AcademySA Accelerator › Module 02

Design Decisions & Tradeoff Documentation

Architecture decisions that aren't documented don't survive the SA's departure from the engagement. The developer who extends the integration six months later doesn't know why the pattern was chosen, what alternatives were rejected, or what constraints would break if the design is changed. Architecture Decision Records exist to carry that context forward — and AI makes them fast enough to actually write.

⏱ 30–35 min 3 knowledge checks ADRs, tradeoff analysis, constraint documentation
02
SA Module
Your progress
0%
1

Why architecture decisions disappear — and what it costs

The SA on a Guidewire implementation makes dozens of significant architecture decisions — which integration pattern to use, how to handle the data migration approach, where to place the transformation logic, how to structure the security boundary. In most engagements, fewer than a quarter of those decisions are formally documented. The rest live in the SA's head, in meeting notes that nobody can find, or in Slack threads that expired six months ago.

When the SA rolls off the engagement and the implementation team needs to extend the integration — or when something breaks in production and the team needs to understand the design intent — those undocumented decisions create expensive problems. Teams make changes that violate constraints they didn't know about. They re-debate decisions that were already made for good reasons. They extend designs in ways that break the assumptions the original architecture depended on.

Architecture Decision Records (ADRs) exist to solve this problem. They're short, structured documents that capture each significant decision: what was decided, why, what alternatives were considered, and what the consequences are. AI makes them fast enough to write during the engagement — not as an afterthought when you're rolling off.

The most expensive undocumented decision in insurance IT

The most common undocumented decision that causes production problems in Guidewire implementations is the data migration strategy — specifically, decisions about which legacy data to migrate, which to archive, and which to leave in the legacy system accessible via integration. These decisions are made early, often under pressure, and rarely documented with the constraint context that shaped them. A developer who later extends the system and doesn't know this history will make assumptions that violate the original migration boundary — with consequences ranging from data inconsistency to regulatory reporting errors.

2

ADR structure and AI-assisted drafting

A good ADR has six components. AI can draft five of them from your notes — fast. The sixth is the one only you can write.

1. Decision title and date
A specific, searchable title that describes what was decided. "Integration pattern for PolicyCenter-to-legacy-claims real-time sync" — not "Integration architecture." Date of decision, status (proposed / accepted / superseded), and who made it.
2. Context
What problem was being solved. What the requirements were. What constraints existed at the time. AI can draft this from your notes — you verify it's accurate and add constraints that weren't written down anywhere.
3. Decision
What was chosen. Stated plainly. Not "we considered options" — "we chose event-driven architecture using IBM MQ with a Guidewire Integration Framework message handler."
4. Alternatives considered
The other options evaluated and why they weren't chosen. This is where AI is most useful — it can enumerate alternatives systematically and articulate why each was rejected given the stated constraints. You validate that the rejection reasons accurately reflect the actual decision context.
5. Consequences — complete
This is the section AI consistently under-writes. Consequences must include what this decision enables AND what it constrains or prevents. What will future architects not be able to do because of this choice? What must they preserve for the design to continue working? What technical debt does this decision create? AI writes benefits; you must add constraints.
6. Review trigger
When should this decision be revisited? What conditions would make this decision wrong — new regulatory requirements, a change in the legacy system's capabilities, a significant growth in transaction volume? Documenting the review trigger prevents good decisions from becoming bad ones silently.
Prompt — ADR for Guidewire integration pattern
Task Generate an Architecture Decision Record for the following decision I made on a Guidewire PolicyCenter integration project.
Decision notes We chose an event-driven integration pattern using Guidewire Integration Framework (GIF) with IBM MQ as the message broker, rather than REST-based real-time API calls or a batch file-based approach. The legacy claims system only accepts inbound MQ messages — no REST capability. The client team has strong MQ skills. The 30-second latency SLA rules out batch. A REST gateway layer was considered but rejected because it would introduce a new middleware component requiring a 3-month procurement process. The MQ approach reuses existing infrastructure and skills. Key constraint: message size limit of 512KB on the existing MQ queues — policy change payloads for complex commercial policies may approach this limit and will need to be chunked. This constraint was discovered late and should be flagged prominently.
Format Full ADR with all six sections: title/date/status, context, decision, alternatives considered (include REST API gateway and batch file approaches with specific rejection reasons), consequences (include BOTH what this enables AND what it constrains — particularly the 512KB message size constraint and its implications), and review trigger. In the consequences section, explicitly document what future architects must preserve for this design to continue working. Flag the message size constraint as a prominent known limitation requiring monitoring.
Knowledge Check
AI drafts the ADR consequences section as follows: "This approach enables reliable message delivery using existing MQ infrastructure, leverages the team's existing skills, meets the 30-second latency SLA, and avoids a lengthy procurement process." Your review identifies that the 512KB message size constraint on existing MQ queues means complex commercial policy payloads may need to be chunked — and this chunking logic will need to be implemented by every system that sends messages through this integration. What should you do?
3

What only the SA can add to the ADR

AI can structure an ADR from your notes. What it cannot add is the context that existed only in conversations — the constraint that emerged in a hallway conversation with the client's enterprise architect, the regulatory guidance that came from a phone call with FSRA that was never written down, the vendor's verbal confirmation that a certain API capability would be available in the next release. This context shapes decisions but rarely makes it into documentation unless the SA puts it there explicitly.

