‘Dirty Data’ Is Hurting Your Company’s AI Strategy — Here’s How This Tech Founder Is Fixing It

Entrepreneur Ventures is an early-stage venture capital firm partnered with Entrepreneur Media that is dedicated to backing passionate and innovative founders as early as day one. In this series, we are profiling the amazing entrepreneurs Entrepreneur Ventures is working with to share their insights on building and growing a thriving business.

Zac Choi has spent two decades building and implementing data and AI systems for companies of every size, from mid-market SaaS darlings to global enterprises — and he saw a big problem coming before almost anyone else did. As AI adoption exploded, Choi predicted the technology would outpace most companies’ ability to actually use it, because their underlying data infrastructure simply wasn’t ready. That instinct has already paid off once: he built and sold his first startup, String AI, to a telecom partner. Now, with Big Context & Company, Choi is going after an even bigger problem — making enterprise data legible for the AI agents that are about to become its primary users.

Dan Bova: What’s the elevator pitch for Big Context & Company?
Zac Choi: We’re heading into an era where the primary user of data will be an agent more so than a human. The infrastructure we’ve built our data estates on — whether you’re a small business or a large enterprise — has been predicated on human users asking predictable questions. Big Context is essentially helping make those data estates legible for AI-scale work. Before you can implement real AI in your company, you have to till the soil, and that soil is generally your data. We’re an AI-native services company, which means we deliver these transformation services at near-software margins, leveraging AI and technology to do the traditionally unscalable work of services

Can you break that down in simpler terms?
Data comes in from everywhere, at different levels of organized state. If you’re selling on Shopify or through a retailer, or you’ve got a POS system for your restaurant, every one of those sources is capturing data — some of it easy to use, some of it not — you have to become an expert of both the tool and the data underneath. That’s been the issue for the last three decades: data gets produced, but it’s still pretty clunky to use. Now you add AI, which is really just conversational retrieval of data. But when an AI agent looks at a messy data estate, it can only assert what it thinks something means — it has no idea how to guess appropriately. That’s where a  lot of AI implementations are falling short of expectations. It goes back to the old mantra: garbage in, garbage out. We’re basically the garbage men — going in, cleaning house, tilling the soil, so your data is ready for reliable use by AI.

What made you think this was the right business to build?
I think of it in two parts. First, going top-down, there is a perfect storm of market opportunity. I predicted AI implementations would fail before the first studies came out, because LLMs are a probabilistic tool and data work is deterministic — structurally, they’re not compatible. Billions of dollars were going into AI projects, and nobody was cleaning their data fast enough. Second, it’s such a new technology that there’s a dearth of talent — not enough people have successfully implemented this before because it hasn’t existed before. You’d have to have been working on this for the last 2 to 3 years AND already been an expert in data infrastructure. And honestly, it’s something my team and I have deep expertise in: I studied data science at Wharton, spent 20 years building and executing data and AI services, including large data transformation efforts at McKinsey, and my last startup was a production-grade AI product helping small and medium businesses triage their incoming texts and voicemails. 

Where did that background come from — how did your career take shape?
My parents were immigrants from Korea who came here in their early 20s. My dad’s first job — around the time my mom was pregnant with me — was driving around New York City delivering fruit to grocery stores. Humble beginnings. My dad was a quantum physicist and my mom was a child psychologist who traded her career to raise my brother and I. I studied Cognitive Science at UPenn, and my first job out of college was doing research for Wharton, mostly stats and data science. That led to General Mills, then Clorox, where I led a team helping big-box retailers grow entire product categories — this was “forward deployed” work before it became popular in Silicon Valley. From there, I went to Green Dot, a fintech company, where I built their BI, trade and rev ops functions from the ground up through their IPO. After that, I went to McKinsey, where I led data transformations and eventually launched a service line helping mid-market enterprise tech companies — many of whom went on to become unicorns —  innovate, several years before their exits. Watching their trajectory is what convinced me to bet on myself. I left to build String AI, which got acquired by a telco about four and a half years later. A year after that, I started what’s now Big Context & Company.

You’ve raised multiple venture financing rounds and also sit on the investor side of the table. What have you learned about fundraising?
I’ve been fortunate to raise several rounds for several companies, and I’ve also been an advisor and board member to a number of startups. I’m always looking for ways to help other builders and operators too — and actively advise other startups, operators, and even creators on ways to transform their audience into business. In 2024, I met Alex and Leila Hormozi, who run acquisition.com, and convinced them to start a venture fund together — I’ve made more than 30 investments over the last two years and seen a couple thousand deals a year. My take: if you have founder-market fit, that’s a great start. In the past, ideas were cheap and building was hard. Now, building has become cheap too, so everyone’s building something now — which means you need some deep expertise, a real pain point attached to large value pools, and the ability to convince investors and buyers that you’re the one for the job. I’ve noticed that VCs are scrambling to figure out where to invest and at what level of conviction given how quickly things are changing with AI. So, another thing investors weigh is how much of their capital you’ll burn on rookie mistakes, which is why there’s a bias toward second-time founders. It was much easier to raise for my second company than my first.

What are the most common mistakes you see first-time founders make?
Team is number one. Typically in software, most of the capital you burn through is talent — who you surround yourself with, the energy and passion they bring. My co-founder Joji John is also a 20+ year veteran and was a pioneer in BI & data warehouse technology, led AI & Data at Rakuten and most recently built applied knowledge graphs for AI analytics. When you can get the right team in place against a massively urgent & difficult problem, that helps immensely, but you also want to make sure that you can be agile together, make decisions quickly, and have a way forward when disagreements inevitably arise. I’ve been in situations where two people produced more than a team of six or seven. Alignment is the other piece: there’s an unwritten wall around the 15-to-20-month mark where tension develops, usually because someone isn’t satisfied with the output they’re getting for the input they’re putting in — and that’s hard to discover on day zero. The second big mistake is distribution. If it’s easy to create supply for what you’re producing (especially these days when AI is leveling the playing field), distribution becomes the moat. A lot of first-time founders focus too much on product and too little, too late, on distribution. My advice is to treat the market, or customer, like another co-founder. In my second company, I was selling the offering before I’d written a single line of code.

How do you approach problem-solving, whether it’s a business dilemma or something going wrong?
I call it the Zero-One-Two-Three framework: zero-based, first principles, second opinions, third eye. Zero-based means clearing my head of preconceived notions — a VC told me the idea sucked, a customer says that’s the way it’s always been done — and getting crisp on what the actual problem is and who it solves it for, because some problems seem interesting but aren’t worth solving at all. First principles is how we were trained to think at McKinsey: break down the problem to its roots – ask yourself why, over and over, instead of accepting inherited constraints that don’t actually apply to your problem. And second opinions matter, and not just from ChatGPT or Claude, because AI conversations can become an echo chamber — if you keep a conversation going long enough, it starts to drift, and you can talk yourself into believing you’re right. You need a body of people you trust who can think clearly alongside you and can play devil’s advocate too. On top of all that, you develop your third eye: i.e., it’s much better to make a directionally correct decision quickly than to wait for the perfect answer. Over time, with this framework, you develop a real intuition and your decisions tend to be better more often. You’re going to be wrong sometimes, but since you have to move fast, you need to get your shots on goal up — and that’s a discipline you can build.