DISCLAIMER — FICTIONAL RESEARCH DOCUMENT. This report is a speculative adaptation study produced as an example of what a structured market adaptation audit looks like when applied to AI literacy tool deployment in a West African context. It was not commissioned by, requested by, or produced in collaboration with Anthropic or Claude. All case study profiles are fictional. All deployment recommendations are illustrative. This document exists to demonstrate what rigorous adaptation analysis looks like — not to represent any actual product roadmap or partnership.
Mandatory first step. This audit will not proceed without it. Botspeak + Claude is not a single-market deployment. Before any analysis, declare which Côte d'Ivoire this product serves.
| Population Segment | Description | Botspeak Viability |
|---|---|---|
| (a) Abidjan formal sector | Plateau / Cocody / Zone 4; university-educated; French text fully viable; Orange Bank Africa integration possible | Text-first Botspeak viable; standard French Claude output serves this population |
| (b) Abidjan informal sector | Yopougon / Adjamé / Abobo / Koumassi; ~3.5–4.5M residents; Nouchi vernacular; standard French NLP fails as primary interface | Voice-first mandatory; Nouchi ASR required; text output requires oral read-back; literacy acquisition dimension applies |
| (c) Interior / cocoa belt / north | Estimated 11–13M; Dioula trade corridor; Baoulé cocoa heartland; Burkinabè migrant workers; Mooré-primary communities | Voice-first mandatory; offline-first required; Mooré deployment before other indigenous languages; CCC compliance required for any cocoa-adjacent product |
This audit is scoped to serve populations (b) and (c) — the populations a standard French-text Botspeak deployment would silently abandon. Population (a) is acknowledged as viable with minimal adaptation and is not the analytical focus. Any deployment that claims to serve (b) and (c) while designing for (a) is making a political choice about who counts. Products that claim to serve all three with a single interface have silently chosen to serve (a) adequately and (b) and (c) inadequately.
The single most important condition shaping Botspeak + Claude deployment in Côte d'Ivoire:
Côte d'Ivoire is not a language problem with a French solution. It is a multilingual reality in which French is the language of elite administration and standard NLP, Nouchi is the actual vernacular of Abidjan's majority population, Dioula is the language of cross-ethnic commerce, and Mooré is the language of the workers doing most of the manual labor in the cocoa belt. Botspeak's supervisory model — teach the human to direct Claude, evaluate its output, and own the result — is structurally sound for all three populations. The failure mode is not conceptual. It is infrastructural: if Claude cannot hear the user's language accurately, the entire supervisory chain collapses at step one. The Nouchi ASR gap and the Mooré deployment opportunity are not equivalent problems. Mooré has a production-grade solution (4.24% WER). Nouchi does not. This asymmetry drives the entire deployment architecture.
The primary risk if that condition is ignored:
A Botspeak deployment that launches on standard French NLP in Abidjan will reach the 15–22% of the metro population in formally educated, French-literate employment and will be described in internal metrics as a success. The 3.5–4.5M residents of Yopougon, Adjamé, Abobo, and Koumassi — the intended beneficiaries of an AI literacy tool that reduces dependence on scribal intermediaries — will not be reached. They will not complain. They will simply not show up in the user data. The audit machinery that could detect this failure — language-disaggregated uptake metrics, commune-level activation rates, Nouchi comprehension testing — will not be built unless it is required at the design stage. This report requires it.
