VoiceInsights Africa
Company & Capability Profile
VoiceInsights Africa
Voice-first Research, Data Collection, M&E and Decision Intelligence Infrastructure
“Every Voice. Every Language. Every Insight.”
Who we are
VoiceInsights turns community and field evidence into traceable, decision-ready institutional intelligence.
VoiceInsights Africa
The institutional brand, and the party an organisation contracts with. It carries the research and delivery capability: instrument design, field operation, quality control, analysis and governed publication.
VoiceInsights Workspace
The platform that capability runs on. It holds the instruments, the collection routes, the evidence chain, the review workflow and the publication engine, and it is the system an institution licenses when it wants to operate the capability itself.
The distinction matters in procurement. A buyer commissioning a study contracts for evidence; a buyer licensing the Workspace contracts for the system. Both are supported, and the engagement models on the next pages set out what each asks of your team.
The institutional problem
Organisations rarely lack data. They lack evidence they can defend when somebody asks where a number came from.
Fragmented systems
Instruments in one tool, field data in another, analysis in a spreadsheet and the report in a document nobody can trace back to a response.
Evidence that will not hold
A finding survives a meeting but not a donor’s follow-up question, because the chain from statement to source was never kept.
Multilingual complexity
The language a respondent speaks, the language a team works in and the language a report ships in are three different decisions, routinely collapsed into one.
Connectivity
Field work happens where the network does not. Collection has to continue offline and reconcile later without losing the record of who collected what.
Too many routes in
Phone, messaging, web and paper-replacement apps each arrive in a different shape, and the differences surface at analysis.
Slow reporting
The gap between the last interview and a decision-ready publication is measured in weeks, and the decision has usually been taken already.
What follows is what VoiceInsights does about each of these, stated at the level the registry supports and no higher.
How we work
Three engagement models. The evidence chain is the same in all three; what changes is who operates it.
Managed research delivery
We run the study
For an organisation that needs the evidence, not the software. Instrument design, translation of the instrument into the collection languages, enumerator arrangement, collection, quality review, analysis and the governed publication.
Technology-enabled delivery
Your team runs it, with us alongside
For an organisation with its own M&E or research team that wants the instrument, the field workflow and the evidence chain to be somebody else’s engineering problem.
Enterprise platform
The platform as institutional capability
For an institution standardising how evidence is collected, governed and published across programmes, countries and teams.
The delivery lifecycle
- 1Design
- 2Field preparation
- 3Collection
- 4Getting the work back
- 5Evidence management
- 6Quality control
- 7Analysis
- 8Reporting and decision support
Research, surveys and evaluation
36 commissionable study types across 8 categories, each carrying the status the registry gives it.
| Study type | Status |
|---|---|
| Surveys | |
| Household survey | Supported, bounded |
| Population survey | Supported, bounded |
| Kap survey | Proven |
| Farmer survey | Supported, bounded |
| Assessments | |
| Facility assessment | Proven |
| Health facility assessment | Supported, bounded |
| Needs assessment | Supported, bounded |
| Rapid needs assessment | Supported, bounded |
| Wash assessment | Supported, bounded |
| Protection assessment | Supported, bounded |
| Gender inclusion assessment | Supported, bounded |
| School assessment | Supported, bounded |
Causal impact evaluation. A causal impact claim requires an identification strategy and a counterfactual. The analytical engine implements neither and refuses the request by na
Census enumeration. A census enumerates every member of every unit, which is a repeat group — and repeat groups are deliberately not supported, because flattening one int
Conjoint pricing research. Conjoint, MaxDiff and constant-sum allocation are question types the collection surfaces do not implement, and the estimation behind them is not in th
Qualitative research
Qualitative is a registered research mode, and the interview formats sit inside it rather than beside it.
| Study type | Status |
|---|---|
| Qualitative | |
| Kii study | Proven |
| Idi study | Proven |
| Fgd study | Supported, bounded |
| Stakeholder consultation | Supported, bounded |
| Community dialogue | Supported, bounded |
| Community feedback | |
| Citizen feedback survey | Proven |
| Community scorecard | Supported, bounded |
Key informant interviews, in-depth interviews and focus group discussions are formats within qualitative research, not separate designs. There is no separate KII module, route or table, and the qualitative model is instrument-agnostic: transcripts code against the same codebook whoever was in the room. Presenting a format as a module would claim a capability the system does not have.
