Klipy: COVID Debt, Six Wasted Years — How a 3-Person AI CRM Reached 5-Figure MRR and 4,000 Companies

Dollar figures include a rough conversion at ¥150 to $1.

Success stories usually only tell you about the climb. What makes this one worth engaging with is that the detour in the middle, ‘six wasted years’, is described specifically, in the founder’s own words. Jung Hong Kim, originally from Korea and now based in Hong Kong, had already sold two companies built around machine-vision retail analytics. But his third venture was wiped out by COVID, along with the market it served, leaving him in debt. From there he spent roughly six years as a management consultant before launching the AI CRM ‘Klipy.ai’ in November 2024. As of July 2026, it has roughly 4,000 client companies and 5-figure monthly revenue (over $10,000/month, roughly ¥1.5M+).

The disclosed numbers

What Klipy sells

In his own words, Klipy is positioned as ‘an AI Chief Revenue Officer that automates all the back-office work of enterprise and consulting-style sales.’ At the implementation level, an AI agent automatically logs email, LinkedIn, and meeting activity and pushes it into the CRM. It’s designed to eliminate the manual data entry that salespeople hate, rather than make that entry easier.

The idea has two sources. One is his observations from six years of consulting work. The other is his own frustration using HubSpot at scale. The former supplied the language for the structural problem. The latter supplied the felt pain point.

The decisive move was reframing the question

The turning point in this case was neither fundraising nor virality but the moment he shifted how the problem was defined.

Having spent years watching companies implement systems as a consultant, Kim concluded: ‘Most failures in business information systems come from a mismatch between how people think and the format in which data gets stored.’ CRMs go unused, in this diagnosis, not because they lack features, but because the shape of human workflow and the shape of data entry don’t fit together.

The move that follows from this diagnosis isn’t ‘build an easier-to-use CRM.’ It’s ‘eliminate the act of data entry itself.’ Klipy automatically converting email, LinkedIn, and meeting records into CRM entries is the direct consequence of that diagnosis.

Looking at the timeline, the period between reframing the question and actually building it was short. He wrote the MVP himself in three months and launched in November 2024. About a year and eight months later, he’d reached roughly 4,000 companies. Even so, the source doesn’t disclose month-by-month progression, so exactly when growth accelerated can’t be traced. In fairness, it should be said plainly: the speed of the initial ramp is confirmed, but the shape of the growth curve is not.

And what set up this turning point wasn’t the failure itself, but the six years that followed it. Paying down debt while observing other companies’ system failures as a consultant sharpened the precision of the question for the next venture. As he puts it plainly, ‘It took me about six years to get out of debt’, the part success stories tend to skip.

Breaking down how the launch was built

The acquisition channels weren’t one thing. Several distinct kinds ran in parallel.

Launch platforms, Product Hunt, Microlaunch, Reddit, Appsumo. What’s notable here isn’t just using them, but that ‘I researched what had ranked well on each platform beforehand and built an attractive offer before launching.’ He read what each platform rewards before using it as an exposure mechanism.

Lifetime deals, not treated as a profit mechanism. His reasoning: ‘Lifetime users expect permanent access, which means we get permanent testers.’ The framing is that of building a base of 100 core users and affiliates, an investment in a long-term feedback-giving cohort, not a discount play.

Cold, direct outreach, described as ‘critical’ in the early days.

LinkedIn ads, effective when well matched to the target segment. He also used LinkedIn to reach competitors’ users directly.

Content and SEO, support articles, explainer videos on YouTube. He’s also been building in public. This is framed less as an acquisition tactic and more as a way to counter skepticism toward AI products by demonstrating that the thing is real.

Pricing wasn’t settled in one shot either. After trying add-on-based, token-based, and lifetime models, he landed on ‘outcome-based pricing (per-user tokens).’ ‘Get them to pay first, then make them feel well-supported. In B2B, what people ultimately pay for isn’t features,’ he explains. ‘It’s peace of mind.’ A deliberate choice not to run the usual sequence of let-them-use-it-free-then-charge-later.

What didn’t work

The biggest self-criticism he offers is, unexpectedly, that he saved too much. ‘The biggest issue was bootstrapping with three people. It considerably constrained our growth. We focused too much on being lean and doing everything with AI. Once we saw signs of PMF, we should have brought on a good agency or freelancers sooner.’ He states directly: ‘We won’t make that mistake twice,’ and ‘we didn’t take enough operational risk to scale.’

When running lean becomes an end in itself, supply-side capacity can’t keep up the moment demand takes off. This is a point worth carrying forward as the flip side of ‘start small.’

There was a technical learning curve too. He wasn’t experienced with the frontend framework (Next.js) at first, and got through it via fast iteration and building out proper observability. Even with two prior exits under his belt, he was learning as a beginner in a new domain.

How far does this generalize

What’s easiest to reproduce is the sequencing: research what gets rewarded at a given launch venue before you launch there. Treat a lifetime deal as a tester-acquisition tool rather than a profit tool. Charge before offering the free-then-upsell path. None of these require capital. The standard he cites is broadly applicable too: ‘Most of the feedback you get for free is nice-to-have requests. Spend 80% of your time on the urgent needs of customers who are actually paying you.’

On the other side, the non-replicable conditions are clear. First, having sold two companies and having enterprise-architecture consulting experience is what underpins the precision of his problem definition. It’s natural to assume cold outreach worked well early on precisely because the person doing it understood the language of enterprise sales. Having all three co-founders able to work full-time, and having a pre-seed investor on board, are also premises that differ from a solo side project. And the least reproducible thing of all is the six years itself. It was only by watching other companies’ failures while paying down debt that the question, ‘the mismatch between how people think and how data is stored’, emerged. This isn’t really a story about ‘AI CRM hit big.’ It’s more accurately read as a story about how what he observed during the detour determined the precision of his next shot.

Related reading

• How ScrapingBee, still a small-team SaaS, reached a sale
• How Plausible reached $1M ARR with zero ad spend