Enterprise capability statement
VoiceInsights Africa is research and decision-intelligence infrastructure for institutions collecting, governing, analysing and acting on evidence across African contexts.
What the platform is for
An institution running programmes across African countries has the same problem in every one of them: the people whose lives the programme is meant to change are the hardest to hear from, and by the time their answers become something a director can act on, the answers are old, the method is undocumented and nobody can say which response supports which finding.
VoiceInsights is the infrastructure between those two points. A person answers in their own language, on the device they already have. What they said becomes governed evidence with its source attached. That evidence becomes analysis, a publication, a recommendation and a recorded decision — and every step back to the original answer stays walkable.
Who it is built for
- NGOs and INGOs running multi-country programmes with M&E obligations to donors.
- Governments collecting citizen feedback, assessing facilities and monitoring service delivery.
- Donors and development partners needing comparable evidence across a portfolio.
- Research firms and institutions running studies that have to survive peer challenge.
Capability at a glance
| Capability | What we can say | What that means |
|---|---|---|
| Research designs — live | 4 — Structured survey, Facility assessment, Citizen feedback, Qualitative research | Governed structures and routes built specifically for the design. |
| Research designs — supported | 3 — Mixed methods, Observation and field assessment, Longitudinal and panel | Run on the platform’s shared research model. Each states what it cannot do. |
| Collection routes | 5 — Phone call, WhatsApp, SMS, Web link, Offline field app | Each collects a response today, with consent captured on the route before the questions. |
| Integrations exchanging data | 3 live, 5 configured per deployment | What moves data today, or is set up with your team during implementation. |
| Formats implemented, never connected | 2 — XLSForm instruments (ODK, KoboToolbox), DHIS2 | We read or write the format and have contacted no server of that kind. Connecting one is implementation work. |
| Language capability axes | 6 — Text, Voice, Transcription, Translation, Analysis, Reporting | Recorded separately per language. A language can pass one and fail the next, and we do not add them up. |
| Institutional editions | 5 — English, French, Portuguese, Swahili, Arabic | The interface and institutional publication editions. Collection reaches further than this; validated analysis reaches less far. |
Research designs
The distinction the table above draws between a live design and a supported one is not marketing hedging: a live design has governed structures and routes built for it; a supported one runs on the platform’s shared research model, and each states exactly what it does not do. Every design, with its limits.
Interview formats — key informant, in-depth and focus group — are formats within qualitative research rather than separate designs, because no module, route or table exists for any of them. Baseline and endline are study phases, not designs.
Collection
Every route listed above collects a response today. Consent is captured on the route itself, before the questions, on every one of them. What each route reaches, and what it costs you.
The offline application is the one that decides whether a programme works: assignments and questionnaires download to the device, interviews are captured with no connection, and the queue syncs when a connection returns. Cached state never mints server authority — every action is re-authorised by the server on reconnection.
Language
“Do you support my language” is six questions, not one — text, voice, transcription, translation, analysis and reporting — and a language can pass one and fail the next. We record them separately and never add them into a single number of supported languages. The registry carries every language we have registered; the number VoiceInsights has itself measured is reported separately, per capability, and is currently zero on every axis. The registry, read live.
The platform performs no automatic language detection, publishes no speech-recognition accuracy figure for any African language, and governs translations rather than producing them.
Multi-country programmes
One programme can operate across several countries inside one organisation. A country deployment resolves through the canonical country registry and an unrecognised code is refused. Projects attach to a programme by reference and never leave their tenant. Reporting is structurally supported at project, country and programme level.
Access is narrowed by explicit grant: one person, one scope, one stated purpose, an expiry. A grant can only narrow what a role allows and can never reach past the organisation. How the structure works.
Evidence governance
- Every reported statistic is computed by deterministic code, never by a language model. A model may explain a result; it never produces one.
- A non-probability sample is refused representativeness regardless of what the study declares, and cannot be post-stratified.
- Every figure states its weighting state, its method, its n and its interval — the four travel together.
- A quotation stays attached to the response it came from; a translation is never presented as the original.
- A finding whose evidence does not resolve does not publish. The export is refused rather than produced with a caveat.
- Retention, erasure and legal hold are an executable lifecycle with a dry run and an approval, and a legal hold blocks an approved erasure.
What the platform produces
From one evidence base: interactive dashboards, research and evaluation reports, publications, executive and decision products, datasets with their codebooks, evidence packs, communication material, knowledge products and AI-assisted intelligence. Twenty-eight product families across nine areas, each carrying its real production status rather than a uniform claim. The catalogue.
Deployment, security and assurance
VoiceInsights is delivered as managed software. An organisation is approved and provisioned; people inside it are invited; roles are granted in the platform by someone who may grant them. There is no public sign-up.
Every tenant-scoped route filters by organisation inside the SQL predicate, and a request for another tenant’s object returns not-found rather than forbidden, so the existence of another customer’s data is never disclosed. Enterprise SSO uses OIDC with PKCE; SCIM 2.0 provisioning is implemented against the RFC.
What we do not hold
- No SOC 2 report, no ISO/IEC 27001 certificate, no independent penetration test, and no independently audited accessibility conformance. Each is listed in the Procurement Center as requiring an independent party.
- No executed data processing agreement is recorded with any subprocessor.
- Jurisdiction controls exist in the underlying infrastructure and are not currently configured; we will not tell you data stays in a named country.
The trust register, with every claim and its verification state · Subprocessors and processing locations
Implementation and support
Five stages — scope, configure, pilot, train, operate — with what each produces and what we need from your team at each one. Training is role-based and runs against the deployed product using the same guides published in the Help Center, so training material and documentation cannot diverge. The approach in full.
Commercial engagement
VoiceInsights is bought as an implementation bundle scoped to a programme, and no price is published: six bundles describe coherent ways to operate the platform, and the commercial terms are shaped around the programme in a proposal. Engagement runs through that proposal and a purchase order rather than a card checkout, because that is how the institutions this is built for actually buy. The bundles and the engagement model · Procurement Center.
How to take this further
Send us the programme, the countries and the teams, and we will map it onto what exists — including telling you where the platform would leave work to your team. If you have a due-diligence questionnaire or a security assessment template, send that too: we answer from the registers, so the answers match the pages.