What is produced
Audio evidence retained with its consent record, transcripts bound to the response they came from, thematic coding against a project codebook, and narrative findings that resolve to the material they came from.
What is not
Speaker identification in a group recording is not performed. Focus group audio is preserved and attributable to the session, not to the individual speaker.
Monitoring, evaluation, research and learning
Programme evidence that holds together between rounds, and a record of how each figure was produced.
| Study type | Status |
|---|---|
| Monitoring evaluation | |
| Baseline survey | Proven |
| Midline study | Supported, bounded |
| Endline survey | Supported, bounded |
| Post distribution monitoring | Supported, bounded |
| Programme monitoring | Supported, bounded |
| Evaluation study | Supported, bounded |
| Programme review | Supported, bounded |
| Government | |
| Service delivery survey | Supported, bounded |
| Public consultation | Supported, bounded |
Baseline and endline are programme phases, not designs. A baseline is a structured survey run at the start of an intervention and publishes as a baseline publication; an endline repeats the baseline instrument so the two can be compared. Change is reported descriptively — a value, a later value, and the difference between them — because the platform implements no identification strategy and makes no attribution claim.
Field operations and collection routes
6 registered routes into one governed pipeline. What arrives is shaped the same way whichever route it came through.
| Route | Status |
|---|---|
| Phone Call | Proven |
| Proven | |
| SMS / Feature-Phone Fallback | Proven |
| Web Platform Link | Proven |
| Offline Mobile Application | Proven |
| Proven |
Field operations
Enumerator assignments bound to a project, a separate field sign-in using a project code, enumerator ID and PIN, offline collection with later synchronisation, consent recorded with the response, server-side validation and supervisory review.
Operational requirements differ by route
The web link and the offline application work wherever the platform is deployed. Phone, WhatsApp and SMS require a communications provider to be configured for your organisation — one account covers all three.
Published limitations, carried on every study type
- No roster. One record per unit on every collection surface. Member-level rosters — a household enumerated person by person — are held relationally by the platform, and there is no roster control on the field app or the web link, so a roster instrument is configured with your implementation team rather than built and fielded unaided.
- No causal claim. Change is reported descriptively — a value, a later value, and the difference between them. No causal or attribution claim is produced, because the platform implements no identification strategy and refuses to imply one.
- No media capture. Photographs, GPS points and file attachments are not question types a respondent or enumerator answers. Location is recorded about a submission, not asked as a question.
- Bounded logic. Skip logic is one condition on one earlier question — “if they said no, do not ask the follow-ups” — evaluated by the server as well as the device. Multi-condition expressions, arithmetic across questions and cross-repeat references are refused when the instrument is imported rather than accepted and ignored.
Voice and African-language research
“Do you support my language?” is six questions, not one, and we answer them separately.
| Axis | What it asks |
|---|---|
| Text | Can a respondent answer in writing in this language? |
| Voice | Can a respondent answer by speaking in this language? |
| Transcription | Can that speech be turned into text? |
| Translation | Can it be translated for analysis? |
| Analysis | Can free text in this language be analysed? |
| Output | Can a deliverable be produced in this language? |
The language a respondent speaks, the language your team works in and the language a report ships in are three separate decisions. The live capability state of every registered language, on each of the six axes, is published on the website and read live from the capability registry rather than restated here, because a number in a PDF ages and a registry does not.
Data management, quality and evidence governance
A publication that cannot resolve its evidence is refused rather than produced.
Validation
Instrument rules evaluated on the device and again on the server, so a submission cannot bypass them by arriving from an unusual route.
Consent records
Consent is recorded with the response, in the collection context and language it was given in.
Quality review
Review and approval steps sit between collection and analysis, with the reviewer recorded.
Evidence traceability
A published statement resolves to the responses it came from; where it cannot, the publication is refused.
Access control
Role-based authorisation with tenant isolation. An organisation’s evidence is reachable only inside that organisation.
Retention and export
Retention follows law, consent and contract. Governed dataset export in open formats.
