Africa's Regulatory Failure Is Not a Law Problem

The Public Benefit Organizations Act passed by Kenya's National Assembly in 2013 is, by the standards of civil society legislation on the continent, a well-drafted instrument. Its policy framework is clear: public benefit organisations should be recognised, enabled, and held to a transparent accountability standard. Its drafters consulted widely with civil society practitioners. Its definitions are more precise than those of the legislation it replaced — the Non-Governmental Organizations Co-ordination Act of 1990. Its regulatory architecture — an independent authority, a defined registration process, a set of obligations that are enforceable but not punitive by design — is what enabling civil society regulation should look like.

Registration rates did not meaningfully improve in the years following the Act's operationalisation, and §4 of this paper gives that claim an actual number rather than leaving it as an impression.

This is the puzzle this paper is about. And the puzzle is not unique to Kenya. Across East and West Africa, the past twenty years have produced a generation of civil society legislation that is substantially better-designed than what preceded it, and substantially less complied with than its architects anticipated.¹ The gap between the quality of the law and the compliance rates it achieves is consistent enough to invite a systemic explanation rather than a country-specific one.

The explanation this paper argues for is not the quality of the law. It is the absence of the translation layer.

Better legislation, replacing worse legislation, and the registration-rate line doesn't move.

1. What the PBO Case Can Carry, and What It Can't

What it argues from. One well-documented Kenyan case — the PBO Act's transition — with an actual post-commencement registration figure (§4). A cross-domain pattern inside Kenya's own regulatory landscape (Companies Act, KRA's iTax, Water Act reporting) offered as repetition of the same structure, not as independent statistical evidence. Two pieces of US tax-compliance research used for a general mechanism — that software reducing the effective complexity of a rule system measurably changes compliance behaviour — not as evidence about Kenya specifically.²³

What it does not argue from. It does not argue from a controlled comparison of registration rates under "good law, no translation layer" against "good law, with translation layer" — no African jurisdiction has run that comparison, as far as this paper's research found. It does not argue that the US tax-filing research generalises mechanically to Kenyan civil-society registration; the domains differ (individual taxpayers resisting a cost versus civil-society organisations navigating a one-time registration requirement), and §7 names this directly as a limitation on the analogy rather than treating it as settled.

What it does not claim. It does not claim the PBO Act's registration shortfall is caused solely, or even primarily, by the absence of a translation layer — §7 takes seriously the competing explanation that fee levels, awareness, and organisational capacity unrelated to any software gap also matter, and that a 2025 High Court ruling complicates the "fees are the barrier" story in a way this paper's own prior work has flagged.⁴ It does not claim the specific cross-domain examples in §3 have been independently measured to the same standard as the PBO case.

What would refute it. Given in full in §9. The sharpest version: if registration or filing rates under PBO Act-equivalent frameworks in other African jurisdictions improve substantially without any translation-layer infrastructure being built — through enforcement, fee removal, or awareness campaigns alone — the translation-layer explanation this paper argues for would be shown to be unnecessary where it claims to be central, and the paper's framing would need to be narrowed substantially rather than merely qualified.

2. The Translation Gap Defined

A regulatory framework has two components. The first is the normative component: what the law requires, what the penalties for non-compliance are, what the rights and obligations of regulated parties are. This is what legislators and lawyers produce. It is what gets debated in parliament, published in the Kenya Gazette, and argued about in courts.

The second component is the operational component: how a regulated party, sitting in an office or a community centre or a rural clinic, navigates from "I need to comply with this" to "I can demonstrate that I do." This is the translation layer — the set of tools, knowledge, advisory services, and institutions that make the normative framework navigable for the organisations it governs.

In well-resourced regulatory environments, the translation layer is dense. US tax compliance, for example, is supported by commercial and free-assisted software that encodes the tax code's requirements into a process a non-specialist can navigate — and, as §6 shows with a directly measured figure, that encoding changes error rates, not merely convenience. The law is not simpler in these environments. But the translation layer is developed enough that the effective complexity — the complexity actually experienced by someone trying to comply — is low.

In Kenya, and across much of Africa's regulatory landscape, the translation layer is thin. This is not primarily a criticism of regulators, who are typically operating with limited resources and cannot themselves provide the advisory and support infrastructure that would make compliance accessible. It is a structural observation about where investment has and has not gone. Significant effort has gone into improving the normative component: better laws, clearer regulations, more enabling policy frameworks. Comparatively little effort has gone into building the translation layer — the infrastructure, software or otherwise, that would make those better laws navigable.

3. The Pattern Across Domains

This pattern repeats across regulatory domains in Kenya with enough consistency to be worth treating as structural rather than sector-specific — though none of these, other than the PBO case taken up in §4, has been independently measured to the standard this paper's central claim would need.