| Dimension | Ivorian Condition | Gap Assessment | Priority |
|---|---|---|---|
| 1. Linguistic Architecture | Nouchi (6–8M Abidjan); Dioula (7–10M); Mooré (1–2M cocoa belt); Baoulé (3.5–4M); French (formal/elite only) | Mooré: production-grade ASR (4.24% WER). Nouchi: zero production NLP. Dioula: Bambara transfer viable, unvalidated. Baoulé: no corpus. | Critical |
| 2. Interface & Interaction | 78–85% literacy in Plateau/Cocody; 25–32% in northern departments; cocoa belt migrant standard requires Mooré/offline-first | Text-first reaches ~15–22% of target market. Voice-first with Nouchi + Mooré reaches majority. | Critical |
| 3. Infrastructure | 85% 4G coverage nationally; north and far west limited; 50–55% mobile internet (highest in WAEMU); Orange Bank Africa unique | Abidjan: sophisticated connected Botspeak viable. Interior/cocoa belt: offline-first mandatory; session interruption handling required. | High |
| 4. Financial Integration | Four-player mobile money market; CCC price floor; PI-SPI June 2026 deadline; Orange Bank Africa banking-grade integration | Botspeak without payment features: lower complexity. Cocoa records: CCC compliance mandatory. PI-SPI enables cross-institutional settlement. | High if cocoa-adjacent |
| 5. Regulatory & Data Sovereignty | APDP-CI most developed DPA in WAEMU; enforcement-capable; ARTCI for USSD/VAS; CCC for cocoa; Ivoirité undercurrent | APDP-CI notification required before launch. Cross-border data pipeline audit required for Burkinabè worker features. | High |
| 6. Cultural & Social Architecture | Four-track social license: Dioula merchant networks; Akan chieftaincy; Christian church; Islamic community; Burkinabè solidarity | No single endorsement channel reaches all populations. Post-conflict north-south fracture creates structural adoption resistance. | Critical |
Mandatory launch gate — no items may be left unresolved.
| Check Item | Requirement | Status |
|---|---|---|
| Visual identity | Spokespeople pan-ethnic, not southern-default | Unverifiable Requires explicit design brief specifying Akan, Dioula, Sénoufo, and Baoulé representation; default photography will skew southern without active intervention |
| Voice persona | Accent does not code to one region | Unverifiable Claude TTS default will not produce Nouchi or Dioula register; a French metropolitan or Plateau-coded voice signals elitism |
| Distribution strategy | Launch geography balanced | Inferred All natural launch partners are Abidjan-HQ'd; cooperative networks in Korhogo, Daloa, Man must be activated before or concurrent with Abidjan launch |
| Gatekeeper engagement | Both Islamic north and Christian south represented | Unverifiable Requires documented community review panels from both geographies before launch — a design requirement, not a post-launch task |
Failing any item is a launch risk. Ivorian users are highly attuned to ethnic coding signals. A product that unintentionally codes as "southern" will face adoption resistance in the north that no amount of subsequent marketing can reverse.
Observed Nouchi is the primary vernacular of Yopougon, Adjamé, Abobo, and Koumassi — the four communes accounting for the majority of Abidjan's approximately 6–7M metropolitan population. It is not degraded French. It is a distinct creole with its own grammar, phonology, and lexicon, drawing on French, Dioula, Baoulé, and English registers in ways that standard French NLP systematically misprocesses.
Observed There is no production-grade Nouchi ASR. No published WER. No corpus of scale. Standard French models will hallucinate, flatten code-switches, and misidentify grammatical structures that carry semantic weight in Nouchi but are absent from standard French training data.
Inferred A Botspeak deployment in Yopougon relying on standard French ASR will produce a failure mode in which the user's prompt is partially transcribed, Claude generates a response to a mangled version of intent, and the user — who cannot read the output independently — receives an audio read-back of a document that does not match what they asked for. This failure is invisible in standard UX telemetry. It will be attributed to "low adoption" rather than "NLP mismatch."
Required action: Nouchi corpus collection is not optional for a population (b) deployment. Minimum viable corpus: 5,000–10,000 annotated utterances of spontaneous Nouchi speech in task-relevant contexts. Target deployment context: Adjamé market and Yopougon cooperative networks. Radio call-in archives (RTI, NCI) are the highest-yield existing audio source. Partnership with Université Félix Houphouët-Boigny linguistics department is the recommended institutional anchor.
Observed Dioula is spoken by approximately 7–10M people across Côte d'Ivoire. It is the language that crosses ethnic lines in commerce — in Adjamé market, in the trade corridor from Abidjan to Bouaké to Korhogo to Bamako. A Botspeak product that earns Dioula merchant network endorsement in Adjamé acquires a distribution channel that reaches from the lagoon to the Sahel.
Observed Bambara (Bamanankan) is mutually intelligible with Dioula at the functional trade register level. The Bambara ASR literature (46.76% WER for untrained global models; significantly improved with fine-tuning on Mali-specific corpora) provides the closest documented transfer learning baseline.
Inferred Bambara fine-tuned ASR will perform meaningfully better on Dioula than generic French or global multilingual models. The transfer is not guaranteed and requires empirical validation — Ivorian Dioula has phonological and lexical divergences from Malian Bambara. This validation is a Phase 1 deliverable.
