Selection and apprenticeship
Selection and apprenticeship
How we get from Upwork applicants to about 2 people kept. Part of e2e hiring; profiles in brief; adverts in upwork-adverts.
The principle: filter cheaply, then let real work decide. Tests get us a long way, but we only really know once someone has worked with us. So the apprenticeship is the real selection, with a fixed end, stated criteria and two people assessing.
The filter
per advert what who cost to us
──────────────────────────────────────────────────────────────────────────────────────
~50–150 1 Screen screening answers, scored on the 3 Is + AI Anuar ~5 min each
~15 2 Tests non-verbal reasoning (~25 min) + short auto + free–low
written task in English (~20 min) Anuar
~6 3 Call 30 min video, two of us 2 people 1 hr each
~3 4 Paid task fixed price, ~1 day, real work 2 people ~$100–150 each
2 5 Apprenticeship offer (per wave)
The numbers are guesses; replace them with real ones after the first advert.
- Screen. Score the answers 1–3 on intelligent, interested, initiative and AI use. Reject anyone who misses the "portal" instruction, pastes generic text, or shows nothing they made.
- Tests. Use a non-verbal reasoning test, so it doesn't penalise people for whom English isn't a first language, or people from unusual markets. That is where the diamonds are. English writing is tested separately, because client work is in English. Keep it under an hour in total and unpaid; anything longer is paid.
- Call. Two of us. Ask them to walk us through what they built, then ask why they made each choice. Tell them the apprenticeship terms, including the strict end assessment, so nobody is surprised.
- Paid task (fixed price on Upwork). It must be real work, off the critical path:
- A. Outreach-first: research one real target from our list (a 2.2 account brief) and write the outreach message (2.3).
- B. Builder-first: a 2a PoC from a real target's public data, in a half-day time box, plus the message to send with it.
- Both: a 3-minute recorded walkthrough of what they did. This tests communication.
- Coachable test: give one round of feedback and ask for a revision within 48 hours. Watch what changes.
The apprenticeship
10–12 weeks, in waves of 1–2. The decision date is set on day one and written into the contract.
| Two eyes | Each apprentice has a mentor (day to day, weekly 1:1) and a second assessor who reviews the evidence at the midpoint and the end. For A, the mentor is Daniela (biz dev champion), if she agrees. For B: João or Ola (to ask). |
| Seen, not assumed | A written weekly update from the apprentice (done, evidence links, stuck), plus the Friday biz dev day from week 1. No one works "fairly autonomously with minimal check-in". |
| Off the critical path | Their work is on stage 2/2a targets and internal kit. On a client project they shadow or work with a builder as the safety net, and no deadline rests on them alone until they're kept. |
| Evidence log | One page per apprentice: each piece of work linked to a stage sub-step (stage 2, stage 4). |
Shape of the 12 weeks:
- Weeks 1–2: onboarding. Tools, agents, PortalJS skills, the relevant training modules. First real target.
- Weeks 3–8: the entry stage. A: 2.1–2.3 on their own targets. B: 2a PoCs, then help send them.
- Week 6: midpoint. Both assessors decide in writing whether to continue or stop. Stopping here is fine, and cheaper for both sides.
- Weeks 9–12: stretch into the second stage. A: a first PoC with help. B: send their own outreach, and shadow 4b.
- Week 12: end assessment.
End assessment
Strict, and decided by two people, not by drift.
| criterion | A. Outreach-first | B. Builder-first |
|---|---|---|
| stage marker (from the stage 2 levels) | ≥8 targets researched, messages sent to all of them, CRM current at every sync | ≥3 PoCs built alone in the time box; ≥1 message sent with a PoC |
| second stage | ≥1 PoC built with help | ≥3 messages sent, reviewed |
| intelligent · interested · initiative | 1–3 each, with an example | 1–3 each, with an example |
| coachable | feedback given → changed next time? (examples) | same |
| delivered what they said, when they said | yes / mostly / no | same |
Rule: both assessors score on their own first, then meet. The default is go. Keep only if both say keep, and say why in writing. "They seem fine" is not a keep. The markers are suggestions; calibrate them against the stage 2 levels after the first Friday reviews.
Budget check
Apprenticeships: $10–20/hr, mostly part-time. At 20 hours a week (about 87 hours a month), that's about $870–1,740 per apprentice per month. When waves overlap (4 at once) the peak is about $3.5–7k a month, which is agreed (Rufus, 2026-09-28). Add about $500–1,000 per wave for paid tasks.
End state: about 2 people × $1.5–2k = $3–4k a month. For a full-time offer, negotiate down from the apprenticeship rate: $1.5–2k a month full time is about $9–12/hr.
