The morning after a general election, a campaign office in a Nairobi ward is being emptied. Plastic chairs stack by the door. Someone pulls the wall map down — every polling station marked on it, turnout from the last three cycles pencilled in the margins, annotated in four different hands over six months — and folds it into a bin bag. Eleven WhatsApp groups, one per sub-location, holding the numbers of roughly four hundred people who had knocked on doors, will buzz for another fortnight and then go quiet. The notebooks the ward coordinators carried, the ones in which they wrote down what people said they needed — water here, a footbridge there, three schools without a perimeter wall — are in a box under a trestle table. Nobody will open the box.
This happens in every ward in the country, in the same week, every five years.
The conventional reading is that a campaign ended and normal life resumed. I want to argue that something substantial was destroyed in that room, that the destruction is structural rather than careless, and that it explains a failure this body of work has already documented from the opposite end.
Built from nothing in six months. Gone the moment it succeeds — the data goes with it.
1. Direct observation, not a survey, and what would break it
This is an argument paper built on direct observation of campaign operations across cycles, the documentary record of how Kenyan election and data-protection law actually applied in 2022, and reasoning from institutional design — not a survey, stated plainly before the argument rather than left for the footnotes to imply a stronger evidentiary base than exists.
What it argues from. Three things. First, first-hand observation of how competitive constituency-level campaigns are actually staffed, organised and run, gathered across more than one cycle rather than from a single case. I have watched teams build this; I have not run a structured study of it, and where a claim depends on how representative my observations are, §12 says so. Second, the documentary and reported record of Kenya's 2022 general election as the first conducted under the Data Protection Act, 2019 — the Act itself, the 2022-election final report jointly published by the Carter Center and Privacy International, and the constitutional and statutory counts of constituencies and wards, all cited below rather than asserted. Third, an argument from institutional design: that a predictable, repeated loss can be structural without being anyone's fault, which is the series' own prior method rather than a new one.
What it does not argue from. No systematic survey of campaigns, and no measured distribution of what a "needs assessment" gathered this way is actually worth against a proper enumeration. No count of how many constituency offices, if any, have ever inherited a structured record from a preceding campaign — §9 states the negative claim as the limit of what could be established, not as a verified zero. No enforcement data specific to campaign contact-list provenance; §11's citation register carries that gap forward rather than filling it with an inference.
What would refute it. Given in full in §12, in decreasing order of damage. The headline version: if a constituency office is found that began its term holding a structured, sourced record of what the preceding campaign's constituents asked for, the central claim — that this has never happened and is structurally prevented from happening — is wrong, and I would want that found rather than continue asserting a negative I cannot fully verify.
2. What a campaign is, operationally
Set the rhetoric aside and look at the org chart.
A serious campaign for a parliamentary seat assembles, inside two quarters: a field organisation with coverage down to sub-location; a volunteer roster in the hundreds; a logistics function moving people and materials daily across a constituency; a finance operation handling many small disbursements under time pressure; a communications team; and a continuous stream of first-hand information about what constituents say they need, gathered by people who went to their doors and asked.
I have watched teams do this with no prior institutional existence, no permanent staff, and no systems beyond a phone. It is genuinely impressive, and it is not rare — it is what campaigning at that level requires, so it happens in hundreds of constituencies simultaneously. The scale is worth stating precisely rather than impressionistically: Kenya has 290 constituencies under Article 89 of the Constitution of Kenya, 2010, subdivided into 1,450 county-assembly wards1 — an average of exactly five wards per constituency, though the real distribution varies. A competitive general election is, by construction, several hundred of these organisations being built from nothing at once, on the same six-month clock, almost all of which will be dissolved in the same week.
The scale a campaign's own data is built, and lost, on each election night (Article 89, Appendix A rows 1-2).
