Jobab — Design-Partner Pilot Kit
Purpose: get three Dhaka social-commerce merchants using Jobab for four weeks, generate the §13 eval set from their real conversations, and learn whether anyone actually pays.
This is not engineering work. The engineering team's job during the pilot is to (a) keep the system up, (b) instrument what matters, and (c) ship one fix per merchant per week — no more.
Files
interview-script.md— 45-minute first-call scriptpilot-plan.md— week-by-week execution plan + success criteriamerchant-consent.md— consent + data-handling agreementpilot-outreach.md— recruiting message templatesweekly-checkin.md— structured weekly merchant check-in
Target merchant profile
- Sells fashion / lifestyle / accessories on a Facebook Page
- 50+ customer DMs per week (real volume, not vanity)
- One owner + 0-2 staff
- Already taking orders via Messenger (so the AI has something to do)
- In/near Dhaka so we can do an in-person session in week 1
- Not an existing customer of a competitor bot (no contamination)
Pilot success = green light for paid launch
| Metric | Floor (decide-to-kill) | Target (decide-to-launch) |
|---|---|---|
| Merchants completing 4 weeks | ≥1 of 3 | 3 of 3 |
| AI autonomy ratio (turns AI handled / total) | ≥50% | ≥75% |
| Orders created via AI / week / merchant | ≥3 | ≥15 |
| Merchant-reported "saves me X hours/week" | ≥3 hours | ≥10 hours |
| Willingness-to-pay (qualitative) | "maybe ৳500/mo" | "I'd pay ৳2,000/mo today" |
| Bangla eval score on collected set | ≥60% | ≥80% |
If we hit the floor column, iterate. If we hit any single metric below the floor by end of week 4, we should consider whether the wedge is right.