What you need to know
An architect can understand a trade-off perfectly and still fail to get a decision, because the people deciding do not think in tokens, endpoints or chunking strategies. They need to know what they are choosing between, in terms of the things they are accountable for.
Three elements of every trade-off
- What you gain with this option.
- What you give up: cost, capability, speed, risk.
- What reversal costs: what it would take to change course after the system is built around this choice.
The third element is the one most often left out, and it is what turns apparent agreement into real agreement. A stakeholder who approves an option without knowing it is hard to undo has not really decided.
| Example decision | Gain | Give up | Reversal cost |
|---|---|---|---|
| Load whole documents instead of building retrieval | Simpler to build and launch | Higher cost per call as documents grow | Building retrieval later, under production cost pressure |
| One delivery route for all workloads | Simpler identity, logging and operations | Flexibility for regional or regulatory needs | A late residency requirement can block go-live |
| Lighter logging to save latency | A faster experience | Visibility into individual decisions | Rebuilding audit trails after an incident or audit finding |
Speak the decider's language
| Audience | Frame the trade-off in terms of |
|---|---|
| Finance | Run cost, payback, cost per outcome, budget risk |
| Operations leaders | Throughput, SLAs, staff workload, peak capacity |
| Compliance and legal | Obligations met or at risk, evidence, liability |
| Executives | Business outcome, timing, risk, what is irreversible |
| Engineering | Complexity, reliability, maintainability |
A weak answer to "why can't we use one platform for everything?" lists endpoints and regions. A strong answer says: one platform is cheaper to build and run; it limits our ability to meet different countries' data rules; if such a rule arrives after launch, moving regulated workloads could delay rollout by months. "I recommend choosing the primary platform by our strictest compliance need, and adding a second route only where a workload needs it."
Demos and proofs of concept
A demonstration built on the stakeholder's own scenario, with realistic (anonymized) data, shows more than a generic one. State one or two known limitations up front: discovering them halfway through damages trust, while naming them shows control of the scope.
In the exam
Trade-off questions describe a decision, an audience and some options. Choose the presentation that states gain, cost and reversal cost in the audience's terms, includes a recommendation, and respects stated requirements (an explicit audit-trail requirement cannot be traded away for speed). Watch for options that omit reversal cost or hide a requirement violation inside a performance gain.
Check your understanding
A proposal to cut logging detail for faster responses lists the speed gain and the loss of detail. What else must it state?
Show answer
Answer: D. The reversal cost is the element stakeholders most need and most often do not get. Restoring audit detail after the system depends on the speed gain can be expensive.
A CFO asks why the design needs two delivery routes. Which answer is best?
Show answer
Answer: C. It states the business trade-off: lower cost against a regulatory need that one route cannot meet, which is what the CFO needs to decide.
Which TWO practices make a trade-off presentation decision-ready?
Show answer
Answer: A, C. Decision-ready means real options, all three elements, in the audience's language. Hiding options or limitations produces false alignment.
Apply it: Present Larkfield's SLA trade-off
Larkfield's VP of Claims Operations is frustrated: the acknowledgement SLA is still at 90%, short of 98%, because low-confidence extractions wait in the handler queue. The VP proposes letting low-confidence claims go straight through, "since handlers can fix things later".
- Lay out at least two options with gain, give up and reversal cost for each.
Done when: the VP's own proposal is one of the options, presented fairly. - Write your recommendation in two or three sentences addressed to the VP.
- Name the evidence you would bring to support it.
Show model answer
1. Options
| Option | Gain | Give up | Reversal cost |
|---|---|---|---|
| A. Low-confidence claims go straight through | SLA likely above 98% at once | Wrong fields enter ClaimsCore; more rework later; compliance exposure if wrong policy facts reach customers | Finding and correcting bad records after the fact; regulator questions if acknowledgements were wrong |
| B. Keep the review, but acknowledge first: send a holding acknowledgement within the SLA for every claim, and review fields afterwards | SLA met; data quality protected | A second customer message for some claims; handler workload unchanged | Low: a template and a routing change |
| C. Keep the review and add handler capacity at peak times | SLA improves; quality protected | Staff cost at peaks | Low |
2. Recommendation. I recommend option B: every customer gets an acknowledgement within a day, which is what the SLA measures, while uncertain claims are still checked before they become records. Option A meets the SLA by moving errors into ClaimsCore, where they cost more to fix and can reach customers. If B does not get us to 98%, we add peak capacity (C).
3. Evidence. The share of low-confidence claims that handlers actually corrected last month and the average rework time per error (35 minutes); SLA performance by day of week; and a sample holding acknowledgement approved by compliance.