Then consider what replaces it. The constituency office that the winner takes over has a handful of staff, a caseload it did not choose, and — this is the part that matters — none of the six months of accumulated knowledge that preceded it. The field organisation is gone. The volunteer roster is a set of WhatsApp groups nobody administers. The ward map is in a bin bag. The notebooks are in the box.
The most capable version of that political organisation existed the day before it took office.
3. The largest needs assessment nobody keeps
Of everything a campaign builds, the piece I would argue is most valuable is the one least likely to survive: the record of what people asked for.
Campaigning at ward level is, functionally, door-to-door primary research. Over six months a team will have thousands of unstructured conversations in which residents state, unprompted and specifically, what is wrong where they live. Nothing else in Kenyan public life systematically collects that at that density. Not the census, which asks different questions. Not the county, which mostly hears from people who already know how to complain. Not the constituency office, which hears from whoever walks in.
Every five years, this country conducts its largest informal needs assessment, and then discards the results on election night.
This information is not rigorous — it is self-selected, it is collected by people with an obvious interest in hearing certain answers, and it over-represents whoever was home and willing to talk. It is not a substitute for proper enumeration. But it is real, it is specific, it is geographically precise, and there is a very large quantity of it — on the order of a thousand or more such conversations per constituency across a campaign season, by my own observation rather than any count I can cite. As a starting backlog for a constituency office, it is better than the nothing that office currently starts with.
4. Why it dies
The loss is not caused by carelessness. Three structural reasons, none of which are anybody's fault in particular.
The campaign and the office are legally and organisationally distinct. A campaign is a temporary vehicle with its own funding, staffing and lifespan. The office is a state-adjacent institution with its own rules. There is no handover procedure, because in law and in practice these are not the same body. Nothing is designed to pass between them.
The people disperse. Campaign staff are hired for a season. Volunteers return to their work. Whatever they knew leaves with them, and most of it was never written anywhere except in a notebook or a group chat.
There is no format for it. Even where the intent to keep something exists, the material is unstructured — chat logs, handwritten notes, a map with pencil marks. There is nothing to migrate it into. A constituency office has no case-management system waiting to receive a backlog, because as Democracy Is Infrastructure argued, the office mostly has no operational infrastructure at all.
Each of these is individually reasonable. Together they guarantee that institutional memory is lost by construction, on a five-year clock.
5. The cost compounds
If this happened once it would be waste. Happening every cycle, it produces something worse: a political system in which no constituency ever accumulates operational knowledge about itself.
Consider the second-term MP. She has now run two campaigns and served one term. In principle she should be starting her second term with a decade of structured understanding of her constituency — which wards raise which problems, which of the promises made in the last cycle were delivered, which were quietly dropped, what changed. In practice she starts each cycle rebuilding the same map from the same blank paper, because the previous version went into a bin bag in 2022.
And the citizen's side of this is the one that corrodes trust. A promise made at a door in April cannot be tracked, because there is no record it was made. Not because anyone concealed it — because nobody wrote it down in a form that survives. The widespread belief that politicians simply lie about their intentions is, at least in part, a description of a system with no memory. You cannot hold anyone to a commitment that was never recorded.
6. The part that survives, and how
Something does survive election night, and the manner of its survival is instructive, and it is the sharpest piece of evidence in this paper.
Phone lists survive. They persist in spreadsheets, get copied between staffers, and reappear in the next cycle with nobody able to say where they came from. In the last election it was common to encounter contact databases of tens of thousands of numbers whose provenance nobody in the room could account for.
That matters legally, and it is worth being precise about the legal record rather than gesturing at it. Kenya's 2022 general election was the first conducted under the Data Protection Act, 2019, which came into force in November 2019 and applies squarely to political campaigning, party recruitment and voter-register inspection, under the oversight of the Office of the Data Protection Commissioner (ODPC).2 The law is not the missing piece; it existed, with time to be understood, before the campaign season that this paper is about.
