Consider a regional bank where SME credit decisions take long enough that the bank is losing the SME relationship to faster competitors. Manual underwriting is the bottleneck. Pilots of automation have stalled at the model-risk management gate. Use this playbook to frame the work, sequence the analysis, and avoid the common traps in a brief like that.
The consulting question
SME underwriting sits between two pressures. The commercial team wants speed — customers compare days-to-decision and the bank that takes nine days loses to the bank that takes one. Risk and compliance want defensibility — every automated decision needs a model-risk pedigree, an audit trail, and a credible explanation under supervisory review. Pure automation is regulator-blocked in most markets; pure manual underwriting is commercially uncompetitive.
How to approach it
Segment the portfolio honestly. A meaningful share of SME applications are obvious-yes or obvious-no decisions where the model output is genuinely trustworthy. A smaller cohort is on the boundary and needs human review. A third cohort is genuinely complex and needs senior underwriting. Build three lanes, not one.
Automate the obvious lane. Production-grade scoring with full model-risk pedigree — challenger models, monitoring, recalibration cadence, and a documented governance pack the regulator can stress-test on a day's notice.
Human-in-the-loop on the boundary. The decision system flags decisions where the model confidence is genuinely uncertain and routes them to underwriter review with the model's input visible. The underwriter signs the decision.
Senior judgment on the complex lane. Genuinely complex SME credit — bespoke covenants, sector concentration, founder risk — stays with senior underwriting. The model contributes data, not the decision.
Suggested workplan
Months 1–3: Portfolio diagnostic and segmentation. Model design, MRM (model-risk management) framework alignment with the bank's existing standards.
Months 4–9: Build challenger models, set up monitoring infrastructure, write the governance pack. Pilot on a low-risk segment first.
Months 10–14: Scaled rollout. Underwriter team retraining to operate in the new three-lane structure. Regulator alignment on the live system.
Questions to pressure-test
- What percentage of SME applications are genuinely automatable by today's regulatory standard?
- Where does the current decision time get lost — data, underwriter capacity, queueing, or covenant review?
- Is the model-risk function ready to govern a production model, and if not, what does it need first?
- What is the bank's stance on adverse-selection risk if competitors automate faster?
- Who owns model monitoring after launch, and do they have the staffing?
A strong answer includes
SME decisions move from days to hours in the automatable lane. Decisions in the boundary lane move from days to a single day. Senior underwriters spend their time on the complex credits that actually need them. Default detection improves because the model surfaces signals the manual review used to miss. The regulator is comfortable enough with the governance pack that the bank can expand the automation perimeter rather than defending it.
Common traps
Model-risk management is the binding constraint, not modelling skill. Banks that get this right invest in MRM capability before scaling the model.
The underwriter team changes shape, not just size. The lane structure means fewer junior underwriters and more senior ones. Plan the transition.
Adverse selection is real. If competitors get faster, the bank that stays slow attracts the riskier end of the SME market.
Regulator alignment is part of the project. Bring the supervisor into the design conversation early. Surprise launches do not end well.
Use this playbook when a bank wants faster SME decisions but model-risk governance is the constraint. It helps consultants structure automation around explainability, controls, and human review.