The African entrepreneur-support sector produces an enormous quantity of reporting and a startlingly small quantity of learning. Annual reports carry founders trained, sessions delivered, jobs claimed and capital facilitated, all of it output and almost none of it evidence, because output tells you what a programme did and evidence tells you whether doing it worked. The distinction is not academic. A programme that has trained five hundred founders without ever checking which of them finished, which of them were still trading two years later, and what those two groups had in common on the day they arrived, has run five hundred experiments and recorded none of the results. The data usually exists. It sits in intake forms, attendance registers and the WhatsApp groups where alumni quietly report what happened, and it is almost never assembled.
Key Takeaways
- Outputs are not evidence. Founders trained is a count of activity, not a test of design.
- The cheapest useful study a programme can run is retrospective: hold the historical file and check what predicted completion.
- Most programmes already hold the data, unassembled, across intake forms, registers and alumni contact.
- Findings should change the design, and if none ever do, the evaluation is decoration.
- Small samples are not a reason to skip the work. Ninety-nine records were enough to show a three-fold effect.
- Publishing what did not work is the contribution the sector most lacks and most needs.
What does an accelerator already know without collecting anything new?
More than it thinks. Intake forms carry business status, sector, stage and stated ability to pay. Registers carry attendance. Records carry who completed and who left. Alumni contact, however informal, carries what happened next. Assembled into a single sheet with one row per founder, that is a longitudinal dataset, and the first pass through it will usually surprise the people who ran the programme.
Ours did. Across ninety-nine alumni, the business a founder held on the day they arrived predicted whether they finished with a force nothing else came close to: fifteen percent drop-out for registered trading ventures against forty-five percent for founders arriving with no business. That was not the conviction going in. Several of us would have argued that motivation and coachability mattered more, and the file disagreed.
The value of a finding like that is that it is actionable in a way an impact number never is. It changes intake weighting, it changes which product is offered to whom, and it changes what the programme promises. An output figure changes nothing, which is why output figures are comfortable.
Why do so few programmes do this?
Three reasons, all understandable and none sufficient. The first is funder incentives: reports are written for people who asked for outputs, so outputs are what gets produced, and the internal question of whether the design works never acquires an owner. The second is that the answers may be unwelcome. A programme that discovers its most enthusiastic intake segment halves is looking at a finding with budget consequences, and it is easier not to look. The third is capability: assembling the sheet takes a week of unglamorous work that nobody’s job description contains.
The honest response to all three is that a programme claiming to build evidence-based businesses while running itself on conviction has a credibility problem it will eventually be asked about. The same discipline it teaches founders, name the decision the number would change, applies to the institution. If no plausible result would alter the programme’s design, the evaluation is theatre.
What should change when the evidence arrives?
Selection, product mix and the promise. Selection, because weights derived from observed completion beat weights derived from interview impressions, and the difference shows up in the third month. Product mix, because a finding that a segment does not complete the intensive is an argument for a lighter product rather than a rejection: a founder with an idea and no trading history is better served by a cheaper tier and a return route than by a seat they will vacate in November.
The promise changes too, and this is the part institutions resist. A programme that knows its completion rates by segment cannot honestly market a uniform outcome, and should stop. Saying plainly that founders arriving with a trading business complete at a much higher rate is a harder sentence than the brochure version, and it is the sentence that makes everything else the programme says believable.
The last change is cultural. Publish what did not work. The sector is thick with success stories and thin with honest post-mortems, which means every new programme rediscovers the same failures at full cost. A single well-documented account of a cohort design that underperformed, and why, is worth more to the ecosystem than another set of headline figures, and it costs only the willingness to be seen learning.
FAQ
Is ninety-nine records enough to act on?
Enough to weight, not to legislate. It showed a three-fold gap holding across four categories. Treat it as the best available evidence and revisit annually.
What if we cannot reach alumni?
Start with what you hold: intake status and completion. Those two columns alone answer the most useful question, and neither requires contacting anyone.
Does this conflict with an access mission?
No. It sharpens it. Knowing which segment struggles lets you design a route for them instead of admitting them into a format that fails them at forty-five percent.
Who should own this?
One named person, annually, with the authority to change intake. An evaluation nobody owns is an evaluation nobody acts on.
Related Reading
- Who actually finishes: what 99 alumni say about selection
- When a diagnostic earns trust, and when it becomes a sales instrument
- The asset nobody counts: human capital in a small African firm
- Selection versus treatment: what accelerators actually change
Sources and Evidence
- Internal alumni outcomes dataset, 99 participants 2022 to 2026, business status at entry against programme completion.
- Blacksmith, N. and McCusker, M. E., Data-Driven Decision Making in Entrepreneurship, on evidence-based selection and retrospective validation.
- Clevenger, M. R. and Fortunato, M. W-P, Entrepreneurial Communities and Ecosystems, on the institutional conditions under which support programmes accumulate learning.