Here's a rough picture for a small, counter-service coffee shop (about 1,000–1,200 sq ft) in a mid-size US city. All figures are ballpark.
Startup costs (one-time): about $55,000–$60,000
| Item | Estimate |
|---|---|
| Espresso machine, grinders, brewers, fridge, ice machine, small equipment (some used) | $18,000–$22,000 |
| Build-out and renovation, signage | $10,000–$15,000 |
| Lease deposit and first month | $4,000–$6,000 |
| Permits, licenses, insurance, legal | $2,000–$4,000 |
| Opening inventory, POS system, smallwares | $4,000–$6,000 |
| Branding, website, opening marketing | $2,000–$3,000 |
| Total | ~$40,000–$56,000 |
That leaves you only $4,000–$20,000 of working capital. That is the main weakness of this plan. I'd aim to keep at least $15,000–$20,000 in reserve. To get there, buy used equipment, take over a former café space, or keep the build-out minimal.
Monthly operating budget (steady state)
| Item | Estimate |
|---|---|
| Revenue target (~300 customers/day × ~$6.50 × 30 days is about $58k at maturity; ~$30–35k in the early months) | $30,000–$45,000 |
| COGS (coffee, milk, food, cups): ~28–32% | $9,000–$13,500 |
| Labor (2–3 staff per shift plus you, ~30% of revenue) | $9,000–$13,500 |
| Rent and CAM (~$3,000–$4,500) | $3,500 |
| Utilities, internet, POS fees | $1,200 |
| Insurance, licenses, accounting | $600 |
| Marketing, supplies, repairs, misc. | $1,200 |
| Loan payments (if any) | $0 |
| Total expenses | ~$25,000–$34,000 |
Your own pay comes out of whatever is left, or out of the labor line if you work behind the counter.
Break-even
- Monthly break-even revenue: roughly $27,000–$32,000. That is about 140–165 transactions a day at a $6.50 average ticket.
- Timeline to hit monthly break-even: typically 6–12 months. Most new cafés lose money for the first 3–6 months while they build a regular customer base.
- Timeline to recoup the $60k: realistically 3–5 years. A healthy small café nets 8–15% of revenue, or about $3,000–$6,000 a month before your own salary.
Main risks
- Thin working capital. A slow first 6 months could use up your cash.
- Location. Foot traffic and the morning commute matter more than anything else.
- Labor and rent. If either goes above ~35% and ~10% of revenue, break-even moves out sharply.
Ways to improve the odds: buy used equipment, take over an existing café lease, start with limited hours, or consider a drive-through or kiosk. They have lower rent and lower build-out costs.
A few things worth a second look:
- How would the break-even change if rent is $5,500 a month instead of $3,500, in my city?
- Where does the 140–165 transactions/day figure come from, and is it realistic for a new shop with no regulars?
- I didn't say whether I'd work the counter myself. How does that change the labor line and my own pay?
Real run recorded with claude-code / claude-sonnet-5-5. Output is shown verbatim, unmodified.
What it does
When the model gives an answer you might act on (advice, estimates, a plan, a draft pitch or email, data analysis), this Skill makes it append two or three short follow-up questions at the end. Each question points at something concrete in that answer: a figure to verify, an assumption to probe, or context the model had to guess.
How it works
- The model finishes its normal answer first, in full.
- If the answer qualifies, it adds a fixed lead-in line, "A few things worth a second look:", followed by 2-3 plain bullets written as questions you can send back.
- It does this at most once per conversation.
When it stays silent: simple lookups, creative writing, casual chat, code you will run, purely educational explanations, and cases where you asked it to verify or cite, asked for the quick version, asked it to review your own work, or handed it the material to reformat or summarize.
Good for: anyone who tends to take a confident, well-structured AI answer at face value.
Pure instructions: no scripts, no network access and no file writes. It only changes how answers end, adding a few lines at the bottom of some replies. The follow-up questions are prompts for you to check, not verified facts.