Required action: Dioula ASR validation study against Bambara-trained models with native Ivorian Dioula speakers. Minimum test set: 500 utterances covering Botspeak task types. Adjamé market wholesale section is the recommended recording site.
The Burkina Faso fine-tuned Mooré ASR achieves 4.24% WER — production-grade performance. Developed on 88,000 utterances / 150 hours of audio, with companion TTS at 4.36/5.0 MOS. This is the highest-performance indigenous NLP in the AKWABA framework family.
Observed Burkinabè migrant workers constitute an estimated 25–30% of the cocoa belt labor force. In many cooperatives in the Daloa, Gagnoa, and San-Pédro supply shed, Burkinabè workers represent the majority of manual harvesters. "Cocoa farmer" without nationality specification systematically excludes the people doing most of the work.
Observed No cocoa belt product has deployed Mooré voice features. The combination of production-grade NLP and a large underserved user population in a defined geography constitutes a structural first-mover advantage.
Required action: Mooré voice deployment is the highest-leverage indigenous language investment for any population (c) Botspeak deployment. Deployment sequence: integrate Burkina Faso Mooré ASR → test with Burkinabè cooperative members in Daloa region → document WER on cocoa-specific vocabulary → iterate before broader launch.
Observed Baoulé is spoken by approximately 3.5–4M people, concentrated in the cocoa heartland of Yamoussoukro, Bouaké, and the central plateau. Cooperative presidents and Akan chieftaincy structures in this region are Baoulé-speaking. The word Akwaba is Baoulé/Akan. A product that cannot address cooperative leadership in Baoulé cannot earn the endorsement of the people who control cooperative membership and payment distribution.
Observed There is no Baoulé corpus of scale. No published ASR WER. No TTS in production.
Inferred Baoulé NLP development is a 12–18 month investment minimum. It cannot be the first indigenous language layer — it is a Phase 2 or Phase 3 investment. Until ASR is viable, the product must reach Baoulé-speaking cooperative presidents through a different channel (literate intermediary in Baoulé, or French for formal-sector cooperative leadership).
Observed French is viable as a text interface for approximately 15–22% of the Abidjan metro population with functional French literacy. It is not viable as the primary interface for the majority of intended Botspeak users in populations (b) and (c).
Tier 1 — Abidjan formal sector (population a). Text prompt input viable; standard French Claude output readable; Botspeak supervision teachable through text-based curriculum; Orange Bank Africa integration possible. Not the focus of this audit.
Tier 2 — Abidjan informal sector (population b). Voice prompt input required; Nouchi ASR is the mandatory interface. Claude output must be delivered as audio read-back; text display secondary. Supervision skills taught through oral evaluation — "does this sound right for what you asked?" not "read this and mark corrections." Shared device model viable and expected. Observed Literacy in Yopougon, Adjamé, Abobo, and Koumassi is estimated at 35–50% functional French literacy; audio-primary design is the access condition for at least half of residents.
Tier 3 — Interior / cocoa belt / north (population c). Voice-only input required; offline-first mandatory; session interruption design required. Device ownership rate lower; group device and mediated access models are primary. Mooré + Dioula + French audio output required; Baoulé for cooperative leadership via intermediary. Cocoa calendar alignment: harvest season (October–January) is the primary deployment window.
Inferred Oral TCC structure can be taught through group demonstration with 2–3 practice cycles. The pedagogical entry point is familiar: it mirrors the structure of any intermediary negotiation in Ivorian market culture. The cognitive shift is naming the three elements explicitly so the user can evaluate Claude's output against their original intent.
Observed The literate intermediary is already present in target deployment contexts. In cocoa cooperatives, the cooperative secretary is typically the only literate member. In Adjamé market, the association secretary plays this role. In Yopougon tontines, one member is typically designated for record-keeping.
Inferred Botspeak's highest-leverage entry point is not the individual user but the literate intermediary. The critical design requirement: the tool must transfer supervision agency from the intermediary to the group member, not reinforce intermediary dependency. A cooperative member who can say "that's not what I told you to write" is exercising Botspeak supervision even if they cannot produce the text independently.
Observed Côte d'Ivoire has 85% 4G coverage nationally and 50–55% mobile internet penetration — the highest in the WAEMU family. Abidjan metro supports sophisticated connected features. The north and far west (Man, Odienné, Bondoukou) have materially lower connectivity.
