Skip to content

Unified customer architecture — the strategy

Sourceclient-resources/unified-customer-architecture.pdf
TypePDF, 25 pages, presentation-style long-form
OriginCanada Segment Enterprise Architecture
Dated2026-06-13 · marked for review
ClassificationManulife internal

What it is

The most complete statement of a target state in anything we have been given. It is a strategy document, not a solution design: it defines the layers, the ownership boundaries between them, and the discipline that keeps those boundaries honest. Its opening claim sets the tone — "the goal is not modernization, it is determinism" — and it returns to that repeatedly: identity is verified, never inferred; schema is approved, not invented; where uncertainty exists the architecture suppresses rather than guesses.

The starting position it describes

Nine secure portals. Five mobile apps. Fourteen discrete experiences. Between three and twelve manual steps to link a product. Zero unified customer identifiers at runtime.

Today identity is created at the product or business-unit level rather than as an enterprise relationship; product relationships resolve independently inside BU systems and never centrally; profile and preference data is duplicated with no single current view; consent lives in compliance artifacts rather than runtime decisions; and operational customer context does not exist, so channels either query mastering systems directly or render canned data.

The four layers

The architecture is layered, and the document is emphatic that each layer does one job:

  • Experience composes.
  • Access and authorization govern.
  • Operational profile serves.
  • Enterprise data (MDM) masters.

"Mastering does not serve UI. Serving does not master." The read side and the write side are two stores with two jobs: MDM is a property graph of nodes and typed edges optimised for resolution, survivorship, lineage and ontology evolution; the operational profile is a denormalised document keyed by relationship id, optimised for low-latency reads by channels, BFFs, widgets and agents. The projection between them is event-driven, idempotent, replayable and keyed on (rel_id, ontology_version, event_id).

The named components

ConcernComponent named
Identity providerForgeRock CIAM (already operated)
Mastered relationship graphMDM, elevated to governed ontology graph
Consent system of recordOneTrust, with authenticated user context
Policy decision pointPingAuthorize (formerly Symphonic), PEP in the BFF tier
Runtime customer contextOperational profile, projected from MDM

Identity and authorization deliberately stay inside a single Ping support boundary. Every policy decision is logged with a full attribute trace.

Visibility and action are separate decisions

The section that most directly shapes the portal: "what the customer sees is not what the customer can do." The PDP consumes relationship, product, consent, entitlement and jurisdiction inputs and emits two answers per request — can the customer see this, and can the customer act on it — with stricter thresholds on the second. A product may be owned, and ready, and still not actionable.

Identity resolution

A five-stage pipeline, deterministic at the boundaries and probabilistic only in the middle (K-means sits in stage three). Verification is mandatory where ambiguity exists. Recovery must not mint a new identity when an existing one should have been recovered.

The seven workstreams

The document decomposes delivery into seven independently shippable workstreams across three phases — and explicitly notes that the landing-page MVP already follows this pattern, with interim correlation keys swapped for enterprise relationship ids over time.

#WorkstreamOwnerPhase
01Canonical ontology and contractsMDM1
02Match/merge and mastered graphMDM1
03Product auto-associationMDM + BU2
04Consent management integrationOneTrust2
05Operational profile and runtime APIsPlatform1
06Policy and entitlement (PDP)Identity2
07Experience integrationChannels1

Phase 1 is backbone and shell; Phase 2 is auto-link and consent-aware; Phase 3 is AI-assisted reasoning over the governed graph — candidate-only, written through governed flows.

It also traces a single login end to end: twenty-one numbered interactions across eight participants in four phases (bootstrap, authenticate, compose, render), with two failure variants that fork inside a phase and rejoin at the next boundary. Group Benefits gets its own ontology section, because a member holds a reference to a sponsor's policy rather than owning one — which generalises to every future audience.

Why it matters here

This is the document the SOW means by "the existing target state vision and architecture direction". Our WS 2 target state architecture has to either build on it, extend it, or argue against it — and say which. Three things make that non-trivial:

  • It is dated June 2026, before the ADS and before both critical reviews. Nothing tells us whether it is still the segment's position.
  • Its Phase 2 content — ECID, MDM, consent, PDP/UES entitlements — is precisely what the GARB ADS excludes from MVP1. So the MVP is Phase 1 of this plan, or it is something else, and the material does not say which.
  • Workstream 07 assumes "forward-compatible contracts so identity-key swaps are non-breaking." Whether the MVP's contracts actually have that property is an architectural question we can answer, and it is worth answering early.

Open questions raised

  • Is this still the endorsed Canada Segment target state as of today, and who owns it?
  • Is the MVP being built as Phase 1 of this architecture, or independently of it?
  • Are PingAuthorize, OneTrust and the operational profile funded and staffed, or aspirational?
  • Are the MVP's BFF and Integration API contracts forward-compatible with an ECID swap?

Internal working knowledge base — not for external distribution.