4 · Insights & Observations
| Slide budget | 3 slides, 6 minutes |
| Skeleton asks for | Key findings from client docs · early learnings and direction decisions from WS 1 and WS 2 · key trade-offs made · risks identified, with mitigations |
| Assembled from | Client resources · open questions · recommendations |
| Status | 🟡 Outline — slides not built |
The line the section has to land
Manulife's own reviewers have already named the gaps. Our value is reconciling them into one target state, not re-finding them.
Everything in this section is traceable to a document in the room. The segment architecture assessment recommends return for re-architecture on 20 blocking defects across eleven domains; the EA readiness review returns NO GO on the October date. Both post-date the GARB ADS, and nothing we have been given reconciles them with it. That unreconciled gap is the engagement.
Say the validation caveat once, here: nothing on these slides has been confirmed with the authors yet, and that confirmation is WS 1's deliverable rather than a formality.
Slide 4.1 — What we read, and what it told us
Claim: eight artifacts, six altitudes, one live dispute — and no agreed order between them.
| Layer | Artifacts | What it establishes |
|---|---|---|
| Strategy | Unified customer architecture (Jun 2026) | The most complete North Star that exists: ECID, MDM as governed ontology, OneTrust as consent SOR, PingAuthorize as PDP |
| Options | The Architecture Decision | Single BFF vs federated GraphQL vs CQRS — framed, not chosen |
| Build | GARB ADS (MVP Phase) | Next.js shell, per-BU micro-frontends, Dashboard BFF, Integration API, Redis — explicitly excluding ECID, MDM, consent and centralized authorization |
| Review | Segment assessment · readiness review | Return for re-architecture · NO GO |
| Estate | API landscape · experience map | ~4,200 API operations across two gateways; the target homepage maps onto 42 real APIs, ~20% net-new |
| Enterprise | API Mediation proposal (Mar 2025) | The only enterprise position on API architecture: a centralized Inter mediation layer over federated per-LOB service meshes. Arrived outside the original seven, on 16 Sep |
Speaker note. The insight is the ordering, not the inventory. Read build-first, the programme looks nearly done. Read strategy-first, it looks barely started. The MVP is not building toward the strategy document, and no one in the material says which of the two is of record — and the mediation proposal, the oldest document here and the only enterprise one, is cited by none of the four that followed it.
Do not put the words "return for re-architecture" and "NO GO" on the slide in isolation. They are the client's own ratings and belong in the narration, attributed, with the point that we are working from them rather than adding to them.
Slide 4.2 — Five observations that change the architecture
The filter for this slide: each one moves a boundary, not a detail, and each is cheaper to settle now than after the MVP hardens around it.
| # | Observation | Evidence | Why it is structural |
|---|---|---|---|
| 1 | No agreed North Star | Readiness review concern #2 · assessment F-02 | WS 2 documents an architecture whose authority is undefined until someone is empowered to say which target state is of record — Q8 |
| 2 | Aggregation relocated the authorization decision | Assessment B-01 (blocker) · ADS "no change to authorization" | Customer scope arrives as a query parameter (GET /api/v1/investments?ciamId=), not an authenticated claim. Textbook OWASP API1. Exposure: 1,435,783 logins/month, 3,309,891 with mobile (L-05) |
| 3 | The availability commitment is arithmetically unreachable | Assessment C-01 (blocker) | 99.99% committed; thirteen serial dependencies at 99.9% computes to 98.71% — ~129× the committed downtime. Only designed degradation closes it, and none is specified |
| 4 | The programme unifies the smaller channel first | Assessment L-01, L-02 (blockers) | 56.6% of logins are mobile (1,874,108 vs 1,435,783). Manulife Bank is 82.9% mobile. Micro-frontends are browser-native — a native client cannot load one |
| 5 | An eighth artifact arrived, and nothing written since cites it | API Mediation proposal · Q5, Q29 | Manulife's own enterprise API architecture marks JWT Exchange mandatory — the mechanism observation 2 needs and the MVP omits. Its two-tier model also puts two policy-enforcing hops under each of the eight fan-out calls, while rate limiting is marked optional |
Carried, but not on this slide — raise if asked, and hold for the working sessions:
- No volumetry anywhere. The ADS commits 1,200 rps against a single documented downstream ceiling of 25 TPS at FCC — a 48× gap (C-02), with the dashboard alone projected at 61% of that ceiling once mobile onboards (L-04). Eight downstream calls per page composition, generated by page load rather than customer intent (L-03).
- Two BFFs are described in one estate. The ADS specifies Next.js route handlers; the experience map documents
di-dashboard-services, a Fastify service already running in DEV and already wired to homepage zones. Same component or two is not stated anywhere — Q19. - The accountable owner is an organisation, not a person. Shell, BFF, Integration API and Redis have no business-unit home — Q14. This is an operating-model gate, and SOW §3.0 puts operating model design out of our scope: we surface and sequence around it.
Speaker note. Observations 2, 3 and 4 all resolve at the same architectural seam, and observation 5 is that seam one tier down, in the client's own document. That is the setup for section 5's first four themes; the two carried items above — two BFFs in one estate, and no owner for the shared runtime — are the setup for the fifth. Do not resolve any of them here — name them and move.
Slide 4.3 — Trade-offs we have made, and the risks we are carrying
Two halves on one slide. The offsite asks for both, and neither fills a slide alone.
Direction decisions taken in WS 1 and WS 2
| Decision | What it buys | What it costs |
|---|---|---|
| Anchor every position to a finding Manulife's own architects rated BLOCKER | The problem statement is not in dispute — the argument starts at the remedy | Slower, and it constrains us to concerns already raised |
| Recommend inside the MVP's shape, not a rebuild | Deliverable within the current programme; no pause requested | Some strategy-layer capability (MDM, consent, PDP) stays deferred |
| Record what is unresolved rather than closing it with a plausible guess | 32 open items with named owners; the register is the evidence trail | The deck shows open questions where leadership may expect answers |
| Knowledge base as the working surface, deck as an export | One source; the diagram in the room matches the document behind it | Handoff format is still unagreed — Q1 |
Risks, with the mitigation we are running
| Risk | Mitigation |
|---|---|
| Nothing is validated with the authors yet | WS 1 validation sessions — this offsite is part of that |
| The June 2026 strategy may no longer be Canada Segment EA's endorsed position — Q10 | Ask directly, in the room, at this offsite |
| Volumetry absent; capacity gate unquantified — Q24 | Current APIM per-tenant limits already exist and are the fastest partial answer — Q27 |
| Environments unprovisioned — the programme's own #1 schedule risk — Q25 | Sequenced explicitly in WS 3 rather than assumed |
| Ownership vacuum over the shared runtime — Q14 | Named as an architectural dependency and a roadmap risk; we do not design the operating model |
Speaker note. The trade-off half is the confidence-builder — it shows the direction was chosen, not defaulted into. The risk half is what earns the right to ask for decisions in section 5.
What this section must not do
- Grade the delivery team. Every finding cited belongs to Manulife's architecture function. The framing is reconciliation, not audit, and the delivery team may be in the room.
- Present the count as the finding. "20 blockers" is a headline that invites a defence of the count. Four named consequences invite a conversation about the architecture.
- Show an unregistered diagram. See Our Progression.