Observed The cocoa belt has variable connectivity. Major processing centers (San-Pédro, Daloa) have adequate 4G. Remote cooperatives and field workers may operate in 2G or offline conditions during harvest.
Inferred A two-mode architecture is required: connected mode (Abidjan metro, major cocoa towns) with real-time Claude API and audio streaming; offline-first mode (remote cooperatives, northern departments) with cached prompt templates, batch processing, and local document storage with sync.
Observed Transsion brands (Tecno, Infinix, Itel) dominate the mass-market device tier. Mass-market devices have 2–4GB RAM. Audio processing must be server-side with low-bandwidth compression (Opus codec preferred over WAV).
Observed Orange Bank Africa's integration of banking-grade savings, credit, and insurance APIs with mobile money infrastructure is unique in the WAEMU family. Claude-generated financial documentation can potentially be linked directly to Orange Bank Africa account data. This integration pathway is available in no other WAEMU market.
Session interruption design: In cocoa belt field conditions, a voice prompt session may be interrupted by network dropout mid-response. A cooperative secretary who loses connection after dictating 6 of 8 harvest records must not lose all six. Checkpoint saving at sentence level is the minimum requirement. This is a known failure mode of cascaded pipeline architectures in low-connectivity environments.
Scoping note: Botspeak is a documentation supervision framework, not inherently a payment product. Financial integration relevance depends on which use cases are activated.
| Use Case | Financial Integration Requirement |
|---|---|
| Cooperative harvest record documentation | CCC price floor compliance if records are used for payment calculation |
| Tontine / savings collective records | No direct payment integration required; records are internal |
| Loan application documents | BCEAO / SFD regulation if submitted to microfinance institution |
| Cocoa payment receipts (farmer to aggregator) | CCC compliance mandatory; PI-SPI settlement architecture if digital |
| Insurance claim documentation | CIMA licensing required; 6–12 month timeline |
Observed Côte d'Ivoire has four active mobile money operators: Orange Money (primary; widest agent network), Wave (aggressive urban entrant; low-fee model), MTN MoMo (third player; urban share), Moov Africa (fourth; northern pockets). Any Botspeak payment feature integrating a single operator provides inferior coverage.
PI-SPI integration with the BCEAO June 30, 2026 deadline enables cross-institution settlement across Orange Money, Wave, MTN, and Moov; early integration provides settlement architecture that late entrants inherit as table stakes.
Observed CCC (Conseil Café-Cacao) regulates the cocoa supply chain including minimum farmgate price floors. Any Botspeak documentation recording cocoa transactions operates within CCC's regulatory authority. Documentation that facilitates below-floor payments, even inadvertently, exposes the deploying entity to enforcement risk.
Observed Côte d'Ivoire's data protection authority is the most developed in the WAEMU family and is enforcement-capable. Any Botspeak deployment processing user voice data, storing AI-generated documents, or maintaining user interaction logs requires APDP-CI notification before launch.
Inferred Voice data of Burkinabè and Malian migrant workers collected in Côte d'Ivoire is subject to Ivorian data sovereignty. Cross-border data pipelines — for example, processing Mooré voice data through servers outside Côte d'Ivoire — require explicit cross-border transfer authorization. The Burkina Faso Mooré ASR model was likely developed and hosted outside Côte d'Ivoire; the deployment architecture must document the data flow and obtain appropriate authorization.
Ivoirité undercurrent: The political history of Ivoirité (Ivorian nationality as eligibility criterion) means any product collecting nationality data is making a political positioning choice, not a neutral technical one. A Botspeak product requiring national ID for registration will systematically exclude Burkinabè and Malian workers — the population most in need of documentation assistance — and will be correctly read as an exclusionary design choice. Tiered KYC (not national-ID-only) is the required design standard.
Observed ARTCI regulates USSD and VAS services. Any Botspeak voice channel deployed over USSD rather than mobile internet requires ARTCI licensing. USSD is the relevant architecture for reaching users without smartphones — which includes a significant share of the cocoa belt target population.
Observed CCC compliance is mandatory for any product touching the cocoa supply chain. This includes documentation of harvest weights, payments, cooperative membership, and price records. CCC has enforcement capability and has used it.
Inferred If Botspeak use cases include preparing insurance documentation, CIMA licensing is required. The 6–12 month licensing timeline means this process must be initiated in Phase 1 if insurance features are planned for Phase 2 or Phase 3.
