Skip to content
Zambia Dev Wiki Zambia Dev Wiki

Where the data comes from

The majority of this wiki’s seed content was produced by mining the WhatsApp exports of two Zambian developer communities and writing down what they know. This page is the record of that exercise: the sources, the period covered, the method, and the limits.

71,492messages analysed
2WhatsApp communities
Mar 2023 – Jul 2026combined archive span
Jul 2026analysis performed

The two sources

Zambian Programmers

Developer community — building software in Zambia

Messages
40,366
Archive
20 Jun 2024 → 19 Jul 2026
Unique senders
370
Mined for
Payment gateways, mobile money, card acceptance, hosting, tooling, regulation

Tech Jobs & Opportunities

Jobs board & market chatter

Messages
31,126
Archive
6 Mar 2023 → 19 Jul 2026
Unique senders
806
Mined for
Salaries, rates, hiring demand, remote work, scam patterns

The two groups share their core membership — the most active poster in each is the same person — so the true number of distinct people across both archives is lower than the sum of the sender counts. Treat every figure here as describing the archives, not the whole Zambian developer ecosystem.

How the mining worked

  1. Export and parse. Both chats were exported in full and parsed into dated message records, folding multi-line messages and excluding system events (joins, adds, leaves). Nothing was sampled: the analysis is a census of the exports, not a skim.
  2. Quantify. Theme share-of-traffic was computed per quarter — payments, AI tooling, hiring, salaries, languages — so that repeated questions could be counted, not remembered. That is how we know the payment-gateway question was asked over 30 separate times.
  3. Hand-read the important counts. Where a number mattered, automated classifiers were checked by reading every candidate message. Example: the classifier found 274 candidate job ads; reading them left 115 real ones. The hand-read census wins wherever the two disagree.
  4. Distil. Recurring questions, verified-in-thread answers, fees, dates and disputes were extracted into wiki pages. The conversation stays behind — see what we deliberately did not take.
  5. Grade. Everything enters the wiki as hearsay until a human re-checks it against a primary source. The grading system is described in how we verify claims.

What the mining surfaced

The findings below are why this wiki exists.

  • “Which payment gateway should I use?” was asked over 30 separate times in three years in a single group — and was still being asked on the last day of the archive. It is now answered once, with dates.
  • “What does DPO charge per transaction?” was asked in September 2023 and never answered — not once, in 71,492 messages. It is tracked as an open question.
  • Payment-gateway integration is the single most durable pain point in both groups, because Stripe does not operate in Zambia and card acceptance keeps breaking. See card payments.
  • Settlement and withdrawal complaints recur for years against the same providers without resolution — the most repeated operational complaint in the archive.
  • Several questions were asked and received zero replies — withdrawal fees and minimums, Flutterwave’s recurring-payment rates, Government Service Bus integration. Each is now listed as an open question on the relevant page, so the next person can answer it once, in public.

Why everything starts as “hearsay”

The archives are genuinely knowledgeable — and verifiably wrong in places. The mining exercise documented, among other things:

  • A fee table for four payment providers that an AI assistant hallucinated into a group chat in May 2024. It read exactly like the real thing and nobody challenged it.
  • A licensing answer about Lenovo given to someone asking about Lenco — followed by a second answer about audio patents. Both were pasted AI output.
  • The same provider’s setup fee quoted at $500 and at $1,000 five months apart; another quoted at K5,000 by one person and “no setup fee” by four separate people on the same day.
  • Bank of Zambia licensing cost estimates ranging from “less than K5,000” to “$70,000 or more”, stated confidently by different people over two years, never reconciled.

This is why the wiki grades every claim, why hearsay pages carry a caution banner, and why upgrading a claim to a primary source is the most valuable contribution you can make. The full catalogue of traps is on the verification page.

What we deliberately did not take

The wiki carries the knowledge, not the conversation:

  • No names. Members were talking in a semi-private space, not publishing.
  • No phone numbers. They appear throughout the exports; they appear nowhere here.
  • No attributed quotes. Pages say “merchants report approval taking 2–3 weeks” — never who said it.

The raw exports and the intermediate analysis notes are not part of this repository and are not published, because they contain other people’s personal data. What you read on this wiki is the distillation. The privacy rules that apply to all contributions are on the contributing page.

Limitations

  • Two groups are not the ecosystem. Plenty of Zambian developers are in neither group, and what is said in a chat is not the same as what is true in the market.
  • Classifier counts are directional. Theme and sentiment figures come from keyword-based classification with known error rates; the trends they show are large enough to survive that error, but no single count should be treated as exact. Hand-read censuses (like the 115 job ads) are exact for the period they cover.
  • Membership overlaps between the two groups, so combined “unique voice” figures are approximate.
  • The archive ends on 19 July 2026. Anything that changed after that date is not reflected here until a contributor updates it — which is the point of the wiki.