AI, analytics and decision intelligence
AI assists governed analysis. People retain authority, and the evidence remains the thing that decides.
- 1Authorized evidence
- 2AI contribution
- 3Candidate output
- 4Human review
- 5Accepted, modified or rejected
- 6Governed intelligence
- 7Publication approval
Where AI assists
Transcription, translation, classification, summarisation, thematic coding and drafting. Every contribution is candidate intelligence until a named reviewer accepts, modifies or rejects it.
Where it does not
AI does not approve evidence, make institutional decisions or authorise publication. It must never fabricate observations, consent, authority, approvals, audit history or provenance.
Reporting and publication
The output is a publication, not a data dump — and it is produced in the formats an institution actually circulates.
Publication types
Executive summary, full report, evaluation report, monitoring report, policy brief, donor brief, management report, statistical annex, evidence annex and board-level output, according to what the evidence base supports.
Governed publication
Evidence lineage, human review, versioning, approval, correction and withdrawal apply to released intelligence. A publication carries the conditions under which it was produced.
Enterprise, government and integration
Stated at its implementation state, because a format implemented is not a system integrated.
| Integration | State |
|---|---|
| Channels | Live |
| Offline app | Live |
| Email delivery | Live |
| Rest api | Configured per deployment |
| Exports | Configured per deployment |
| Sso | Configured per deployment |
| Scim | Configured per deployment |
| Bi tools | Configured per deployment |
| Instrument import | Format implemented, never connected |
| Dhis2 | Format implemented, never connected |
Multi-country programmes
One programme across several countries stays one institutional record, with geographic scope enforced in authorisation rather than by convention.
Workspaces and roles
Organisation and project workspaces, role-based access, and a multi-tenant architecture in which tenant isolation is an implemented boundary.
Security, privacy, governance and safeguarding
Nothing on this page is a certification, and the page says so before a reviewer has to ask.
Applicable law
Implementations are designed to support obligations under Tanzania’s Personal Data Protection Act, No. 11 of 2022, alongside the applicable agreement. Where a programme falls under the EU GDPR or another regime, those obligations are assessed and documented during implementation. No independent assessment of GDPR alignment has been carried out, and none is claimed.
Implemented controls
Authentication, scoped authorisation, encryption in transit and at rest, tenant isolation, auditability, retention and deletion controls, subprocessor and processing-location registers published and read live.
Safeguarding, research ethics, responsible AI and publication governance each have a published position, and the procurement pack lists 15 published documents of 25, with the remainder shown at their real state rather than omitted.
Who we serve
NGOs and INGOs
Programme evidence that survives a donor’s follow-up question.
Governments and public sector
Citizen evidence a ministry can act on, with the account of where it came from.
Development partners and donors
Impact data from grantees rather than their self-reports.
Research institutions
Evidence a client can defend, with the trail attached.
Private sector
Customer, employee and market evidence where the registry supports it.
Multi-country programmes
One programme, several countries, still one institutional record.
Private sector and market research
| Study type | Status |
|---|---|
| Market enterprise | |
| Customer satisfaction | Proven |
| Employee engagement | Supported, bounded |
| Market research | Supported, bounded |
| Financial inclusion survey | Supported, bounded |
Sectors
HealthEducationAgricultureHumanitarianLivelihoodsGovernanceCitizen engagementYouth and employmentMarket research
Procurement and engagement
Routes in, and what each one needs from you.
| Route | What it suits |
|---|---|
| RFI / EOI | Early market engagement. We answer at capability level and name what requires an independent party. |
| RFP / Tender | A specified study or platform requirement. The procurement pack supplies the standing documents. |
| Framework agreement | Repeat commissioning across programmes or countries under agreed terms. |
| Pilot | A bounded first study, scoped so the result is usable whether or not it continues. |
| Annual platform agreement | The Workspace as institutional capability, with implementation and support. |
| Managed research engagement | A study delivered end to end, with the evidence and the publications as the deliverable. |
Contact us
“Every Voice. Every Language. Every Insight.”
Tanzania — Head Office
Ali Hassan Mwinyi Road, Kisutu3rd Floor, Haidary Plaza
P.O. Box 77139
Dar es Salaam, Tanzania