💬

Verbal commitments and confirmations

Vendor claims about roadmap capabilities, client confirmations about system behaviour, regulatory guidance from informal conversations. These shaped your decision and they need to be in the ADR — with the caveat that they're verbal and the confidence level attached to them. Future architects need to know the decision rested partly on an unverified commitment.

🕐

Time-bounded constraints

Some constraints exist because of a specific circumstance that may not persist. The procurement process that ruled out Option B. The team skill gap that made Option C non-viable. The regulatory review backlog that made the timeline constraint real. Document when these constraints were true — a future architect revisiting the decision needs to know whether the constraint still applies.

⚖️

The real reason, not the official reason

Architecture decisions sometimes have a political or organisational dimension alongside the technical one. The option that was technically preferred but rejected because a key stakeholder had a vendor relationship. The approach that was chosen partly because it was the one the client's lead developer was comfortable with. The ADR should reflect the actual decision context, including the non-technical factors that were real influences.

🔮

Known future risks

What you know at decision time about how this decision might become wrong in the future. The legacy system's end-of-life timeline. The expected growth in transaction volume. The regulatory changes being discussed that would affect this pattern. Document what you know about the horizon — the review trigger section is where this lives.

Knowledge Check
Your ADR for the MQ integration pattern includes this review trigger: "Revisit if transaction volume grows significantly or if the legacy system is replaced." During the engagement, the legacy system vendor verbally confirmed to you that end-of-life for the system is planned for 2028 — three years out. Should this be in the ADR? How?
4

Tradeoff documentation discipline — the complete picture

The consequences section of an ADR is where most AI-generated documents fail. AI writes the benefits of the chosen approach thoroughly and treats the constraints as minor caveats. In practice, the constraints are exactly what future architects need to understand — because they're the things that will break if someone makes a change without knowing they exist.

A complete consequences section documents three things: what the decision enables, what it constrains, and what it requires. "Requires" is often the most important — the things that must be true for the design to keep working, that aren't obvious from the design itself, and that would be violated by a seemingly reasonable future change.

Prompt — completing the consequences section
Task I have a draft ADR for a Guidewire integration decision. The consequences section currently only describes benefits. Help me complete it with a full picture.
Context Decision: event-driven integration via IBM MQ with 512KB message size limit, chunking required for large payloads. Team has MQ skills. Legacy system EOL expected 2028 (verbal, unconfirmed).
Format Rewrite the consequences section with three explicit sub-sections: (1) What this enables — benefits and capabilities this decision creates. (2) What this constrains — what future architects cannot do, what must be preserved, what would break if changed. (3) What this requires — the ongoing conditions that must be true for this design to keep working. Be specific. Avoid generic statements. Every constraint should be actionable enough that a future architect knows what to avoid.
Knowledge Check
A developer joining the Guidewire project six months after your engagement ends reads the ADR you wrote for the MQ integration pattern. They want to add a new high-priority endorsement notification that needs to reach the claims system in under 5 seconds — faster than the current 30-second SLA. They consider using a REST API call alongside the MQ integration for this high-priority path. Your ADR's consequences section says: "Future additions to this integration pattern must use the MQ interface — the legacy system does not support REST inbound calls." They follow the ADR guidance and raise the constraint with the architecture review board rather than implementing the REST path. What does this demonstrate?
5

Module summary

ADRs are the SA's professional legacy

Decisions that aren't documented don't survive the SA's departure. The constraint that shapes every future extension of the design needs to be in writing, in a place future architects will find it, before you roll off the engagement.

AI drafts structure; SA adds context

AI produces a solid ADR skeleton from your notes. What you must add: verbal commitments with confidence levels, time-bounded constraints, non-technical decision factors, and the specific consequences that only you know from the engagement context.

Complete consequences: enables, constrains, requires

AI writes benefits. You add constraints and requirements. Every hard limit — message size, system capability, team skill boundary — belongs in the consequences section. Future architects need to know what would break, not just what works.

Review triggers carry strategic context

Document when and why the decision should be revisited. Include what you know about future conditions — vendor EOL timelines, expected volume growth, regulatory changes in discussion — with the confidence level attached to each. Future architects need the strategic horizon you saw from your engagement vantage point.

Ready for Module 03

Module 03 — Integration Architecture in Insurance IT — goes deep on the integration design challenges that define most insurance IT SA engagements: Guidewire integration patterns, legacy system connectivity, data migration architecture, and the specific constraints of the Canadian insurance regulatory environment.

Module 02 Complete

Design Decisions & Tradeoff Documentation is done. Continue to Module 03: Integration Architecture in Insurance IT.