The Business Challenge
The business situation the organisation faced before the project began.
The client's sales team needed a single system to find, research and contact prospects. Lead data was scattered across public sources, prospect research and outreach writing were manual, and approvals, notifications and lead routing between team members were not automated.
Why AI Was the Right Solution
Why AI was the right approach — and what alternatives were considered.
Prospect research and outreach involve unstructured information: company websites, documents and public records that vary widely in format. Rules and templates cannot summarise a company, understand a free-text search, or write a relevant first message. LLMs can, and when combined with structured pipelines and scoring they turn scattered public data into research and outreach the sales team can act on immediately. Voice AI extends the same automation to phone conversations.
How the Solution Was Delivered
Discovery, design, development, testing and rollout — the journey, not the tooling.
- 1Worked directly with the client to understand how the sales team found, researched and contacted prospects
- 2Built the core platform (React, TypeScript, Supabase/PostgreSQL) so the team had one place to work
- 3Built automated pipelines that ingest leads from public sources on a schedule, with deduplication and entity resolution
- 4Added a configurable lead-scoring engine so the team could prioritise the best prospects
- 5Added LLM features: natural-language prospect search, AI-generated research briefs and outreach drafts
- 6Built and deployed AI voice agents to handle live phone calls
- 7Automated approvals, notifications and lead routing
- 8Secured the platform with row-level security, role-based access and verified endpoints
- 9Shipped in iterations, gathering change requests from the client and releasing updates that matched their expectations
Key Technical & Architecture Decisions
Architecture, model selection, workflow and trade-offs.
One platform, not point tools: search, research, outreach, scoring and voice live in a single system on Supabase/PostgreSQL, so data flows between them without manual exports.
Data quality before AI: deduplication and entity resolution run before scoring and LLM features, because AI output is only as good as the lead data underneath it.
Buy where it is commodity, build where it is differentiating: proven voice platforms (Deepgram, Voiceflow, Retell) and automation tools (n8n, Zapier) were used for telephony and workflow plumbing, while search, research, scoring and the core platform were custom-built.
Security by default: row-level security, role-based access control, JWT-verified endpoints and restricted CORS from the start, since the platform holds business data.
Challenges & How They Were Solved
Obstacles hit along the way and how they were overcome.
Public lead data arrives in many formats and contains duplicates of the same company or person. Scheduled scrapers with HTML and PDF parsing, plus deduplication and entity resolution, produce one clean record per lead.
Different team members needed different levels of access to sensitive data. Row-level security and role-based access control enforce this in the database itself.
Requirements evolved as the team used the platform. Working directly with the client on change requests, and shipping in small iterations, kept the product aligned with what the team actually needed.
User Adoption & Change Management
How users responded — training, change management and feedback loops.
Adoption was driven by working directly with the client: gathering change requests from the people using the platform and shipping updates that matched their expectations. Because the platform replaced manual steps in their existing workflow rather than adding new ones, it became part of the sales team's daily routine.
Business Outcomes & Impact
The measurable outcomes from the underlying AI Deployment — with the story behind them.
In daily production use by the client's sales team. Manual prospect research and outreach drafting replaced by AI-generated research briefs and drafts; lead ingestion, deduplication and scoring from public sources fully automated on a schedule; live phone calls handled by AI voice agents; approvals, notifications and lead routing automated end to end.
Business Outcome Categories
The story behind the numbers
The platform is in daily production use by the sales team, which is the clearest signal that it solves a real problem.
Work that used to be manual is now handled by the system: prospects are found by asking in plain language, research briefs and first-draft outreach are generated by AI, and lead data from public sources arrives already deduplicated and scored. This lets salespeople spend their time on conversations rather than research and data entry.
AI voice agents handle live phone calls, and approvals, notifications and lead routing happen automatically, so leads reach the right person without manual hand-offs.
Lessons Learned
What surprised the team, what worked well, and what would be done differently.
- Data quality (deduplication and entity resolution) is the foundation for useful AI features
- Working directly with users and shipping iteratively drives adoption
- Combine custom-built differentiators with proven platforms for commodity capabilities
- Security controls belong in the first version, not a later one
- AI is most valuable when it removes steps from an existing workflow
Future Opportunities
Where this solution could go next.
Feedback loops from outreach results into lead scoring, so the scoring improves over time.
Agentic workflows that research a prospect, draft outreach and schedule follow-ups for human approval.
CRM integrations so data flows into the team's other tools automatically.
Usage and outcome dashboards to measure time saved and conversion impact.
Professional Reflections
Guidance for another builder tackling a similar problem.
Fix the data before adding AI: clean, deduplicated, well-structured lead data made every AI feature more reliable.
Use proven platforms for commodity capabilities such as telephony and workflow automation, and spend engineering time where the business is different.
Stay close to the users. Regular change requests and fast iterations matter more than a perfect first release.
Build security in from day one when handling business data.