Botspeak cannot be endorsed by a single institution across all three target populations. It requires four parallel endorsement tracks.
Track A — Dioula merchant networks (Adjamé / trade corridor). Entry point: Adjamé wholesale market association leadership. Endorsement signal: product endorsed as a "record-keeping tool for honest trade" — not as an AI product. Propagation: merchant networks from Abidjan to Korhogo to Bamako; endorsement travels with goods. Inferred Dioula merchant endorsement in Adjamé is the highest-value single endorsement act for reaching both urban informal sector and interior trade corridor.
Track B — Akan chieftaincy + cooperative presidents (cocoa belt). Entry point: cooperative presidents in Daloa / Gagnoa / Yamoussoukro region. Endorsement signal: product introduced as a tool giving farmers "their own record" — protection against payment disputes with aggregators. Baoulé-language endorsement from cooperative presidents is the requirement; intermediary-mediated until Baoulé ASR is production-ready.
Track C — Christian church networks (Abidjan and south). Entry point: Protestant and Catholic congregation literacy programs. Churches already run adult education programs. Inferred Church networks provide the highest-reach distribution channel for populations (b) in communities that distrust commercial or government-affiliated launches.
Track D — Islamic community structures (north and Dioula Muslim communities). Entry point: Islamic community leadership in Korhogo, Odienné, and Abidjan's northern communes. Endorsement signal: tool for documentation consistent with Islamic principles of honest dealing. Observed The north-south fracture means a product entering exclusively through Track C will face adoption resistance in Track D territory. Both tracks must be activated before launch, or the product will be ethnically coded from day one.
Observed Burkinabè migrant workers in the cocoa belt are organized through worker solidarity associations and ethnic community networks. These associations function as trust infrastructure, dispute resolution, and remittance coordination networks.
Inferred Botspeak introduction to Burkinabè cooperative workers requires endorsement from solidarity association leadership, not cooperative management (which is typically Ivorian). The use case that earns trust: cross-border remittance documentation to Burkina Faso families — a primary financial need that the Ivorian formal banking system serves poorly.
Observed Côte d'Ivoire is Francophone West Africa's cultural powerhouse. Coupé-Décalé music and its associated visual and linguistic register define urban Abidjan youth culture. A Botspeak product targeting youth in populations (b) that adopts an institutional or European register will be read as culturally foreign.
Inferred Product voice persona, onboarding flow, and example prompts for urban Abidjan youth should be developed with Ivorian cultural consultants fluent in Coupé-Décalé register. Nouchi register signals cultural authenticity. Plateau French signals exclusion.
Observed The mass-market Ivorian device ecosystem supports voice input through standard mobile OS voice APIs. WhatsApp voice messaging is the dominant voice communication mode in Yopougon and Adjamé — more familiar than any purpose-built application.
Inferred A WhatsApp-adjacent voice interface — record a voice message, receive a voice message back — is the lowest-friction entry point for Botspeak voice interaction. Users already understand the interaction model. The technical layer (ASR → Claude → TTS) is invisible. The pedagogical work is teaching users what to say in the voice message to get a useful output — which is the Botspeak supervisory skill.
Inferred Code-switching — Nouchi to French to Dioula within a single utterance, which is normal in Adjamé market conversation — is the specific failure mode that standard single-language ASR cannot handle. The system will assign the entire utterance to one language model and misprocess every segment that belongs to the other.
Example well-formed oral Botspeak prompt (Yopougon tontine secretary context):
"Claude, fais moi un papier. Écris que la réunion de notre tontine c'est samedi prochain, chez Mama Adjoua à Yopougon Selmer. On commence à 14h. Dis-leur d'amener leur cotisation — c'est 5000 FCFA ce mois-ci. Fais ça court, facile à lire à haute voix."
This prompt contains all three TCC elements: Task (make an announcement), Context (tontine meeting, time, place, host), Constraint (short, audible register). The constraint facile à lire à haute voix is the Botspeak insight: the user already knows the output will be consumed by voice, not by reading, and specifies it. This is supervision in action.
Observed Côte d'Ivoire's adult literacy programs include DAENF and NGO-led programs through Catholic and Protestant literacy missions. These operate in French and, in some northern regions, Dioula. Mooré-medium adult literacy is largely absent from formal Ivorian programming.
