Opal Balance: How Real Is the 2025 Provision Line

What This Follow-up Is Testing

The main article argued that Opal Balance's real test moved beyond raw growth and into funding discipline, capital structure, and the quality of earnings. This continuation isolates the second line item that is easy to misread inside that broader story: the 2025 expected credit loss line.

On first read, the number looks almost too good. Expected credit loss expense fell to just NIS 429 thousand, down from NIS 1.689 million in 2024 and NIS 12.08 million in 2023. Management's presentation builds a clear narrative around that outcome: better underwriting, fewer bounced checks, stronger collections, and a lower macro overlay. All of that is real. It is also only half the story.

The 2025 provision line is not a pure accounting trick, but it is not a clean steady-state run rate either. It was built by three forces working at the same time: real improvement in collections and restructuring compliance, lower return rates, and a lighter macro coefficient inside the general provision model. The right question is therefore not whether the line is "real" or "fake." The right question is how much of the improvement belongs to better credit quality, and how much belongs to a more favorable modeling year.

The easiest number to miss sits exactly at that junction: gross credit rose to NIS 599.0 million from NIS 431.2 million, while the total allowance stock barely moved, to NIS 29.8 million from NIS 30.1 million. That means 2025 was not a year in which credit risk disappeared. It was a year in which the book expanded quickly under a lighter allowance burden.

Where the Improvement Really Came From

Collections and workout discipline drove the specific line

The company itself explains the drop in expected credit loss expense through four factors: lower bounced-check rates, lower specific provisions as the war's impact on customers faded, successful recovery of debts that had previously been provided for at 100%, and ongoing collections from customers that kept honoring settlement arrangements even though they had been classified into groups B and C. The investor presentation says the same thing more bluntly: the specific provision declined because of ongoing collections and compliance with settlement payments, while lower return rates were attributed to tighter underwriting and portfolio refinement.

That is not just slide language. The movement table gives those collections real weight. The opening allowance balance for 2025 was NIS 30.095 million. During the year, NIS 863 thousand was added from a business consolidated for the first time, NIS 2.767 million was removed through write-offs of customer credit, and NIS 1.620 million was added through remeasurement of the allowance. That matters because it shows the improvement did not come from simply pretending the risk was gone. It came from real progress in the book alongside an ongoing need to revisit problem credits.

The specific allowance split supports the same read. By year-end 2025, the specific allowance stood at NIS 27.404 million, versus NIS 27.927 million at the end of 2024. In other words, the troubled part of the book did not disappear, but it also did not expand alongside the rest of the portfolio. That is a real improvement, but it is not the same thing as saying the risk line has become structurally irrelevant.

The macro overlay eased just as the book got much larger

The second leg of the story comes from the general provision. This is where a relatively small number materially changed the reading of the year. The add-on from the macro coefficient fell to NIS 274 thousand in 2025 from NIS 473 thousand in 2024. The company explicitly ties that decline to macroeconomic indicators and Bank of Israel forecasts that reduced the uncertainty coefficient.

But before that macro overlay, the base general allowance actually rose. The allowance before the uncertainty factor increased to NIS 2.133 million from NIS 1.695 million, simply because the book itself got bigger. Put differently, the core model did not say credit risk had vanished. It said the book was larger and therefore needed more base provision, and then the macro layer came down and offset part of that increase. The end result was a general allowance of NIS 2.407 million versus NIS 2.168 million, only about an 11% increase despite much faster book growth.

That is the core thesis of this continuation. If you only look at the annual expense line of NIS 429 thousand, you can miss that the book grew much faster than the reserve base, and that part of the gap came from a lower macro cushion. 2025 therefore supports a read of genuine improvement, but not a read in which the provision model has suddenly become immaterial.

What Still Has Not Been Settled

The good news is real. Groups B and C combined were almost flat in absolute shekels, NIS 79.9 million versus NIS 78.0 million, while the total book grew by NIS 167.8 million. That is a meaningful sign that growth was not accompanied by broad deterioration across the portfolio. The total allowance ratio also fell to about 5% of gross credit from about 7% a year earlier.

But the ageing table should stop readers from turning that into a claim that every layer of the book improved at once. Credits that were not past due rose to NIS 478.0 million from NIS 346.9 million, which is clearly positive. At the same time, deferred credits jumped to NIS 62.2 million from NIS 28.7 million, the 0-90 day bucket rose to NIS 10.6 million from NIS 4.9 million, and the 90-180 day bucket rose to NIS 3.5 million from just NIS 118 thousand. Only the heavy tail of 181 days and more declined, to NIS 44.6 million from NIS 50.7 million.

That does not mean credit quality deteriorated. It does mean the improvement came mainly from shrinking the heavy tail and collecting against older problem assets, not from every front of credit risk disappearing. Some pressure has simply moved forward inside the monitoring pipeline. The annual P&L line of NIS 429 thousand does not tell you that on its own.

There is also a real protection layer that deserves recognition. By year-end 2025, the group had 28 customers in risk levels 2 and 3 for whom recognized credit losses amounted to about NIS 7.31 million, and management says that without the relevant collateral the group would have had to recognize an additional loss of roughly NIS 15.844 million. That matters because the low loss line is not supported only by collections and macro assumptions. It is also supported by collateral realization value. From an earnings-quality perspective, that is both a strength and a reminder that some of the calm depends on asset recovery values, not just on borrower behavior.

The fourth quarter was already less flattering

One more data point keeps the annual number honest. Full-year 2025 showed expense of only NIS 429 thousand, but inside the year there was provision income of NIS 1.010 million in the second quarter and a further NIS 144 thousand of provision income in the third quarter. In the fourth quarter, the direction turned back into expense of NIS 1.336 million.

That does not prove the improvement has already reversed. It does prove that the annual number smooths over meaningful in-year volatility. Anyone taking the NIS 429 thousand line and pushing it straight into 2026 as if it were a stable annual run rate is leaning on shaky ground.

There is one more reason to stay disciplined here. The auditors explicitly identified expected credit loss provisioning as a key audit matter. Their reason was straightforward: classification across the portfolio, expected cash collections, collateral realization, and macro assumptions all require significant judgment. That is not an accusation against the number. It is a reminder that this line is especially assumption-sensitive and should not be read like a mechanical output.

Conclusion

The 2025 provision line is real, but only in a qualified sense. There is operational and underwriting improvement here that should not be dismissed: return rates were lower, collections worked, the worst tail came down, and groups B and C did not expand with the book. At the same time, the flattering picture also relied on a lighter macro overlay, quarters that included provision income, and a collateral layer that softens loss severity.

The right reading of 2025 is therefore not that Opal solved credit risk. The right reading is that Opal bought itself time. It showed that the book can grow without an immediate blow-up in the provision line, but it still has not proved that NIS 429 thousand is a clean base that can be carried forward. The real test now shifts into 2026: can expected losses stay low relative to a larger book even without the same macro tailwind and without provision income doing part of the work mid-year.