Next best product engine¶


ID	`MOD-107`
System	SD06
Repo	`bank-risk-platform`
Build status	Not started
Deployed	No

What it does¶

MOD-107 computes a ranked list of next best products for every active customer on a weekly Snowflake run. It takes the eligible product set from MOD-105, scores each eligible product against a set of customer signals, and writes the top 3 ranked products per customer to Postgres for consumption by MOD-108 (offer engine), MOD-083 (agent assist back-office view), and MOD-077 (app insight cards).

Why it exists¶

Commercial reason: Next best product is the primary mechanism for deepening customer relationships. The bank's growth model (BG-003, BG-005) depends on expanding product holdings per customer — an NBP recommendation that surfaces the right product at the right moment drives conversion without requiring a sales team.

Conduct reason: Scoping recommendations to the eligibility matrix (MOD-105) ensures every surfaced product is one the customer can actually obtain. This directly satisfies CON-006 (product suitability) and avoids the mis-selling risk of promoting products a customer would be declined for.

Ranking model¶

Signal	Source	Weight direction
Eligible product (binary gate)	MOD-105	Must be ELIGIBLE — ineligible products excluded
Relationship gap	Products held vs products in same segment peer group	Higher gap → higher rank
Churn / health score	MOD-040	High churn risk → prioritise retention products (savings rate, loyalty offer)
Customer ROTE contribution	MOD-106	Products that improve customer-level ROTE ranked higher
Behavioural signals	Spend categories, idle cash detected, FX activity	Idle cash → savings product; regular FX → multi-currency wallet
Recency of prior offer	MOD-108 offer history	Products offered and rejected recently ranked lower (cooling-off period: 90 days)
Product lifecycle stage	Product register	Products in sunset / under-capacity ranked lower

Weights are configurable per product class in a Snowflake configuration table. Model version is recorded on every output row.

Data model¶

-- product_intelligence.next_best_product (Snowflake — written weekly)
CREATE TABLE product_intelligence.next_best_product (
  party_id            VARCHAR NOT NULL,
  jurisdiction        VARCHAR NOT NULL,
  rank                INT NOT NULL,           -- 1 = highest ranked
  product_id          VARCHAR NOT NULL,
  score               NUMBER(8,6) NOT NULL,   -- 0.0–1.0
  primary_signal      VARCHAR NOT NULL,       -- top-contributing signal name
  model_version       VARCHAR NOT NULL,
  evaluated_at        TIMESTAMP_NTZ NOT NULL,
  PRIMARY KEY (party_id, rank, evaluated_at)
);

Results are written back to Postgres (product_intelligence.nbp_current) for fast app reads — only the current week's top 3 per customer are held in Postgres; Snowflake retains full history.

Output destinations¶

MOD-108 reads NBP rankings to prioritise which eligible products generate offers.
MOD-083 (agent assist) surfaces the top 3 NBP in the agent's customer 360 view.
MOD-077 (app dashboard) reads NBP for insight cards — surfaces as contextual prompts.

Fairness monitoring¶

Weekly fairness report computes recommendation rate by jurisdiction, age band, and customer segment. Any product where recommendation rate differs by more than 10 percentage points across demographic groups triggers a nbp.fairness_breach alert to the compliance team for model review.

Module dependencies¶

Depends on¶

Module	Title	Required?	Contract	Reason
MOD-105	Product eligibility engine	Required	—	Eligible product set from MOD-105 is the input constraint — NBP ranks within the eligible set only.
MOD-040	Churn & health score engine	Required	—	Churn probability and financial health score from MOD-040 are ranking inputs — retention-risk customers receive retention-oriented product recommendations.
MOD-106	ROTE engine	Optional	—	Customer ROTE from MOD-106 informs ranking to favour products that improve overall relationship profitability.
MOD-104	AWS shared infrastructure bootstrap	Required	—	AWS shared infrastructure provisioned by MOD-104 is required before this module can be deployed.

Required by¶

Module	Title	As	Contract
MOD-108	Product offer engine	Hard dependency	—

Policies satisfied¶

Policy	Title	Mode	How
CON-006	Product suitability and governance	`CALC`	NBP recommendations are scoped strictly to the customer's eligible product set from MOD-105 — no product outside the eligibility matrix can appear as a next best product.

CON-001	Customer Fairness & Conduct Policy	`CALC`	Ranking model is periodically tested for demographic fairness — no protected characteristic may systematically suppress product recommendations for a customer group.

Capabilities satisfied¶

(No capabilities mapped)

Part of SD06 — Snowflake Analytics & Risk Platform Compiled 2026-05-22 from source/entities/modules/MOD-107.yaml