-learning loop has added 2 to 7% incremental recovery on top of the initial automation lift, compounding as portfolio history deepens.
Want your portfolio modelled against these benchmarks? Request a sample portfolio analysis. No call required.
Pilot on a slice, measured against your current benchmark, with definitions agreed up front.