We spent four years training African creatives. Here is what was missing.

Most companies publish the result. We are going to publish the bet.

The founding cohort of the Senpai Collective is forming. It is the second time we have tried this. The first time worked and we could not keep it, and the reasons we could not keep it are the reason the platform exists at all.

This note is here so that when we report back, you can check it against what we said beforehand.

It worked the first time

Between 2017 and 2021 the original Senpai community grew to more than 2,000 members across Nigeria and Africa, with a newsletter that reached over 3,000 subscribers. It ran four cohorts across five tracks.

The flagship was Design 101, a six week course. Across its cohorts it took 157 students, split into WhatsApp groups with mentors so nobody was ever in a group larger than about fifteen. 45 people finished it.

At least 30 of those 45 now earn in dollars, through a mix of freelance gigs and direct hires, and some run their own companies. That happened with no job board, no employer relationships and no placement programme of any kind. They went and built careers on the foundation the course gave them.

Two things about that are worth saying plainly. The first is that most of them are doing better today than the person who taught them, which is exactly what should happen and still takes some getting used to. The second is that none of that value came back to the thing that created it. We trained people and watched other companies hire them.

What we could not do was keep it

It is important to be accurate about what went wrong, because the obvious reading is the wrong one.

The programme did not fail. Design 101 worked. The teaching was good, the outcomes were real, and the people who came through it went and built careers. By the only measure that matters to a student, it succeeded.

What failed was everything around it. The value could not be captured and the operation could not be sustained, because there was no institution underneath the programme, only a person. Four specific things were missing.

There was no bird’s eye view. You could message any individual member. You could not see the community. Who was active, who had gone quiet, who was ready to be pushed further, who was about to drop out. At 100 people you hold that in your head. At 2,000 you cannot, and what replaces management is triage: responding to whoever is loudest that day.

Opportunities had nowhere to land. One of the promises was that being in the community would open doors, and it did, badly. Someone would need a designer. Finding the right member meant messaging people one at a time and discovering that this one already had a full time job and that one was not taking clients. Who was available, what they did, what they were looking for: that lived in individual WhatsApp threads and nowhere else. The single strongest thing a community of 2,000 builders has to offer could not be queried.

The rules lived in one person’s head. There were values and there was culture. There was nothing documented and nothing enforced structurally, so every norm could only be applied through direct founder intervention. That does not scale and it makes the whole thing fragile.

Placement was real but accidental. The most important outcome, graduates getting work, happened without any system anyone controlled. Which means it could not be measured, could not be improved, and could not be promised to the next cohort. The value was real. The pipeline did not exist.

Underneath all four, one sentence: the community ran on the founder’s energy, and energy is not infrastructure. When the only system is a person, you do not burn out gradually. You stop.

What we built this time

The platform is not a nicer version of the old community. It is a direct answer to those four failures, feature by feature. That mapping is the whole design:

What was missing in 2017	What exists now
No bird’s eye view	An append only activity log behind every action, member and collective feeds, an idleness detector that surfaces who has gone quiet, and admin views over cohorts, pods and tasks
Opportunities could not be routed	Structured member profiles with skills, verification status and availability, a job board, and projects with defined teams. The community became queryable
Rules lived in a head	A server side Pledge every member signs, published Community Guidelines separate from the manifesto, and roles with actual permissions rather than informal authority
Placement was accidental	The verification system, which is the keystone. Skills are verified through reviewed real work, so the record exists whether or not anyone asks for it
Value leaked out of the community	A business model built on a take rate from real work rather than course fees, so the collective earns when members earn
₦4,999 could not fund the operation	Training is not the product. Work is. Capability gets built through reps on real work, where the output is an asset the collective keeps

The verification system deserves its own explanation, because it carries the thesis.

• The unit of reputation is the skill, not the role. A job role is a recipe of required skills. You are verified as a role only by being independently verified in every skill it requires. It is a computed rollup, never a credential someone granted you.
• Self assessment counts for nothing. A skill moves through claimed, nominated, verified, dormant. You claim it, reviewed real work nominates it, and only then does it verify.
• Reviews are weighted. A rater already verified in the skill they are scoring counts three times a rater who is not.
• The gates are real. A task review requires you to be that task’s named reviewer and the work to be complete. A project review requires the project to have shipped and both people to have actually been on the team.
• Work that did not go through the platform does not exist for verification, because nobody accountable reviewed it.