Inferred The "reading through listening" pathway — hearing Claude's audio read-back while following along with text on screen — is viable for users at the emergent literacy stage. For functionally non-literate users, the audio-only loop (dictate → hear output → evaluate → revise) provides the Botspeak supervisory skill without requiring reading.
Inferred The highest-leverage literacy acquisition use case is the revision cycle. When a user hears Claude's output and says "that's not what I meant" — and then re-dictates a revised prompt — they are exercising the critical evaluation skill. Over repeated iterations, they develop an implicit grammar of what makes a prompt produce a useful result. This is meta-linguistic awareness building through practice, not formal instruction.
Observed In Côte d'Ivoire, oral tradition is strong across all major ethnic groups — Akan oral history, Dioula praise singing (djeliya), Bété oral narrative. A user who dictates a narrative and receives a written letter is performing a translation that their culture already values — the griot function, digitized.
| Language | ASR Status | Interim Architecture | Minimum Viable Interface |
|---|---|---|---|
| Nouchi | Zero production NLP; no corpus; no published WER | Standard French ASR with human review layer; community moderator flags misrecognitions | Human-in-the-loop: ASR → draft → literate intermediary review → Claude → audio output |
| Dioula | No dedicated production ASR; Bambara transfer viable but unvalidated | Bambara fine-tuned model as interim; empirical validation required | Bambara-trained model with validation testing on Ivorian Dioula corpus |
| Mooré | Production-grade: 4.24% WER; 88K utterances; 4.36/5.0 MOS TTS | Deploy now — no interim architecture required | Burkina Faso Mooré ASR integrated directly; cocoa vocabulary layer |
| Baoulé | No corpus; no published WER; no TTS | Literate intermediary in Baoulé reading Claude French output aloud | Intermediary-mediated; Phase 2–3 investment for dedicated ASR |
| French | Production-grade; Claude natively strong | Direct deployment for population (a) | Standard text/voice interface |
Observed Tontines and likelemba in Yopougon and Adjamé are the primary financial community structure for informal sector women. A typical tontine has 10–20 members, meets monthly, and has one designated secretary. A shared group device used in the tontine meeting context — where the secretary demonstrates Botspeak use and members collectively evaluate the output — is the most plausible entry point for population (b) group adoption.
Inferred In the absence of individual critical reading capacity, group listening — playing Claude's audio output to the assembled group — is the primary quality control mechanism. In a tontine context: the secretary reads back Claude's record of the month's contributions; members correct errors by voice; the secretary re-prompts Claude with corrections. This is a functional Botspeak revision cycle conducted collectively.
Observed The post-conflict trust architecture is not historical background — it is a present-tense social fact. The north-south fracture (Gbagbo v. Ouattara political geography; northern Muslim / southern Christian structural tension) remains active. A product introduced through a northern cooperative will be viewed with suspicion by southern associations and vice versa if the endorsement is regionally coded.
Inferred The scaffold/substitute distinction maps onto a specific failure mode in the cooperative documentation context: if the cooperative secretary uses Claude to generate all records without member verification, Claude becomes a substitute for collective memory, not a scaffold for individual capacity. Payment disputes that would previously have been resolved by member recollection become disputes about what Claude was told to write.
Inferred The productive struggle threshold: users develop supervision capacity when asked to evaluate Claude's output against their own intent — not when asked to produce text independently. The stretch task is not "write this yourself" but "tell me three things Claude got right and one thing Claude got wrong." This evaluation task is achievable by functionally non-literate users through audio output.
Observed Griotic tradition in Dioula and Akan communities encodes social memory through formal oral composition with precise structural requirements (opening, acknowledgment, narrative, closing). Inferred This structural competence is directly transferable to Botspeak prompt design: a user who knows how to structure a formal oral address can apply the same structure to a voice prompt. The framework is not teaching a foreign cognitive skill — it is naming a skill the user already has.
Assessment approaches without writing requirement:
Amenan Kouamé, 34, female, cooperative outside Daloa, Haut-Sassandra region. Primary languages: Baoulé, functional Dioula, basic oral French; non-literate in any script. Secretary of a 47-member cocoa cooperative — elected because she attended school through CM2 and can read simple French slowly, the most literate member available. Each harvest season, her records of each member's harvest weight and payment due are the only documentation between the farmer and the aggregator's scale. Disputes happen every season. Farmers who cannot read cannot verify the records.
