Case Study

Global Bank: AI-Designed Workflow. Zero AI at Runtime.

The Client

Wealth advisory team at a leading global bank. 100+ clients per advisor. IT policy prohibits AI in production systems.

The Problem

Five problems. One hard constraint.

Manual client segmentation

Inconsistent service standards across advisors. Same-revenue clients getting wildly different service levels.

Blind capacity utilization

Advisors had no visibility into how many hours their book actually required versus what they could deliver.

Clients falling through cracks

No automatic follow-up triggering. Service touchpoints tracked in heads, not systems.

Decision fatigue

Every tier assignment was a judgment call. Advisors spent cognitive load on classification instead of clients.

The Hard Constraint

NO AI. NO ML. NO cloud services in production. Bank IT policy — non-negotiable.

The Method

Use AI to build something that doesn’t need AI.

Modeled the implicit rules

Used Claude to extract and formalize the bank’s unwritten segmentation logic. Rules that lived in advisors’ heads, not in any system.

Tested against 3+ years of history

Validated multi-dimensional scoring frameworks against actual historical data. The model had to match reality, not theory.

Pressure-tested edge cases

High-revenue/low-trust clients. Low-revenue/high-referral clients. Every edge case that breaks naive segmentation models.

Built purely deterministic output

Excel formulas + VBA macros. No API calls. No cloud dependencies. No AI at runtime. Every calculation auditable by compliance.

The Solution

A complete client management system. In a spreadsheet.

01

6-tier classification

WM: Premier / Core / Standard. IM: Engaged Strategic / Engaged / Service / Transition. Clear, defensible categories.

02

Multi-dimensional behavioral scoring

Implementation & Trust, Responsiveness, Energy Alignment + 5 weighted variables. Revenue alone does not determine tier.

03

Premier Cap Rule

Revenue alone is not enough. Behavioral scores must meet threshold. Prevents high-revenue, low-engagement clients from consuming Premier resources.

04

Automatic service calendars

Touchpoint schedules generated per tier. No more guessing when to call. No more clients forgotten for 6 months.

05

Capacity analysis

Hours per week by tier with breach detection. Advisors know exactly when their book exceeds capacity.

06

Monthly dashboard

Tier distribution, pending tasks, capacity alerts. Management visibility without a BI platform.

The Results

95%+

Prediction accuracy vs. 3+ years of historical advisor behavior

0.5 hrs

Capacity analysis precision — identified over-capacity advisors within half an hour per week

Zero AI

Inside the deployed system. Bank IT approved without exception.

100%

Auditable. Every tier assignment traceable to explicit scoring rules.

The Takeaway

AI is a design tool, not just a runtime tool.

We used AI to build something that doesn’t need AI. The bank’s compliance team can audit every decision. Every score is reproducible. Every tier is explainable. That’s not possible with ML models — but it IS possible with AI-engineered deterministic logic.