Some deliberate decisions that go against the old model. Pods have no mentors, they are all peers. Pods never do client work, because belonging and delivery are different jobs and mixing them corrupts both. And the governance thresholds, the numbers that decide when a skill trips from nominated to verified, are held as tunable constants, because we do not yet know the right values and pretending otherwise would be the first lie.

Why any of this matters outside Senpai

The reason to build the record is not tidiness.

A Nigerian builder’s individual skill is invisible to the global work economy. Not undervalued, invisible. Whole platforms ban us by default. Payment systems turn people away before they have typed a word. To much of the internet, being Nigerian is a risk score, and no amount of individual talent unchecks that box.

Alone, each of us is priced by the worst thing anyone from our country has ever done. The cure is not telling people to get better, because skill was never the constraint. The cure is an institution willing to stake its own name on specific people, and to enforce the standard hard enough that the stake means something.

That is what a verified record is for. It is trust infrastructure for a locked out nation.

What we do not know

The part usually left out.

Whether the standard survives contact. It is easy to hold a high bar with no members. The test is the first person we like who does not meet it.

How to form a pod. Genuinely unresolved. Mix disciplines or concentrate them, match by timezone, ambition, or experience. We do not know.

Whether the verification thresholds are anywhere near right. Three contributions, weighted average above four, at least two distinct raters, at least one already verified. Guesses made in an empty system.

Whether structure kills the thing that worked. The old community had real warmth. Fifty people in a room at Social Media Week Lagos, saying out loud what they wanted to become. None of that came from a platform. There is a version of this where we systematise the magic out of it, and no amount of good schema protects against that.

What we are measuring

Stated now, so it cannot be chosen later to flatter the result.

1. Induction completion, as a proportion of those accepted.
2. Still contributing at 90 days. Contributing means reviewed work on the platform, not presence.
3. Members reaching a first verified skill, and how long it took.
4. Work shipped, and how much of it was peer reviewed.
5. Members who earned money through work that came via the Collective.
6. Whether a verified record changes an outcome outside Senpai.

Targets for all of these are set in advance and written down, including the line at which each one counts as a failure.

The 2017 comparison has to be used carefully, because it is easy to compare the wrong things. Design 101’s 29% completion rate says nothing useful here: a six week course and a standing community are different objects, and that cohort was open to anyone who paid where this one is hand-picked. What does transfer is the outcome. At least 30 of the 45 people who finished went on to earn in dollars, with no platform at all. If the platform does not eventually beat a WhatsApp group with a committed founder behind it, the platform was not the thing that mattered.

Number six is the only one that tests the actual thesis, and it is the slowest. One member getting a meeting they would not otherwise have got, because a client could check a record, is worth more than every other number here.

One expectation worth setting now: the first cohort will produce few verifications and take a long time doing it. The pool of people qualified to rate a skill grows only as people get verified in it, so the curve is slow and then sharp. Cohort one is not supposed to look impressive. It is supposed to produce the baseline that cohort two is measured against.

What would count as failure

If most of the founding cohort stops contributing within three months, the structure did not work. If people contribute but nothing ships, we built a community and called it a collective. If work ships but nobody gets verified, the review mechanism is too tight or nobody is using it. If people get verified and it changes nothing for them outside Senpai, the record we are manufacturing is not the signal we thought it was, and the thesis needs rebuilding rather than tuning.

And if it works but only because one person is holding it together again, then we have rebuilt the 2017 problem with better software, which would be the most expensive way possible to learn the same lesson twice.

Why publish this before we know

Two reasons, and the second is the honest one.

A baseline stated afterwards is not a baseline. We tell clients that a pilot with no measured before is a story rather than evidence. It would be difficult to keep saying that while running our own pilot on vibes.

And writing it down makes it harder to quietly rewrite later. If this goes badly, the temptation will be to find the one metric that went up and lead with that. This post exists to make that awkward.

The founding cohort is forming

If you design, build, write, shoot, or research, and you have been doing it alone, this is the invitation.

We are looking for a small number of people who are already good or obviously about to be. You will be pushed harder than is comfortable. Your work will be reviewed by people who know what they are looking at. And what you build here belongs to you and travels with you, because the entire point is a record no one can wave away.

The last time we did this, most of the people who came through it went on to earn in dollars. We have spent since then building the thing that was missing.

Applications are read by a person, not filtered by a form. The founding cohort is small on purpose, and once it closes the next one is a long way behind it.

Apply to the Senpai Collective

Not a builder, but you know one? Send this to them. That is how the good ones found us last time.

And to everyone else: we will report back, whichever way it goes.