Access: Group-owned Tecno T40 Pro tablet (4GB RAM) purchased through cooperative fund contribution. Connectivity intermittent — 4G from Daloa town reaches the cooperative office; field positions use 2G or none.
Voice prompt in Dioula:
"Claude, sɛbɛn nin koo: Koné Moussa, suro 12. Traoré Adama, suro 9. Sawadogo Issouf, suro 15. Sawadogo i baara i ka cogo ma."
(Claude, write this: Koné Moussa, 12 bags. Traoré Adama, 9 bags. Sawadogo Issouf, 15 bags. Write Sawadogo well, he worked the most.)
The prompt directs in Dioula but the names signal the cooperative's mixed Ivorian/Burkinabè composition — Koné (Dioula), Traoré (Mandé), Sawadogo (Mossi/Burkinabè). Amenan's final instruction — "write Sawadogo well, he worked the most" — is a supervisory constraint about register and respect that Claude must be trained to interpret.
Observed The CCC minimum farmgate price for 2024/25 was set at 1,000 FCFA/kg. Any Claude-generated payment record must incorporate the current season's price floor. Miscalculation exposes the cooperative to CCC enforcement.
Literacy acquisition: Amenan can read simple French slowly. Each Claude-generated harvest record reinforces her recognition of French column headers (Nom, Sacs, Poids, Montant, Date) and numbers 1–100. Over one harvest season (15–20 record-generation cycles), she develops stable recognition. Her most active supervisory skill is numerical verification without reading: she checks Claude's numbers against her mental tally — "je sais que Moussa a amené 12, pas 11" — and flags discrepancies.
Endorsement obligation: Amenan's cooperative must not be the only beneficiary of accurate records. If the tool helps produce records more favorable to the aggregator than to the farmers, it will be seen as an instrument of extraction. The endorsement condition is that the records protect the farmer, not the buyer. This is a product positioning and data governance question: who does Claude's accuracy serve?
Reliability flags:
Gnagnon Bah, 28, female, Yopougon Selmer, Abidjan. Primary languages: Nouchi, Dioula, oral French; non-literate at functional document level. Sells attiéké and poisson braisé from a mobile setup near the Yopougon Selmer roundabout. Income 3,000–15,000 FCFA on a typical day. Elected treasurer of a 22-member tontine. Currently relies on a handwritten notebook and asks her younger sister (who attended lycée) to write the formal summary. She wants to do it herself.
Access: Personal Infinix Hot 40 (shared with her sister for data cost). WhatsApp is her primary communication interface; the Botspeak interface is built on WhatsApp Business API to minimize learning curve.
Voice prompt in Nouchi (WhatsApp voice message):
"Claude, fais moi le papier du mois. Le groupe c'est 22 personnes, cotisation 10.000 chacun. Ce mois Ama Koné a pas payé encore. Les autres ont payé. C'est le tour de Mariam Dosso de prendre. Écris ça proprement pour que je lise à la réunion."
Ça proprement is a register instruction. Gnagnon knows the difference between her Nouchi speech and the register appropriate for a formal group meeting. She is specifying a code-switch in her constraint — "take my Nouchi input and produce formal French output." This is a sophisticated supervisory instruction.
Literacy acquisition: Claude's tontine record output consistently uses the same vocabulary (Cotisation, Versement, Solde, Bénéficiaire). Over 12 months of monthly records, she develops stable recognition of these terms in context. More importantly, she reads the document aloud at the tontine meeting — a reading practice performed under social obligation (22 people listening), which motivates higher engagement than individual practice.
Productive struggle: Gnagnon's current prompt does not include all information Claude needs — she didn't specify the rotation order or the current month. Claude must ask clarifying questions. This dialogue teaches her what information a well-formed prompt must include.
Reciprocal obligation: The tool must not become a mechanism for extracting financial data from informal group savings. Tontine records are private group business. Any data storage architecture must make explicit that tontine records are not shared with financial institutions, insurers, or government entities without explicit group consent. In a context where financial surveillance of informal savings groups has historically been used to pressure them into formal banking, this assurance is not optional.
