Predictive Analytics for Banking and BFSI in 2026, Credit Risk in the UAE and Africa

Why this is now a board conversation

Credit risk modelling in banking and BFSI has moved from a model conversation to a board conversation across the UAE, Nigeria, Kenya, Tanzania and Ethiopia. The combination of IFRS 9 expected credit loss, the central bank model risk governance expectations and the AI Act trajectory means that the methodology, the assumptions and the fairness profile of the credit risk models are now reviewable by the regulator on short notice. The Chief Risk Officer is the named owner, and the Chief Data Officer is the named partner. The board ratifies the operating model.

The Enterprise AI Readiness Assessment Audit consistently shows that BFSI organisations rated at Strategist or Agentic Enterprise level have a documented model risk operating model, a quarterly board review of the model performance and a fairness testing cadence. Organisations at Bystander or Aspirant level have the models but not the operating model, and the gap is visible to the regulator in a desktop review.

PD, LGD and EAD, what predictive analytics improves

Probability of Default is the chance that a borrower will default in the next twelve months or over the lifetime. Predictive analytics improves PD by widening the input set, including behavioural and contextual signals such as payments velocity, channel mix and life-stage events, and by replacing linear assumptions with non-linear models that capture interaction effects.

Loss Given Default is the share of the exposure that the bank will lose after recovery. Predictive analytics improves LGD by modelling the recovery path as a sequence rather than a point estimate, and by incorporating collateral revaluation under stress scenarios.

Exposure at Default is the amount the bank expects to be exposed to at the moment of default. Predictive analytics improves EAD by modelling the customer behaviour in the run-up to default, including the credit conversion factor for revolving exposures.

The three together drive the expected credit loss under IFRS 9 and the regulatory capital under Basel. The accuracy gain from modern predictive analytics translates directly into a lower provision for the same risk appetite, or a higher risk appetite for the same provision. Either reading is a board conversation.

IFRS 9 alignment in 2026

IFRS 9 expected credit loss requires twelve-month and lifetime ECL estimates for performing and underperforming exposures respectively, with the staging transitions governed by significant increase in credit risk. Predictive analytics improves the SICR detection by replacing rule-based thresholds with behavioural and contextual signals. The improvement reduces the volatility of the staging transitions, which in turn reduces the volatility of the provision and the noise in the earnings line.

The CRO and the CFO together should ratify the IFRS 9 model methodology and the staging rules at the board, and re-ratify them at least annually or whenever the methodology changes materially.

Model risk governance

Model risk governance covers the model development, the independent validation, the ongoing performance monitoring and the model inventory. The CRO is the accountable owner. The model risk function is independent of the model development function. The validation methodology is documented, the validation outcomes are tracked and the model inventory is refreshed on a defined cadence.

Where predictive analytics is concerned, the validation methodology must extend to the data lineage, the feature engineering, the training and test split, the holdout performance, the stability over time and the fairness profile. Each of these is a regulator question waiting to be asked.

• Documented validation methodology, refreshed annually.
• Independent validation function, separate from development.
• Ongoing performance monitoring, with thresholds and escalation.
• Model inventory, refreshed on a defined cadence.
• Fairness profile, with metrics agreed with the model risk function.

Fairness testing, the rising bar

Credit decisioning models must not produce disparate adverse outcomes for protected groups beyond what the underlying credit risk justifies. The fairness testing must be documented, the metrics must be agreed with the model risk function, and the override path for the false positive and false negative cases must be exercised. The expectation is rising across the UAE, Nigeria, Kenya and Tanzania, and CROs should design to the higher baseline.

The fairness conversation is not a compliance overhead. It is a model quality conversation. A model that is unfair is also typically a model that is using a proxy variable that is masking a more accurate signal. Fairness testing surfaces the proxy and the underlying signal together.

The country lens

In the UAE, the Central Bank of the UAE and the DFSA have both raised the bar on model risk governance, and the expectation is now closer to the European baseline than the regional one. In Nigeria, the Central Bank of Nigeria is increasingly focused on model risk for IFRS 9 and Basel, and the on-site examinations have widened. In Kenya, the Central Bank of Kenya has issued guidance on model risk and is following the IFRS 9 implementation closely. In Tanzania, the Bank of Tanzania is moving in the same direction, with on-site reviews focused on the staging methodology. In Ethiopia, the National Bank of Ethiopia is on an earlier trajectory but the direction is the same.

Across all five jurisdictions, the unifying expectation is the same. The methodology, the assumptions and the fairness profile of the credit risk models must be defensible at the board and reviewable by the regulator on short notice.

How the AltaFuturis MasterClasses build the capability

The Applied AI and Predictive Analytics MasterClass takes the CRO, the CDO and the credit risk team through PD, LGD and EAD modelling with modern techniques, the IFRS 9 alignment, the model risk governance operating model and the fairness testing methodology. The Generative AI for CXOs and Business Leaders MasterClass equips the CEO and the executive committee to ratify the methodology at the board. The Adaptive Leadership in an AI-Accelerated Business Environment MasterClass prepares the executive committee to lead the organisation through the model risk operating model uplift.

Cohorts run virtual on July 16 to 18 and August 13 to 15 2026, and onsite on July 23 to 25 and August 19 to 21 2026. Early Bird pricing of USD 650 is open until 30 June 2026.

Five actions in the next week

First, take the Enterprise AI Readiness Assessment Audit and capture the Risk and Compliance and Data and Customer pillar scores. Second, commission an inventory of the credit risk models in production with the validation status of each. Third, schedule the next quarterly board review of the model performance and the fairness profile. Fourth, name the model risk operating model owner at the executive committee. Fifth, reserve seats in the July or August 2026 Applied AI and Predictive Analytics MasterClass cohort before Early Bird closes on 30 June 2026.

Recommended further reading

1. EU AI Act 2026, cross-border obligations for UAE and Africa.

2. AI model risk governance for banks in the UAE and Africa.

3. Applied AI and Predictive Analytics MasterClass.