Impact & grant readiness
Asotele exists to put institutional-grade economic intelligence inside reach of African banks, businesses, and small enterprises that the global data oligopoly was never going to price for. This page is the funder-facing summary of who benefits, how we know, and how grant capital is deployed.
The problem in one paragraph
Bloomberg, Refinitiv, and S&P cover emerging markets thinly, at price points that exclude domestic mid-tier Nigerian banks, regional asset managers, and the millions of SMEs whose FX, inflation, and supply-cost decisions ripple through the real economy. The data itself exists — central bank releases, exchange feeds, food and fuel prices, news, employment, cross-border flows. What's missing is the fusion, modelling, and accessible delivery. Without it, Nigerian institutions make consequential decisions on lagging or fragmented signals, and SMEs make them on rumour.
Theory of change
Asotele is built around a deliberate two-tier structure. Institutional revenue from banks and asset managers is what funds the free SME tier; grant capital accelerates the public layer without creating dependency on it.
- Banks and asset managers adopt the paid tier for treasury, FX risk, and investment-committee work. This is the commercial layer — annual platform licenses, embedded inside Finovamax or deployed standalone.
- Institutional revenue subsidises the free SME tier — multilingual web chat access in English, Hausa, Yoruba, Igbo, and Nigerian Pidgin, grounded in the same forecast engine the banks pay for.
- Open methodology, datasets, and model weights make the platform auditable for regulators, replicable to other emerging markets, and useful to researchers — extending public benefit well beyond the user base.
Grant capital does not substitute for the commercial model. It accelerates the parts that the commercial model can't reach on its own timeline: multilingual interface engineering, open dataset releases, responsible-AI safeguards, public-facing evaluation, and the SME-tier launch.
Who benefits — quantified targets
| Beneficiary group | How they're reached | 12-month target (post grant) |
|---|---|---|
| Nigerian mid-tier and microfinance banks | Paid REST API, Finovamax-embedded, or VPC deployment | 2 production deployments + 3 pilots |
| SMEs & founders (English-speaking) | Free multilingual web chat at asotele.apexgridapps.com | 5,000 monthly active users |
| SMEs & founders (Hausa, Yoruba, Igbo, Pidgin speakers) | Same web chat surface, served in four additional languages | 2,000 monthly active users across the four non-English language surfaces |
| Public researchers & replicators | Hugging Face datasets, methodology page, blog updates | 2 open datasets + 1 fine-tuned model + monthly evaluation reports |
| Regulators & policy researchers | Open methodology, regime-classification outputs, source-grounded briefings | 1 published partnership with a Nigerian university or think-tank |
Use of grant funds
Grant funding is deployed against the parts of the public-benefit layer the commercial model cannot subsidise quickly enough on its own. Indicative allocation:
- Compute & fine-tune capacity — completing the Tier-3 foundation-model fine-tune (Qwen 3.6 35B-A3B + Gemma 4 26B-A4B) on the Nigerian economic corpus. Currently the binding constraint on the SME tier.
- Multilingual interface engineering — web chat front-end, language-routing model, evaluation in Hausa, Yoruba, Igbo, and Nigerian Pidgin.
- Responsible-AI evaluation harness — public benchmarks, hallucination measurement, bias audit, model cards, third-party review.
- Open dataset releases — extending the Hugging Face org with corpus subsets, regime-labelled time series, and the multilingual evaluation set.
- User research & field testing — direct work with SMEs, cooperatives, and trader associations to validate that briefings produce better decisions, not just impressive demos.
How impact is measured
The evaluation cadence is monthly and public. Metrics tracked from day one:
- Reach — monthly active users per language, institutional deployments, dataset downloads, briefings published.
- Quality — forecast error against realised outcomes, hallucination rate on the public evaluation set, regime-classification accuracy, user-reported usefulness.
- Equity — share of users on non-English surfaces, geographic distribution within Nigeria, gender breakdown where users opt in to disclose.
- Influence — citations in central-bank, NGO, and academic outputs; partnerships entered; downstream replication in other emerging markets.
Monthly evaluation reports ship on the project blog and via the engineering newsletter. Funders receive a structured quarterly report with the same numbers plus narrative.
Governance
Applicant entity: Apex Grid Technologies Ltd (RC 9108833) — a private limited company registered with the Corporate Affairs Commission of Nigeria, with registered office in Lagos. Founder and technical lead: Francis Oyakhire.
Advisory council (forming, 2026 H2): Asotele is actively recruiting independent advisors across African macroeconomics, informal-sector labour economics, FX-transmission policy, responsible AI, and Nigerian regulatory practice. Outreach is targeted at Africa-focused academics and ex-central-bank technical staff rather than generic macroeconomics PhDs, so the council can stress-test the regime taxonomy and FX-transmission assumptions from inside the domain. Advisors will be named publicly as they confirm.
Ethics & conflicts: Asotele's open-source posture is intentional accountability. Methodology is public, datasets are downloadable, and the evaluation harness ships alongside the code. Where Asotele forecasts could materially affect Apex Grid commercial outcomes (e.g., publishing market-moving FX forecasts), they are released on the same cadence and at the same granularity to all users — no advance access for paying institutions.
Risk management
| Risk | Mitigation |
|---|---|
| Forecast misuse or over-reliance | Every public output ships with a confidence interval, source citation, and use-disclaimer. Briefings are framed as decision-support inputs, not investment advice. |
| Model hallucination | Source-grounded generation pipeline. Public evaluation set covers factual recall, citation accuracy, and counterfactual reasoning. Hallucination rate is measured monthly and published. |
| Data-quality limitations | Multiple redundant sources per signal (FX from CBN + parallel + crypto-derived). Pipeline degrades gracefully when any source fails. All inputs are dated and versioned. |
| Bias toward formal/English-language signal | Explicit multilingual evaluation, informal-sector signal panel (Sahm-Jobs, layoff-news, WIEGO-style informal-labour data), and documented model-card limitations. |
| Privacy exposure (future SME tier) | Web chat interface launches with privacy-by-design — minimum-necessary data collection, retention limits, no resale of user data, dedicated privacy policy. |
| Unequal access | Free tier is genuinely free — no metering, no premium gating. The cost of inclusion is borne by institutional revenue and grant capital, not by SMEs. |
Status at a glance
- Data pipeline live since November 2025; daily briefings since April 2026
- ARIMA + Markov regime-switching + GARCH forecast engines in production
- VAR + Granger causality engine activating end-May 2026
- Tier-3 foundation-model fine-tune in progress; compute capacity is the bottleneck
- First pilot-bank conversations beginning Q3 2026
- SME-facing free tier (multilingual web chat) launching Q4 2026
- First public evaluation dataset live on Hugging Face; additional releases planned