The Companies Act, No. 17 of 2015, modernised Kenya's company law framework substantially. Its reporting requirements for small and medium enterprises are more clearly specified than those of its predecessor. Anecdotal and practitioner reporting suggests SME compliance rates with annual filing requirements have not improved at the pace the reforms anticipated, though this paper did not find a rigorously measured, published figure quantifying that gap, and does not assert one.

The Kenya Revenue Authority's iTax system is a genuine improvement on the paper-based system it replaced in the sense that it digitises income-tax filing into a navigable interface. Whether it has produced a measured improvement in compliance rates specifically — as distinct from filing convenience, or KRA's own reported subscriber counts — is a claim this paper searched for a verifiable source on and did not find one that met this programme's sourcing standard; the available Kenyan academic literature on the question consists of unpublished graduate theses with figures this paper could not independently verify. This is registered as CITATION-NEEDED in Appendix B rather than estimated or borrowed from a source that could not be checked.

In the environmental sector, Kenya's Environmental Management and Co-ordination Act imposes specific reporting requirements on projects above certain environmental impact thresholds. Reported compliance among small and medium enterprises is low — attributed, in the practitioner accounts this paper drew on, less to ignorance of the law than to the absence of systems that make the reporting process navigable without specialist environmental consultants. This paper has not independently verified a compliance-rate figure for this domain either.

The WASH sector provides the most direct evidence from Flux's own work; the conflict of interest in citing it is worth flagging before doing so. Kenya's Water Act, No. 43 of 2016, imposes data reporting obligations on licensed water service providers. The data utilities are required to report on service delivery, water quality, and network performance is, in principle, available in their operational systems. In Flux's own field deployment experience, many utilities report it manually, using paper aggregation processes that introduce errors and delays that make the data unreliable as a regulatory signal — a constraint AccessWASH's deployment work has repeatedly found to be the absence of operational data infrastructure, not an absence of willingness to report.⁵ This is internal, unaudited observation from Flux's own product deployments, not an independently verified dataset, and should be read with that limitation attached.

4. What Actually Happened to PBO Registration

The puzzle stated in the introduction deserves an actual number, not just an impression.

The PBO Act, assented to in January 2013, did not come into force until over a decade later — commenced by Legal Notice No. 78 of 2024, effective 14 May 2024, which repealed the NGO Co-ordination Act of 1990. Organisations registered under the old framework were deemed transitioned but required to apply afresh under the new one within a defined window; that transition deadline was later extended, by a gazette notice issued 16 May 2025, to 13 May 2026.

Against an estimated population of roughly 14,000 NGOs operating under the old framework, a mid-2025 regulator statement put the number that had completed transition at approximately 4,000 — meaning, on that count, well under a third of the estimated population had transitioned more than a year after the new Act came into force, despite the Act itself being, as the introduction argues, a clearer and better-drafted instrument than what it replaced. In the year to mid-2026, 4,779 organisations were de-registered.

The clearest complication to the "the barrier is cost and process friction" reading of this gap comes from the Kenyan courts, not from this paper's own argument. In Otieno & 2 others v Attorney General & another; Katiba Institute & 9 others, Petition E519 of 2024, the High Court [2025] KEHC 8557 (KLR) struck down the requirement for mandatory re-registration and compulsory donor-data disclosure, and ordered automatic transition without a fresh application or fees — removing, by judicial order, exactly the kind of friction a translation layer would also be built to remove.⁴ The queue still did not clear. That is the single strongest piece of evidence in this paper's own corpus against a pure "translation-layer-and-nothing-else" reading of the registration gap: when a court order removed the administrative and financial barrier directly, registration did not visibly accelerate to match. §7 takes this up as the paper's own strongest objection to itself rather than setting it aside.

What this section can responsibly conclude is narrower than what the introduction's framing implies: registration under better legislation remained low months after both the Act's own grace-period extension and a court order removing its most-cited procedural barriers. The translation-layer explanation argued for in this paper is a candidate explanation for part of that gap — the part concerning organisations that want to comply and cannot navigate the requirements unaided — not a complete account of all of it, and this paper does not have the data to apportion the gap between "cannot navigate" and other causes (awareness, organisational capacity, indifference) with any precision.

5. What Is Encodable and What Is Not

Not all compliance requirements are equally amenable to software encoding. Requirements that involve genuine legal judgment — determining whether an organisation's activities qualify as "public benefit" within the meaning of the Act, or whether a specific transaction constitutes a conflict of interest in a novel factual context — require human interpretation. These are genuinely hard cases, and this paper is not suggesting software replaces the judgment they require.

