Architecture Review Deck

About this template

Ten slides that end with a decision

An architecture review deck is not a design document. The design document explains how something works; this deck exists so that a specific choice is approved, rejected or explicitly deferred before the room breaks up. That is why slide 4 is not background but the list of things being decided today, and why the closing line is “We put the decision in writing”. Ten slides is roughly a 20 minute slot, which is what a single decision usually deserves.

Five agenda lines, two body pairs

Slide 2 runs what we decide today, options compared, proposals that pass and proposals that wait, key metrics, trend analysis. Only the first two carry a divider and a body slide – slides 3 and 4, then 5 and 6. The comparison is slide 7, the cards slide 8, the bars slide 9. Risks accepted, review criteria and the decision record are not built as their own slides, so duplicate the slide 5 and 6 pair for each one you need and continue the SECTION numbering as 03 and 04, adding the matching agenda lines.

The two body slides

• What we decide today: “Approve splitting out the payment module”, “Pick the message queue product”, “If held, set the date to revisit”. The third line is the one people delete and then regret – a deferral with no date is a decision not to decide.
• Options compared: “3 options: keep, partial split, full split”, “Judged on latency, ops load and migration cost”, “Partial split has the lowest 6-month cost”. Name the yardstick before the winner. If the criteria appear after the recommendation, the recommendation looks reverse-engineered.

The metric cards and the cost bars

Slide 8 holds “Options reviewed” 3, “Security review items” 12, “Availability target (%)” 99.9 and “Decision deadline (wks)” 2. The pairing is deliberate: what was compared, what was checked, what has to hold, and by when the room has to answer. Slide 9 is not a time series – it is the four options priced against each other, with “Keep monolith” at 100 as the baseline, “Partial split” 82, “Full split” 128 and “Full rewrite” 190. Because the baseline is an index, the bars only mean something if you say what the index is over: six months of total cost of ownership is the reading the body slide implies. Replace the “Unit: index” label with your own basis, and clear the sample-values footnote under the cards.

The comparison slide

Slide 7 is the review checklist split in two. “To get approved” asks for “Attach the load test results”, “Submit migration and rollback steps” and “Name an owner for the ops handover”. “What gets held” lists “Performance claimed without measurement”, “A design with no way back” and “No operating cost estimate”. Sent round before the meeting, it saves the meeting: proposers see what will get them held before they walk in.

What goes wrong

• Comparing three options against criteria invented for the winner. Fix the yardstick on slide 6 first, then score.
• Presenting 99.9% as a property of the design rather than a target that costs money. The gap between 99.9 and 99.99 usually shows up on slide 9 as a bigger bar.
• Leaving the meeting with a verbal yes. If nothing was written down, the same options come back next quarter with different numbers.