The clearest documented instance of the law meeting campaign practice is a 2021 incident, not the election itself. A 2021 upgrade to the Independent Party Membership database let voters check, via a government portal, which political party they were registered to. The check surfaced a problem: many Kenyans discovered they had been registered as party members without their knowledge or consent — a recurrence of a near-identical incident in 2017, before the Data Protection Act existed to give it a remedy.3 The Data Protection Commissioner stated publicly that her office received over 200 complaints from aggrieved individuals, and that she had met with the Office of the Registrar of Political Parties to have the complainants deregistered.3 The Carter Center and Privacy International's joint final report on the 2022 election independently confirms the same figure and adds the institutional consequence: the complaints, made the year before the election, were acted on in consultation with the Registrar of Political Parties and led to additional consent mechanisms being built into the registration portal.4
Read that sequence carefully, because the shape of it is the paper's whole argument in miniature. The law existed. The regulator responded to a concrete complaint at real volume — over 200 individuals, a number large enough to force a public statement and an inter-agency meeting — within roughly a year of the Act applying to an election cycle. The system worked, for the narrow case of unconsented party-membership listing, because there was a specific portal, a specific registrar, and a specific point of complaint. What it does not show is any comparable mechanism for the much larger and more diffuse problem this paper is about: campaign contact databases of unknown provenance, which are not registered anywhere, do not have a portal to check them against, and have no equivalent of the Registrar of Political Parties on the other end of a complaint. The ODPC could act because the party-membership list is a named, structured, queryable thing. A WhatsApp-era phone list passed between campaign staffers is none of those things, and the same regulatory machinery that worked once has nothing to grip onto in the case this paper cares about.
But notice the shape of the failure this paper is about, which the complaints case throws into relief by contrast. The only artefact that reliably survives a campaign is the one nobody designed for survival, and it survives in the worst possible form — undocumented, unconsented, untraceable. The valuable, benign material (what people said they needed) evaporates. The sensitive material (who they are and how to reach them) persists lawlessly. That is exactly backwards, and it is what happens when an institution has no designed continuity layer: what endures is whatever is easiest to copy, not whatever is most worth keeping.
7. Objections I take seriously
"This is a diagnosis in search of a product." A fair concern, and one I should concede the shape of before defending the substance: Flux is building infrastructure in exactly the direction §10 describes. The mitigation I can offer, in the same spirit as the series' prior commercial-interest disclosures, is that this paper's claim is specified narrowly enough to be checked independently of anything Flux ships — §12's falsification test is answerable by looking at any constituency office in the country, with no reference to Flux's own product at all.
"Campaigns are not neutral collectors, so the needs-assessment claim is overstated." Already conceded in §3: this data is self-selected, gathered by interested parties, and skewed toward whoever was willing to talk. The claim is comparative, not absolute — that it beats the nothing a constituency office currently starts with, not that it rivals a proper enumeration. A reader who thinks even the comparative claim is too generous should treat §3 as the paper's weakest premise and weight §9's falsification test accordingly.
"The destruction might be functional, not wasteful — campaign promises are not supposed to bind, and a system that cannot be held to them is working as designed." This is the most serious objection to the normative thrust of §5, and I do not think the paper fully answers it. If Kenyan political practice genuinely treats door-to-door promises as non-binding rhetoric rather than commitments, then the "memory" this paper wants built would mostly produce grounds for accusations of broken promises, not accountability — a different and more adversarial outcome than the one §5 implies. I take the trust argument in §5 to be about citizens' ability to check, not a claim that every promise should be enforceable, but the line between the two is doing real work in this paper and deserves more scrutiny than it gets here.
"The ODPC complaints case shows the system does work; you are cherry-picking the cases where it does not." §6 tries to pre-empt exactly this by using the complaints case as the strongest evidence for the paper rather than against it — the system worked precisely where there was a named, structured, complainable thing to act on, which is the condition this paper's whole argument says is missing for campaign contact data generally. A reader unconvinced by that distinction should treat it as the load-bearing move in §6 and examine it directly rather than take the framing on trust.
