How we cut device-subscription default from 27% to 8%

The short answer

We fixed device-subscription default at the approval decision. At Zension Technologies (Zaam) in Dubai, I built a weighted scoring model from our own book and set explicit thresholds with an escalation path. The risk inputs got a refresh every quarter. Finance, risk and commercial signed the same thresholds, so every approval ran against one set of numbers. Subscriber default went from 27% to 8%.

Where it was going wrong

We sold devices on subscription. Every approval was a bet on 24 months of payments, and default was at 27%.

Approvals went case by case. Risk and finance each had their own view, and commercial had a target. Each call made sense on the day, and together they added up to a book with no design behind it.

The external credit checks were already there, through AECB and Credicide. They were read differently from one file to the next.

How the scoring works

I built the model from the book's own history: 128 subscriptions and 455 behavioural data points. Each application gets a weighted score from four groups of inputs:

• Credit score.
• Plan value.
• Card and payment behaviour.
• Bank and emirate risk.

Knockout rules sit on top. Some cases fail whatever the score says. I wired the AECB and Credicide checks into the same flow, so the same evidence got the same answer every time.

The history showed where the risk sat. Low-value plans defaulted at more than 10 times the rate of high-value ones. And above a certain credit score, the sample had no defaults at all.

Next to the model I kept country and area risk sheets and a demographic score, refreshed every quarter. A model built once starts drifting as soon as the market moves.

Before anyone had to trust it, I backtested it. It accepted about a third of applications and excluded every customer we already knew as a first-payment default.

Then the thresholds. I set them with an escalation path and took them through finance, risk and commercial until all three signed the same set. That part took longest.

What to check in your own book

1. Split default by plan value. In ours, low-value plans defaulted at more than 10 times the rate of high-value ones. A gap that size means plan value belongs in the score.
2. Find the credit score where default stops. Our sample had a line with no defaults above it. Know where yours sits before you set a threshold.
3. Test whether the same evidence gets the same answer. Pull approvals and declines with similar AECB results and compare them. If the outcomes differ, the decision depends on who read the file.
4. Backtest against the defaults you already know. Run the rules over past applications. If known first-payment defaults get through on old data, they'll get through on new data.
5. Ask who signed the thresholds, and when the inputs were last refreshed. You want one set signed by finance, risk and commercial, and risk sheets dated this quarter. Anything else, and approvals follow whoever is in the room.

Questions people ask

How do you reduce default on device subscriptions?

Decide the risk at approval. Build a weighted score from your own payment history, add knockout rules, and hold every approval to thresholds that finance, risk and commercial have all signed. Refresh the inputs every quarter. That's how we went from 27% to 8%.

What data goes into a subscription credit score in the UAE?

Ours used credit score, plan value, card and payment behaviour, and bank and emirate risk, with knockout rules on top. AECB and Credicide checks fed the same flow. Country and area risk sheets and a demographic score ran alongside it.

How often should a credit scoring model be updated?

We refreshed the risk sheets and demographic scores every quarter. A model you build once drifts as the market moves.

Does a lower default rate mean approving fewer customers?

In our case approval volume was protected and the commercial targets stayed the same. The model changed which applications got accepted.

The case study

The full write-up is in the risk decisioning case study. The other half of the same business, getting corporate clients signed and paying, is in B2B customer onboarding in the UAE.

If your book needs the same work, here's how I work.