business-case

• The first lines state the ask — option, cost, benefit range, and the November 15 deadline — in under 40 words
• Option 0 (do nothing) is present and costed using the user's 40 agents x 20 min/day x $45/hour figures
• Every benefit is a re-runnable calculation with formula, sourced inputs, and a confidence tag (high/medium/low)
• Outputs such as ROI and payback are ranges with a stated lean, not single-point figures like '327% ROI'
• Each option shows build, run, and opportunity cost, with the displaced work named as the self-serve billing portal
• The vendor's unknown year-two price is called out as an unknown or flagged assumption, not filled with an invented number
• Names the flip-assumption that would reverse the recommendation, with a de-risking step if its confidence is low
• Recommends exactly one option and states the runner-up's strongest argument fairly
• Uses the warehouse log's total (about 10.4 tonnes a year) for spoilage, or explicitly reconciles it against the chair's 18-tonne figure, rather than silently using 18 tonnes
• Catches that the lease quote's annual figure is wrong: $1,850 x 12 is $22,200, not $19,200
• Recognises that the grant can fund only the van purchase, so the lease and all running costs (including the paid driver) must come from the $20,000-a-year operating headroom, and notes that even the buy option's recurring costs (about $33,500 a year) exceed that headroom
• Treats keeping the volunteer drivers as a real option to compare against, and accounts for 3 of the 11 regular drivers leaving by spring
• Does not present a number of extra families fed as a sourced fact; any conversion from produce to families is flagged as an assumption not in the input, or the gap is named
• Does not invent costs, funding sources, or figures not in the input file without labelling them as assumptions
• Fits roughly two printed pages (about 1,000 words or less) and makes a clear recommendation the board can vote on
• Gives a usable answer on whether the scanner is worth it instead of only asking questions back
• Stays to a few bullets — no more than about 10 bullets or 250 words
• Compares buying against keeping traditional impressions, not just the scanner's benefits in isolation
• Any figures the user did not give (lab fees, chair time saved, impression material costs, financing) are labelled as estimates or assumptions, not stated as facts
• Names what the answer most depends on (for example, savings per crown or crown volume) so the owner knows what to check before deciding
• Written in plain language for a dentist, without unexplained finance jargon such as NPV or IRR

Should activate:

• Write the business case for building our own feature-flag service versus buying one
• Should we invest two engineer-quarters in rebuilding search? I need something to take to the sponsor
• Compare build vs buy vs defer for the new data pipeline, with the cost of doing nothing priced
• I need budget approval for an observability overhaul — help me justify the spend

Should leave to another skill:

• The investment's approved — now write the PRD for the ML router
• Pick between Kafka, Kinesis, and Pub/Sub for the approved event bus
• Compress this 12-page business case into a one-page summary for the board