8. What follows
If this argument is right, the useful intervention is not better campaign tooling in the sense of helping campaigns win. Campaigns are already, by the standards of the institutions around them, unusually effective. It is continuity: infrastructure whose job is to make the operationally valuable part of a campaign survive contact with the office that follows it, and to make the sensitive part accountable while it exists.
Concretely that means a campaign's record of constituent needs being structured from the moment it is captured, geographically anchored, and in a form a constituency office can inherit on day one as a case backlog rather than as a box. It means promises being recorded when they are made, so that delivery can be checked later by anyone. And it means the contact data that currently outlives every campaign doing so with documented provenance and consent, or not at all — which the ODPC complaints case in §6 shows is both legally required and institutionally achievable once there is a specific thing to hold accountable.
This is the direction Flux's own campaign-infrastructure work is heading, and I will write about that product specifically when there is something to report rather than something to announce. The argument here does not depend on it, and §7 already addresses the conflict of interest directly.
9. The claim I would defend
Narrower than the product. The five-year destruction of campaign capacity is not an unfortunate side effect of electoral politics; it is the largest single source of institutional amnesia in Kenyan constituency governance, and it is repairable without changing any law. Nothing in the Constitution requires the wall map to go in the bin bag. It goes in the bin bag because nobody built the thing that would have kept it.
As far as I can establish, no Kenyan constituency office has begun a term holding a structured, sourced record of what its constituents asked for during the campaign that put it there. That is a negative claim, built on direct observation and the absence of any reported counter-example I could find, not a verified zero across all 290 constituencies, and §11 registers it as such.
10. What would falsify this
Stated as commitments, in decreasing order of how badly each would damage the argument.
A constituency office is found that began its current term holding a structured, sourced record of its preceding campaign's constituent requests. This is the direct test stated in §9. One confirmed instance does not merely soften the claim — it falsifies the "structurally prevented" part of the argument outright, and I would need to explain why the mechanism in §4 failed to apply there instead of asserting it applies everywhere else.
A systematic survey finds that the self-selected needs-data campaigns gather is not, in fact, better than nothing as a constituency-office starting backlog — for instance, because it is so skewed toward vocal or accessible households that acting on it directly would misallocate a constituency office's limited attention. That would not rescue the destruction described in §4–§6, but it would undercut §3's claim about what is actually being lost, and the paper's normative urgency would need to be revised downward.
Enforcement action is found against a political actor specifically for the provenance of an inherited campaign contact database, comparable in specificity to the 2021 party-membership complaints in §6. That would show the regulatory gap identified in §6 is narrower than this paper claims, and the "nothing to grip onto" framing would need revision.
The functional-promise objection in §7 is shown to be the dominant view among the constituents a campaign actually serves — that is, that voters themselves do not expect door-to-door commitments to be tracked or enforced. That would not falsify §4's account of why campaign knowledge is destroyed, but it would remove much of §5's normative force for building a record of it.
11. Where the evidence runs out
No measured value exists for the needs-data campaigns collect, against a proper enumeration or against each other — §3's claim stays comparative and qualitative.
It does not claim a verified count of constituency offices that have, or have not, inherited a structured record from a preceding campaign. §9 states a negative claim bounded by what I could establish, not an audited zero.
It does not claim the Data Protection Act's enforcement record is weak in general — §6 shows it responding forcefully to a concrete, structured complaint. The claim is narrower: that the Act currently has nothing to grip onto in the specific, diffuse case of inherited campaign contact lists, which is a different and more specific claim than "the law is not enforced."
It does not resolve the objection in §7 about whether campaign promises are understood, by the people who hear them, as commitments at all — a question this paper is not equipped to answer from the evidence it has.
And it does not depend on any product Flux builds. The falsifiable claim in §9 and §10 is checkable against any constituency office in the country, with or without Flux's involvement in building the fix.
