Real-time SME credit decisioning at a Tier-1 GCC bank.

At a glance.

All client identifying details are anonymised under NDA. A senior lead reference call is available to qualified prospective clients on request.

The bank is one of the largest universal banks in the GCC by SME book. Its SME credit process was a sequential, document-heavy workflow that took on average eleven working days from application to decision. The bank's commercial leadership had identified SME as a strategic growth segment but was losing applications to faster regional and digital-first competitors. The chief risk officer was unwilling to compress the cycle through any approach that softened the underwriting standard or moved decisioning outside her direct governance.

The bank had attempted an internal automation programme two years prior and had stopped it after the model layer was found to be drifting in ways the risk function could not explain to the regulator. The constraint set was therefore unusually tight: every decision had to be reconstructible from source data, every model input had to be tested against the bank's existing credit policy, and the chief risk officer required a kill-switch she could exercise without engineering involvement.

We deployed a twelve-person pod for seven months, reporting jointly to the chief risk officer and the head of SME banking. We did not build a black-box scorecard. We built a structured decisioning system in which every model input is named in the bank's existing credit policy, every decision generates a human-readable explanation that maps to the policy paragraphs invoked, and the chief risk officer has a single-control kill-switch over the automated path. The bank's underwriters were repositioned as exception-handlers for the ten to fifteen percent of cases the system declines to auto-decision, and the underwriter feedback closes the loop on calibration. The system is governed by a fortnightly committee chaired by the CRO and attended by two pod engineers.

Ninety-two percent of SME applications are now decisioned within four working hours, against the prior baseline of eleven working days. Expected loss on the SME book has not deteriorated — within the noise of normal portfolio movement, it is flat. Qualified SME applications per month have increased two-point-four-times since launch, driven principally by the bank's commercial team being able to commit to fast turnaround at the point of pitch. The regulator's most recent thematic review on credit governance cited the system favourably.

Three. First, in regulated credit, explainability is not a UX feature — it is the entire value proposition; a fast decision the CRO cannot defend to the regulator is worth zero. Second, the kill-switch is what makes the speed safe; the bank ran with a one-percent shadow path for the first eight weeks. Third, repositioning underwriters as exception-handlers is the right organisational answer, but only if the feedback loop from their exceptions is engineered into the system from day one.