-
CustomerFlows CustomerFlows is a modern, AI-powered CRM designed specifically for lead-driven businesses
Inspiration
Enterprise CRMs are built to impress on a feature list, not to be used by actual human beings. Research shows that roughly three-quarters of CRM features go completely unused, and implementations fail up to 70% of the time. Small businesses—like home service providers and tour operators—are forced to pay massive per-seat fees for bloated software, fighting the tool instead of running their business. We wanted to build a lean, focused, AI-native "Lead-to-Revenue" platform that sets up in 10 minutes, features no predatory per-seat pricing, and actually helps businesses convert leads into paid invoices.
What it does
CustomerFlows is a modern CRM that unifies communication, AI lead qualification, and marketing attribution into a single dashboard. • Omnichannel Unified Inbox: Captures every message across Web Chat, WhatsApp, and Email into one provider-independent inbox. • Instant AI Qualification: An AI lead qualifier responds to incoming inquiries 24/7 in seconds, asks industry-specific questions, and automatically populates the pipeline. • Pre-built Industry Pipelines: Drag-and-drop deal pipelines tailored specifically for niche industries (like Home Services and Tour Operators). • True Marketing Attribution: Tracks Profit Return on Ad Spend (PROAS) by connecting the original ad click directly to the final paid invoice, automatically feeding offline conversions back to Google and Meta.
How I built it
We engineered CustomerFlows to be hyper-fast, secure, and infinitely scalable by deploying it entirely on a serverless global edge network: • Frontend Architecture: The user dashboard and web chat widgets are built as a compiled Single Page Application (SPA), statically hosted directly on global edge servers for near-zero latency and instant loading anywhere in the world. • Serverless Backend APIs: Instead of a traditional monolithic server, our entire backend logic runs on distributed serverless edge functions. This allows incoming webhooks from WhatsApp, Email, and Google/Meta ads to resolve in milliseconds at the data center closest to the request source. • Edge Databases & Caching: We utilized a globally distributed SQL database engine for core relational deal data, alongside an ultra-fast key-value store for session management, tracking tokens, and instant deduplication lookups. • AI Core: Integrated Large Language Models (LLMs) tuned with industry-specific prompt structures to handle immediate qualification, context gathering, and automated deal summary generation. • Attribution Engine: Developed an edge-computed data pipeline that attaches source tracking parameters (UTMs/click IDs) to a customer profile upon first visit, persisting that data until an invoice event triggers a secure API callback to ad networks.
Challenges I ran into
• Handling Webhook Bursts at the Edge: When marketing campaigns scale, thousands of ad-click and message webhooks hit our endpoints simultaneously. Managing these asynchronous spikes on serverless functions required architecting a robust distributed message-queueing system to handle throttling without dropping data. • State Management in a Serverless Environment: Because edge functions are stateless, maintaining a continuous, fluid websocket or polling state for the live unified inbox across separate messaging protocols was tough. We resolved this by building a highly synchronized coordination layer utilizing transactional edge storage. • AI Hallucination in Lead Qualification: We had to strictly bound the AI lead qualifier so it wouldn't quote incorrect pricing or promise services outside a business's scope. We solved this by implementing strict system prompts and retrieval-augmented structures based on individual industry templates. • Data Hygiene and Deduplication: When leads come from multiple random channels (a WhatsApp text today, a web chat tomorrow), matching them to the same contact database is incredibly difficult. We designed an automated duplicate resolution algorithm using fuzzy matching logic operating directly on our edge database layer.
Accomplishments that I'm proud of
• Sub-10-Minute Setup: We successfully designed an automated template system where picking an industry instantly configures the pipeline, qualifying questions, and automation blueprints with zero heavy backend provisioning. • The Attribution Loop: Achieving a closed-loop tracking system that accurately passes offline financial conversions back to Meta and Google Ads, giving businesses precise PROAS metrics. • Infinite Scalability, Flat Pricing: By utilizing a distributed serverless and edge-database architecture, our operational overhead doesn't scale linearly with user counts. This technological efficiency allowed us to cleanly offer fair, flat-rate pricing with unlimited contacts and seats.
What I learned
• The Power of the Edge: Moving the database and the compute layer directly to the network edge completely eliminates traditional server bottlenecks and makes global applications feel instantaneously responsive. • Less is More: Building this taught us that users don't want 50 configuration tabs. They want a pre-built blueprint that matches how they already work. • AI is Best as an Assistant, Not a Replacement: AI shines brightest when doing the heavy lifting of 24/7 instant qualification, but the hand-off to a live team member with full conversation summaries is where the actual revenue is won.
What's next for CustomerFlows
• Expanding Industry Blueprints: Launching pre-configured automated pipelines specifically tailored for the Solar and Real Estate verticals. • Expanding Channels: Finalizing the upcoming Facebook Messenger native channel integration and building a "Bring Your Own" carrier portal for SMS / MMS deployment. • Advanced AI Copilots: Expanding beyond the initial intake qualifier to offer deeper core AI copilots that assist live agents with drafting responses and recommending follow-ups inside the unified inbox.
Log in or sign up for Devpost to join the conversation.