But a significant proportion of regulatory compliance requirements are procedural rather than interpretive. They ask: is this provision present? Has this return been filed? Does this board composition satisfy this requirement? These are checkable. They are automatable. And they are scalable in ways human advisory services are not.

The question of where to draw the line is not theoretical — it is a practical question that software builders working in specific regulatory domains have to answer by examining each requirement in detail. PBOMaster's experience building a constitution analyser against the PBO Act's own schedule is instructive here, and is examined on its own terms, independent of this paper's broader argument, in What a language model finds when it reads your constitution: the large majority of the Act's constitutional requirements turned out to be procedural. The hard cases — the public benefit objectives determination being the clearest example — are a small fraction of the total compliance burden. The encodable fraction is, on this evidence, large enough that software can close a meaningful part of the compliance gap even without addressing the genuinely interpretive cases — though "a meaningful part" is doing real work in that sentence, and this paper has not measured what fraction.

6. The Translation Layer, Measured

The argument that a translation layer changes outcomes, not merely convenience, is not original to this paper — it has been tested, directly, in a domain with better data than Kenyan civil-society registration currently offers: US federal income tax filing. The economics-of-compliance literature treats this as a live, contested question rather than a settled one: both the magnitude of non-compliance and the mechanisms that reduce it remain, on the discipline's own account, harder to pin down than headline enforcement numbers suggest, since most of what regulators observe is detected non-compliance rather than the true underlying rate.⁶⁷ That caution matters here: the measured figures in this section describe a specific, narrow outcome (a math-error notice) in a well-instrumented system, not a general solved formula for "compliance."

A 2018 randomized outreach study by IRS and academic researchers sent postcards to a sample of paper filers describing free assisted tax-preparation options (software via the Free File program, or in-person help through the Voluntary Income Tax Assistance program) and tracked the effect on filing method and, separately, reported baseline figures on error rates by preparation method.³ The baseline figures are the ones this paper draws on: in the 2015 filing season, the IRS sent a math-error notice to only about half of one percent of taxpayers who used an assisted preparation method, against over 16 percent of paper filers — a thirty-fold difference in the rate of a specific, measurable error. The same research also documents a large gap in processing cost (roughly $4.00 per paper return against $0.25 for an electronic one) and in refund turnaround (about three weeks against five days).

What this does and does not establish for the Kenyan case is worth separating out clearly. It establishes that, in a well-studied domain with a much denser translation layer than Kenyan civil-society registration currently has, the presence or absence of an assisted/software-mediated path is associated with a very large, directly measured difference in a compliance-adjacent outcome (error rate) — not merely a difference in stated convenience or satisfaction. It does not establish that the same mechanism, at the same magnitude, applies to PBO registration, which is a one-time, higher-stakes process rather than an annual, lower-stakes one, and where the "error" being prevented is a missing or non-compliant clause rather than an arithmetic mistake. The analogy is structural — software reducing effective complexity measurably changes outcomes — not a claim that Kenyan civil society would see a thirty-fold change from the same intervention.

A separate 2006 analysis, written to support a US return-free filing proposal rather than to study software adoption directly, is relevant for a narrower reason: it quantifies how large the pure compliance-cost burden of a complex rule system can be even before considering error rates — US individual income tax compliance costs were estimated at roughly 10 percent of revenue collected, with a proposed simplification projected to save up to 225 million hours and over $2 billion a year for the taxpayers eligible to use it.² This paper cites it only for the general scale such a burden can reach, not as evidence about tax software specifically — the return-free proposal it describes is a different mechanism (government-prefilled returns) from the commercial or assisted software this section's main point rests on.

7. Objections I take seriously

"The clearest counter-evidence to this paper's own thesis is in this paper." §4 is explicit about this: when a Kenyan court order removed the PBO Act's most-cited procedural barriers directly — striking down mandatory re-registration and compulsory donor-data disclosure — registration did not visibly accelerate. If the translation-layer explanation were the dominant cause of the registration gap, removing a major procedural barrier by judicial fiat should have produced a visible response. It is not clear that it did, on the evidence available to this paper. The honest response is not to explain this away but to narrow the claim: the translation-layer argument is best read as an account of part of the gap — the organisations that want to register and cannot navigate the requirements unaided — not a complete account of why registration overall has stayed low. Awareness, organisational capacity, and indifference toward a transition deadline that has itself already been extended once are all live competing explanations this paper has not ruled out.

