负债6年之后，我用3个月开发的AI CRM做到五位数月收入

When a failure propels you toward a new success

I am a Korean serial founder. I began my startup journey in Hong Kong. I built and sold two companies developing machine-vision-based retail analytics software for shopping malls and government properties. I was working on another startup — also heavily focused on retail — but it crashed violently when COVID wiped out the entire retail solution market overnight. This put me into serious debt, and I spent about six years as a management consultant specializing in large-scale infrastructure and enterprise architecture. That's when I met my now cofounders. We worked together on projects, witnessing patterns of failure in enterprise software. And that led to the concept of what we are building now — an AI Chief Revenue Officer that automates all back-office operations for enterprise and consultative sales processes, turning every seller into a 10-person sales team. Most business information system failures stem from the gap between how humans think and how data is saved. Not everyone can think in spreadsheets. LLMs bridge this gap effectively. We eliminate the entire sales data collection process and use AI agents to execute mundane sales processes, allowing sellers to focus on client interactions. The product launched as an automatic CRM in November 2024 and has consistently pivoted and improved. Currently, it serves around 4,000 companies worldwide, mainly in North America and Australia. We're at a 5-figure MRR, and we're targeting $1.5M ARR by the end of 2026.

Building the product

This was a classic dogfooding case. I have been a loyal HubSpot customer throughout my career, but I found it painful to enforce its use when scaling sales teams. CRMs are crucial for business operations, enabling forecasting, planning, and centralized customer record-keeping. However, salespeople often lack the motivation to perform data entry. So, we started the product with one feature: a simple sales CRM that automatically logs communications from email, LinkedIn, and meetings using AI. I hand-coded the first MVP in about three months. The stack is Next.js + Convex as the backbone. We have many side systems built with Go and Rust, hosted on Google Cloud, for integration and various subsystems that our AI agents use for scraping and generating documents. Convex helped me save significantly on DevOps costs because it handles the entire infrastructure provisioning and workload orchestration via a JavaScript-based SDK — basically, Supabase for NoSQL. This helped me focus purely on business logic, which accelerated the process. The biggest challenge was the frontend. When I started, I didn't even know what Next.js was. But like any other engineering problem, I built, tested, and set up good observability to identify problems faster than users, then rapidly debugged them.

Landing on freemium

We have been bootstrapping this company since day one, with everyone full-time. Fortunately, I am both a developer and a seller. So, we kept our costs very low. We leveraged as many government grants as possible to fund the business, keeping fixed costs as low as possible. Since we were all full-time on this, we had ample time. Money came from grants and our own savings until we became profitable. Our current business model is freemium. We started charging from day one because we didn't want to test or build features based on the feedback of someone who's unwilling to pay. That sort of feedback is mostly nice-to-haves. Burning needs from real customers are more important and should take up 80% of your time. The free tier includes 200 tokens per month, a single user, and two channel integrations. The paid tier ranges from $39 per month per seat up to $149 per month per seat. It varies by monthly tokens (cheaper per token on higher tiers) and optional enterprise security features. We tested many pricing models. These included add-ons such as a monthly fee on channels, token-based pricing, and lifetime deals. Ultimately, we settled on results-based pricing because it makes sealing the deal easier. We then engineer the product to keep margins intact. This also allowed us to navigate significant PLG-driven growth by offering free tokens for specific user actions within the product. We still operate with only the three founders. We are currently raising capital to scale the business, with one pre-seed investor on our cap table. My advice? Charge first, then ensure they feel sufficiently supported. In B2B, people ultimately pay for a sense of security, not features. Features create that sense of security. Good support does, too, while you learn what to build.

Launch and growth playbooks

Our launch strategy was simple. We researched the previous month's top performers on Product Hunt, Microlaunch, Reddit, and Appsumo to understand what the audience was looking for. Then, we created super attractive offers for each platform. The offers often included lifetime deals. Those were crucial because lifetime users expect lifetime access, which provides us with lifetime testers. Most people complaining about LTD promotions expect them to be profitable, but making LTD profitable is very difficult, especially with the gross margin of LLM-based products. We just wanted to acquire 100 core users so that we could continuously improve the product based on their feedback and get referrals to grow our revenue. Another benefit of our launch approach was that most of these launch sites have members who are freelancers and small agencies. We turned those into affiliates and subsidized referrals. After that, we used cold direct sales. These were crucial early on. Content can provide some inbound, but think of it as a wide net you set up. Here's the playbook: You should always set up ad pixels first. Then, properly track funnel events with them. For B2B, build a well-matched audience on LinkedIn Ads. Build in public to show people you are a "real person." With AI, this has become more important because there are exponentially more AI flops and scams online. Collect testimonials at all costs. Send emails, use pop-ups, and gating — testimonials are key content for the bottom of the funnel. Lead magnets still work well, but bundle them with an explainer video and upload it to YouTube, which also helps AI discovery. Once you set up these basic systems, drive SEO and AEO through support articles. This helps activation, retention, and AI discovery. Then, scrape competitor LinkedIn page followers and reach out to them. Tell them what you offer differently. They'll get curious faster than a random audience. Once you achieve steady growth from these inbound & outbound channels, leverage the pixel database to scale them with ads. Hire a performance marketer to handle this. Then monitor CAC/LTV. Now, we've started executing the "headless CRM" strategy. Claude is pushing all existing boundaries for GTM point solutions, but it still requires a robust single source of truth to service a team larger than 5 pax. This allows us to run very strong PLG and integration partnership campaigns to accelerate our growth. This upgrade is going live in 2 weeks, and we look forward to working with agencies and freelancers building great solutions around the Claude ecosystem.

Bootstrapping mistakes

The biggest challenge has been bootstrapping with three people. This significantly constrained our growth — we overfocused on being lean and using AI for everything, but I wish we had sourced good agencies and freelancers earlier when we saw PMF signals. Now that we are in growth motion, the risk of hiring the wrong agency has grown. Overall, I don't think we took enough risks on scaling operations — I would not make this mistake again.

Three pieces of advice

1. Don't spend too much time developing solutions. You are not solving your own problem, and you are not your customer. Many people suffer because they don't know what you already know. Find them, solve their problem, and your business will grow as you do this more. 2. Go to more events and meet new people. Something like 2% of the people you meet can give you a completely new perspective. Find them. Being stuck in the office gives you 0% chance of doing this. 3. And maintain health at all costs.