Architectural Decisions
Card fraud on Haat's checkout has climbed to about ₹2 lakh a month. Kabir wants to build fraud rules in-house; Farhan suspects a specialist vendor would be faster. Six months from now someone will ask why the team chose what it chose, and the answer should not be "ask Farhan, he remembers."
The rule
Weigh at least two real options, then write down why you chose.
An engineering manager doesn't have to make every technical call; Farhan, as tech lead, usually will. Meera's job is to make sure the decision process is sound:
- A clear decider. Everyone knows who makes the call, and by when.
- Real options. At least two genuine alternatives, plus "do nothing" as a baseline. A single option presented as the decision is really just a proposal.
- Criteria tied to needs. Cost, time to deliver, how hard it is to run, the team's skills, security, and how well it fits where the product is heading.
- Effort matched to reversibility. Choices that are easy to undo, like a library for one feature, should be made quickly. Choices that are hard to undo, like a database, a public API or a long vendor contract, deserve a slower look.
- A written record. A short decision record captures the context, the options, the decision and its consequences.
This is also where build versus buy questions belong. Build when the capability is core to what makes your product different, or when nothing on the market fits. Buy when it isn't core, mature products exist, and your team lacks the expertise.
A decision record can be half a page:
DR-014: Use a vendor service for fraud scoring
Status: Accepted, 12 August
Context: Card fraud losses are about ₹2 lakh a month. Nobody on
the team has built fraud models before.
Options: 1) Build rules in-house 2) Buy vendor scoring 3) Do nothing
Decision: Option 2, on a one-year contract.
Consequences: Costs about ₹60,000 a month and adds about 100 ms to each
card payment. Revisit if losses stay above ₹50,000 a month.The shortcut
Before any significant decision, ask three questions: Who decides? How hard is it to undo? Where will the reason live?
Who decides?
Farhan makes the call by the end of the month. Meera brings in Rohan on customer impact and finance on cost.
Drills
Answer fast — trust the rule.
Which decision deserves the most careful written comparison?
What belongs in a decision record?
What is the engineering manager's main job in an architectural decision?
Which fact most strongly favours buying rather than building?
Remember
- Weigh at least two real options, then write down why you chose.
- Ask: who decides, how hard is it to undo, and where will the reason live?
- Build what makes you different; buy what is not core and already solved well.