Appendix A — Evidence table
Sourced rows are cited in the footnotes. The unverified rows are measurements this argument needs and does not have; they are left empty rather than estimated.
| # | Claim used in this paper | Value | Source |
|---|---|---|---|
| 1 | Kenya's constituency count | 290, under Article 89 of the Constitution of Kenya, 2010 | C1 |
| 2 | Kenya's county-assembly ward count | 1,450 | C1 |
| 3 | Data Protection Act, 2019 in force for the 2022 general election | First general election conducted under the Act | G2 |
| 4 | ODPC complaints over unconsented political-party-membership registration | Over 200 individuals (2021, stated publicly by the Commissioner) | P3 |
| 5 | Same complaints, independently confirmed; consent mechanisms added to the registration portal as a result | Over 200 complaints, "the year prior to the election" | R4 |
| 6 | Volume of unstructured, specific need-statements gathered per constituency across a campaign season | — (order of a thousand or more, by observation, not a count) | — |
| 7 | Number of constituency offices nationally that have inherited a structured record from a preceding campaign | — | — |
| 8 | Measured value of campaign-gathered needs data against a proper enumeration | — | — |
| 9 | Enforcement actions specifically against inherited campaign contact-database provenance | — | — |
C = Constitution/statutory count. G = governing statute. P = press-reported regulator statement. R = independent report (Carter Center/Privacy International), read directly rather than taken secondhand.
Row 7 is the one that would most directly test this paper: a single confirmed instance falsifies the "structurally prevented" claim outright.
Appendix B — Citation register
CITATION-NEEDED — enforcement action against campaign contact-database provenance. Appendix A row 9. No case has been found, by the author, of a political actor facing enforcement specifically over the provenance of an inherited campaign contact database, as distinct from the 2021 party-membership complaints cited in §6, which concerned a different and more structured dataset. This is stated in §6 as the current limit of the regulatory mechanism's reach, not as a verified absence across all enforcement activity.
CITATION-NEEDED — measured value of campaign-gathered needs data. Appendix A row 8. §3's claim that this data, however unrigorous, is better than the nothing a constituency office currently starts with is argued from first principles and direct observation, not measured against a proper enumeration or against outcomes where such data was and was not used.
CITATION-NEEDED — national count of constituency offices with an inherited structured backlog. Appendix A row 7. §9's negative claim ("as far as I can establish, this has not happened") is bounded by what the author could find reported, not an audited survey of all 290 constituencies.
Constitution of Kenya, 2010, Article 89, which caps the number of constituencies at 290 and provides for their periodic delimitation by the Independent Electoral and Boundaries Commission; county-assembly wards, delimited under the same framework, number 1,450.↩
Data Protection Act, 2019 (No. 24 of 2019), Laws of Kenya, in force from November 2019 and applying to the processing of personal data including in political campaigning, party recruitment and voter-register inspection, with oversight from the Office of the Data Protection Commissioner.↩
Office of the Data Protection Commissioner, public statement, 25 June 2021, on alleged use of personal data in registration to political parties without consent, following a 2021 upgrade to the online party-membership-verification portal that let voters discover party registrations they had not consented to — a recurrence of a similar 2017 incident, before the Act existed. The Commissioner stated her office had received over 200 complaints from aggrieved individuals and had met with the Office of the Registrar of Political Parties to have them deregistered.↩
The Carter Center and Privacy International, Carter Center Election Expert Mission to Kenya: 2022 General Elections Final Report (March 2023). Confirms independently that "the year prior to the election, over 200 complaints were made to the Office of the Data Protection Commissioner by individuals who learned their signatures had been registered as members of political parties without their knowledge or consent," and that "these complaints were acted on in consultation with the Office of the Registrar of Political Parties and led to safeguards, including additional consent mechanisms' being built into the portal." Also confirms the Data Protection Act's November 2019 commencement date and the 2022 election's status as the first conducted under it.↩
