Giving African SMBs an Always-On Growth Team: How We Built Opportunity Radar on Amazon Bedrock
How Fernwood Systems built an agentic business-intelligence platform that finds, scores and drafts action on growth opportunities, so founders don’t have to hunt for them.
The short version
African SMB founders spend a large share of their week on manual prospecting, competitor-watching and scanning for regulatory or funding news. Global sales-intelligence tools are priced and populated for US and European buyers. Opportunity Radar, live at app.useopportunityradar.com, is an agentic platform that watches signals continuously, scores each one against the specific business it works for, and hands the user a ranked list of opportunities with drafted next actions. It runs on Amazon Bedrock in af-south-1 (Cape Town), and a human approves everything before it leaves the building.
The problem: opportunities expire while founders do research
Our customer serves SMBs across Nigeria and West Africa. Four pain points kept coming up:
- Sales: founders spend roughly 15 to 20 hours a week on manual prospecting, and most leads arrive by chance through referrals.
- Talent: hiring is reactive, and the best passive candidates, people who demonstrate expertise in public, stay invisible.
- Competition: competitor moves are discovered after the fact.
- Market signals: CBN and SEC regulatory changes, funding rounds and macro shifts open commercial windows that close before a small team notices.
Tools like LinkedIn Sales Navigator, ZoomInfo, Cognism and Apollo.io have thin African coverage and enterprise pricing. The real gap was not more data. It was context: knowing which of a thousand signals matters to this business this week.
What we built
Opportunity Radar runs AI agents across intelligence modules: Prospect Hunter, Talent Scout, Competitive Pulse, Partnership Finder and Market Pulse. Each follows the same loop:
- Ingest signals continuously from configured sources.
- Understand the tenant. Every business has a persistent profile covering what it sells, who it targets and what it ignores.
- Score each signal against that profile with a Claude model on Amazon Bedrock.
- Rank and surface the best opportunities in a feed.
- Draft the next action, such as an outreach message or a briefing note, for the user to approve.
- Learn from feedback. Actioned, dismissed and rated opportunities tune relevance over time.
The design principle is simple: the agent recommends and drafts, and the human decides and sends. Nothing goes out externally on a tenant’s behalf.
Architecture

Figure 1: Event-driven ingestion feeds a Bedrock-powered agent that scores signals against per-tenant context. Guardrails wrap the model calls.
The signal flow
- Ingestion: Amazon EventBridge schedules and routes signal collection, and Amazon SQS buffers it so bursts are absorbed rather than dropped.
- Agent orchestration: an AWS Lambda function invokes the Amazon Bedrock agent, which reasons over the signal, calls its tools and returns a structured score with a rationale.
- Models: the Claude 5.5 family on Amazon Bedrock, reached through IAM-scoped access with no model infrastructure to run.
- Data: Amazon DynamoDB serves the hot opportunity feed and session state, Amazon Aurora holds relational data, and Amazon S3 holds documents and raw-signal artefacts.
- Identity: Amazon Cognito issues tenant-scoped JWTs. Tenant ID and role are claims, so isolation is enforced at the API and not only in the UI.
Why these choices
| Decision | Chosen | Rejected | Why |
| Model access | Amazon Bedrock | Self-hosted models, single-vendor APIs | No GPU fleet to run, IAM-native access control, Guardrails built in |
| Compute | Lambda, EventBridge, SQS | Always-on containers | Signal processing is bursty. Scale-to-zero removes idle cost |
| Hot data | DynamoDB | Relational store for the feed | Fast reads for the ranked feed, pay-per-request |
| Region | af-south-1 primary | Europe-only | Lower latency for Nigerian users and in-region data residency under NDPA 2023 |
Responsible AI and security
Agentic systems act, so the guardrails matter more, not less.
- Amazon Bedrock Guardrails on model invocations for content filtering, prompt-injection protection and PII redaction.
- Human in the loop. Agents draft and users approve. There is no autonomous outbound action.
- Bias evaluation. Scoring is checked for unfair skew, which matters most in the talent module, where a poor score can disadvantage a person.
- Observability and audit. Amazon CloudWatch and AWS X-Ray trace each invocation, and structured audit events record the agent, the action, the inputs, the output, the guardrail result and the approval status, so any decision can be reconstructed.
- Least-privilege IAM, encryption and per-tenant isolation throughout, designed against AWS Well-Architected principles and NDPA 2023.
What it delivers for the business
- Hours back. The research that used to eat 15 to 20 hours a week of a founder’s time is done continuously by the agents.
- Proactive, not accidental, pipeline. Leads arrive pre-scored against the business’s own profile, not by chance.
- Early warning. Competitor moves and regulatory or funding signals surface when they happen, not after they are common knowledge.
- An African-first price point. A pay-as-you-grow SaaS model built for SMB budgets, not enterprise contracts.
Lessons learned
Plan for regional service topology on day one. Our primary region was chosen for latency and data residency, but not every agentic capability is available there. Designing the cross-region data flow early, and documenting residency implications, avoids a late redesign. It is now part of how we scope every agentic project.
Responsible AI is not a phase. Guardrails, a human approval step and a bias-evaluation process went in from the first release, not as a retrofit. They are now part of our standard agentic delivery method.
Building an AI-powered product on AWS? Fernwood Systems is an AWS Advanced Tier Partner based in Enugu, Nigeria, with an office in Lagos. We design, build and operate production generative-AI and agentic applications.
Opportunity Radar: app.useopportunityradar.com





