AI Platform for Automated B2B Lead Research and Sales Call Preparation
Team Composition:
- Senior Data Scientist
- Senior Python Developer
We partnered with NEXUS aion to design and build an AI-powered platform that automates B2B lead research and sales call preparation. The solution combines human-in-the-loop AI workflows, customizable research steps tailored to the customer’s needs, RAG-based research, and an AI phone agent within a secure, scalable web platform.
- Python (SARIMA, Prophet, XGBoost)
- TypeScript
- OpenAI API
- Remix
- PostgreSQL
- Redis
- Auth0
- Vapi
- Restack Cloud
Background
NEXUS aion GmbH is an early-stage startup building an AI-powered platform for B2B sales teams, with the goal of streamlining how companies research leads, prepare for sales calls, and manage early-stage communication.
The client approached Beetroot with a product idea, technical preferences, and early specifications, looking for a technical partner who could define the architecture, validate feasibility, and move quickly from concept to a working prototype. The collaboration built naturally on a long-standing relationship — the founders had worked with our team across three different companies over the past five years.
From the start, we focused on understanding the business model and user workflows, helping to structure the roadmap, prioritize features, and align development with product goals.
CHALLENGE
B2B sales teams invest significant time researching leads before the first call. Manual research is often inconsistent, lacking the context about industry pain points or relevant case examples to start a meaningful conversation.
NEXUS aion needed a system that could:
- Analyze lead input data such as name, email, company, and job title;
- Generate structured research about the lead, company, and industry;
- Identify likely challenges and relevant value propositions;
- Prepare meeting agendas for sales calls and generate call summaries.
The solution had to support human oversight at each AI step, allow prompt customization, integrate with external systems, and scale into a commercial SaaS product.
Solution
We partnered with the client in a flexible team extension, supporting the product from early concept through MVP release. The project began with a Senior Data Scientist and a Senior Python Developer to validate the architecture and build the foundation of the research engine. Once the core data science phase was complete, the team scaled down to a single Python developer for MVP stabilization and feature expansion, balancing controlled costs with steady delivery. Key contributions included:
-
AI-Powered Lead Research Engine
The platform processes structured lead input and generates research on the person, their company, and industry — covering contextual insights, likely challenges, and tailored value propositions to give the sales team a meaningful head start before every interaction.
-
Retrieval-Augmented Research (RAG)
We implemented vector storage and integrated the client’s case studies and internal knowledge bases into the research workflow, grounding outputs in the company’s own expertise and improving relevance beyond what a general model could provide.
-
Human-in-the-Loop Workflow
Users can review, edit, and refine AI-generated content at each stage of the process. Prompt management tools allow teams to adjust research and call preparation scripts without any code changes.
-
AI Phone Scheduling Agent
We developed a voice agent that can schedule meetings between leads and sales managers, generate post-call summaries, and highlight the core topics discussed. The system supports multilingual interaction and configurable prompt logic, adapting to different sales contexts.
-
Multi-Tenant SaaS Foundation
The MVP includes organization management with role-based permissions, webhook integration, a credit-based usage model, and secure authentication. The architecture is built for incremental scaling and future payment integration as the product grows.
-
Platform & Collaboration Features
The platform supports two organization types (vendors and partners) with lead sharing between them in both directions. User-admins can manage team access and configure notification preferences for leads, demo requests, and research completions. A built-in demo request workflow allows the NEXUS team to provision and manage client demo environments directly through the platform.
Grounding the research outputs in the client’s own case studies and knowledge base, rather than relying purely on the model, was what made the RAG layer actually useful. Keeping humans in the loop at each research step reinforced that: the system gets better as teams refine it, which matters more for a sales tool than raw model performance.
Results
Within one year, the team delivered a stable MVP platform ready for commercial deployment, moving from early concept to operational SaaS without major scope changes. Close communication and flexible task planning kept delivery steady throughout. The client achieved:
-
Automated lead research and meeting preparation
Sales teams can generate structured research outputs and call agendas in minutes, replacing time-consuming manual preparation.
-
40–60% reduction in pre-call research effort
reported by early users, reflecting meaningful time savings across the sales workflow.
-
Integrated AI phone scheduling
The voice agent handles first-touch meeting scheduling, reducing friction in early sales engagement.
-
Controlled and transparent AI workflows
Human-in-the-loop review and prompt management give teams oversight and the ability to refine outputs over time.
-
Production-ready SaaS foundation
The platform supports multi-organization access, role-based permissions, and secure authentication for commercial rollout.
-
Scalable architecture for continued growth
Built to support ongoing iteration, knowledge base expansion, and customer-driven improvements without structural rework.
Your AI product idea deserves more than a prototype:
Discover how Beetroot helps startups and product teams design, build, and launch AI-powered SaaS products — from architecture decisions to production-ready delivery.