"Flux has a commercial interest in this conclusion." Flux builds PBOMaster, the product used in §5 to illustrate what is encodable in the PBO Act's constitutional requirements, and sells software premised on exactly the "translation layer" argument this paper makes. This is a direct conflict. The honest mitigation is narrow: §6's grounding evidence (the IRS outreach study, the Goolsbee compliance-cost analysis) is independent of Flux and of Kenya, and establishes the general mechanism is real elsewhere; it does not establish that PBOMaster specifically closes the Kenyan gap at any particular rate, and this paper has not measured that.

"The US tax-filing analogy may not transfer to one-time civil-society registration." Tax filing is annual, relatively low-stakes per instance, and already backed by decades of IRS and commercial infrastructure investment the Kenyan case does not have an equivalent of yet. PBO registration is a one-time, higher-stakes event for an organisation that may not file again for years. The thirty-fold error-rate difference in §6 is suggestive of what a dense translation layer can do, not a prediction of what a nascent one will do for a structurally different kind of compliance event. This paper treats the US evidence as showing the mechanism is real, not as a forecast for Kenya.

"Attributing the compliance gap to a missing translation layer, rather than to fee levels, awareness, or capacity, risks being the explanation a software company would prefer to be true." This is close to the commercial-interest objection but distinct: even disinterested researchers studying this question might reach for the explanation that is easiest to build a product around. This paper does not have a way to fully rule that bias out in its own reasoning, beyond noting that §4's own evidence (the court-order natural experiment) cuts against the translation-layer explanation rather than for it, which is not what a motivated account would be expected to emphasise.

8. Limitations, and what this paper does not claim

No controlled comparison exists between "good law, no translation layer" and "good law, with translation layer" in any African jurisdiction, as far as this paper's research found.

It does not claim the translation-layer explanation accounts for the full PBO Act registration gap; §4 and §7 both treat the 2025 High Court ruling's apparent lack of effect on registration pace as the paper's own strongest piece of counter-evidence, not as a detail to minimise.

It does not claim the cross-domain examples in §3 (Companies Act, iTax, EMCA) have been measured to the same standard as the PBO case; most are practitioner reporting or unverified, and the iTax compliance-rate claim specifically is left open in Appendix B rather than sourced to an unverifiable thesis.

It does not claim the US tax-filing evidence in §6 predicts a specific magnitude of improvement for Kenyan civil-society registration; §7 states directly why the domains differ enough that the analogy should be read structurally, not quantitatively.

And it does not claim PBOMaster has measurably closed any part of the Kenyan compliance gap; that would require deployment data at a scale and rigor this paper does not have, and Flux's commercial interest in the opposite finding is named directly in §7 rather than left implicit.

9. What would falsify this

Stated as commitments, in decreasing order of how much damage each would do to this paper's claims.

Registration or filing rates under a PBO Act-equivalent framework in another African jurisdiction improve substantially without any translation-layer infrastructure being built — through enforcement, fee abolition, or awareness campaigns alone. That would show the translation-layer explanation is not necessary where this paper claims it is central, and would require narrowing the paper's scope substantially.

A rigorous, published study finds Kenyan PBO registration rates did not meaningfully change in the twelve months following the 2025 High Court ruling that struck down mandatory re-registration and donor-data disclosure requirements, confirming rather than merely suggesting what §4 already treats as the paper's sharpest self-objection. If confirmed with real post-ruling registration data (which this paper does not yet have), this would strengthen the case that non-procedural barriers — not the translation layer — are the dominant constraint, and would require demoting the translation-layer argument from primary to secondary explanation.

A verified KRA iTax compliance-rate figure, once found, shows no measurable improvement in SME filing compliance attributable to the system's introduction. That would remove one of this paper's own cross-domain illustrations (§3) and weaken, without destroying, the general pattern claim — the PBO and US evidence would still stand independently.

PBOMaster deployment data, examined rigorously rather than described anecdotally, shows no measurable effect on an organisation's probability of completing PBO registration successfully. That would not falsify the general translation-layer mechanism (§6's US evidence is independent of PBOMaster), but it would falsify the implicit claim that Flux's own product is a working instance of the mechanism this paper argues for.

Appendix A — Evidence table

Appendix B — Citation register

CITATION-NEEDED — KRA iTax's measured compliance effect. Appendix A row 9. §3's iTax discussion deliberately does not cite a compliance-rate figure; the available Kenyan sources found were unpublished graduate theses whose figures could not be independently verified, which does not meet this programme's sourcing standard.

CITATION-NEEDED — post-ruling PBO registration data. Appendix A row 10. §4 and §9 both depend on what happened to registration rates after the 2025 High Court ruling; no published data tracking that specific window was found.

CITATION-NEEDED — PBOMaster's measured registration effect. Appendix A row 11. §7 and §9 both name this directly: Flux has not measured, at a rigor beyond internal and unaudited, whether its own product changes registration outcomes.