Reliability flags:
Côte d'Ivoire is the highest-opportunity deployment context in the WAEMU family for Botspeak + Claude — and the most likely to fail if its linguistic and social architecture is treated as an implementation detail rather than a design requirement. The opportunity: the highest-sophistication fintech ecosystem in WAEMU; the highest mobile internet penetration; the most developed data protection authority; and a production-grade Mooré ASR that can reach the most underserved labor population in the cocoa belt immediately. The failure mode: a product built for the Plateau/Cocody educated class that launches with a French-text interface and claims to serve Yopougon and the cocoa belt. That product will not fail noisily. It will quietly exclude its intended users and report success metrics drawn entirely from the population that needed it least.
| Priority | Dimension | Immediate Action |
|---|---|---|
| 1 | Linguistic Architecture | Nouchi corpus collection; Mooré ASR deployment; Dioula/Bambara transfer validation |
| 2 | Cultural and Social | Post-conflict balance check; four-track endorsement activation; Burkinabè solidarity engagement |
| 3 | Interface and Interaction | Three-tier architecture; WhatsApp-API voice interface; literate intermediary training |
| 4 | Regulatory and Data | APDP-CI notification; cross-border data pipeline audit; tiered KYC confirmed |
| 5 | Financial Integration | CCC compliance review; PI-SPI integration path against June 2026 deadline |
| 6 | Infrastructure | Offline-first architecture; session interruption handling; low-bandwidth audio compression |
Phase 1 (Months 1–4): Research and Architecture. Nouchi corpus collection initiated (RTI/NCI radio archives; Adjamé market recordings; Université FHB linguistics partnership). Dioula/Bambara transfer validation study (500-utterance test set). Mooré ASR integrated and tested on cocoa-vocabulary task set. APDP-CI pre-notification filing. Post-conflict north-south balance check completed. Four-track endorsement strategy developed. CIMA licensing initiated if insurance features planned.
Phase 2 (Months 5–9): Pilot Deployments. Cocoa belt pilot with Mooré voice Botspeak with Burkinabè cooperative members in Daloa region; literate intermediary training program. Yopougon tontine pilot with Nouchi-adjacent French voice and human review layer; WhatsApp API interface. Adjamé market pilot with Dioula voice for merchant record-keeping. CCC price floor integration. PI-SPI integration completed before June 30, 2026 deadline.
Phase 3 (Months 10–18): Scale and Indigenous Language Deepening. Nouchi fine-tuned ASR first version deployed if Phase 1 corpus threshold achieved. Baoulé intermediary program formalized; Baoulé ASR development initiated. Northern deployment (Korhogo, Odienné, Bouaké) with Islamic community endorsement. Orange Bank Africa integration. WAEMU regional expansion assessment: Burkina Faso (Mooré fully transferable), Mali (Bambara validated in Phase 1).
| Test Item | Status |
|---|---|
| Three Côte d'Ivoires Declaration made | ✓ Populations (a), (b), (c) declared; audit scoped to (b) and (c) |
| Post-conflict north-south balance check | ✓ All four items reviewed; Unverifiable flags placed |
| Nouchi gap assessed | ✓ Zero production NLP; corpus collection plan specified |
| Migrant worker population addressed | ✓ Burkinabè workers addressed in language stack, KYC, financial use case, Case Study 1 |
| Every dimension has documented finding | ✓ All six dimensions covered with evidence labels |
| APDP-CI notification | ✓ Required in Phase 1; cross-border audit specified |
| CCC compliance | ✓ Assessed; price floor integration specified as design requirement |
| CIMA licensing | ✓ Initiate Phase 1 if insurance features planned |
| PI-SPI integration confirmed against June 2026 deadline | ✓ Confirmed in Phase 2 |
| Four-player payment market architecture decided | ✓ PI-SPI as settlement architecture across all four operators |
| Mooré as highest-leverage indigenous NLP | ✓ Confirmed; deploy-now recommendation issued |
| Dioula/Bambara transfer validation scoped | ✓ Phase 1 validation study specified |
About this report. This is a fictional adaptation study produced as a demonstration of the AKWABA methodology applied to AI literacy tool deployment. It was produced by the AKWABA AI adaptation framework, not by Anthropic or Claude. The Botspeak framework is developed by Irreducibly Human. Claude is a product of Anthropic. Neither organization commissioned, reviewed, or endorsed this report. All case study profiles are fictional. All deployment recommendations are illustrative examples of adaptation analysis methodology.
Framework: AKWABA — Côte d'Ivoire AI Adaptation Consulting
Report type: Botspeak × Claude AI Literacy Adaptation Study
For the full framework methodology and other market audits: MoctarDatt.com