Lopus AI Thinks Your Growth Team Deserves Analytics That Actually Understand the Business

The Macro: Growth Analytics Is Still Surprisingly Broken

There is a dirty secret in the analytics industry. Most growth teams do not actually use their analytics tools for growth decisions. They use them for reporting. They build dashboards, they send weekly metrics emails, they have quarterly business reviews with charts that go up and to the right. But when it comes time to make an actual decision, like which channel to double down on, which cohort is churning and why, or what is driving the spike in signups from Brazil, they open a spreadsheet and start from scratch.

The reason is simple. Traditional analytics tools, Amplitude, Mixpanel, even Looker, are built around predefined schemas. You have to know what question you want to ask before you can ask it. You need an analyst to set up the events, define the funnels, build the dashboards. By the time the dashboard is ready, the growth team has moved on to a different question.

Heap and PostHog tried to fix this with auto-capture. Track everything, figure out what matters later. That was an improvement, but it created a different problem: a mountain of data with no semantic structure. You have events, but you do not have meaning. “Button clicked” does not tell you whether that click was a signup, a purchase, or an accident.

The next logical step is an analytics layer that actually understands what the data means in the context of your specific business. Not generic event tracking. Not predefined funnels. Something that knows that when your company says “activation,” it means a user who has completed onboarding and sent their first message within 48 hours. That is what a semantic layer does, and adding AI on top of it turns a dictionary into a conversation partner.

The Micro: Dartmouth Grads Who Built for DARPA Now Building for Growth Teams

Lopus AI is building AI-powered growth analytics with what they call an “agentic semantic layer.” In practical terms, this means you can ask questions in plain language, and the system translates them into queries that account for your company’s specific definitions, metrics, and data structures. It unifies structured data (events, transactions, user properties) with unstructured data (support tickets, reviews, chat logs) into a single queryable surface.

The word “agentic” here is doing real work. This is not just a natural language interface bolted onto a SQL engine. The system is designed to chain together multiple analytical steps, pulling from different data sources and applying company-specific logic at each stage. If you ask “why did retention drop last week for enterprise accounts,” it should be able to decompose that into sub-queries, check multiple data sources, and synthesize an answer. That is meaningfully different from a chatbot that generates a SQL query and shows you a table.

Aamish Ahmad Beg and Danylo Borodchuk founded the company. Aamish is CEO and a CS graduate from Dartmouth with a background in full-stack development, infrastructure, and ML. He contributed to DARPA’s DIGIHEALS initiative, which is a cybersecurity program for the healthcare sector. Danylo brings two-plus years of R&D from the DALI Lab at Dartmouth, with deep experience in AR/VR development and product design. They are a two-person team in San Francisco, part of YC’s Winter 2025 batch.

The competitive landscape is worth mapping. Amplitude and Mixpanel own the traditional product analytics space. Looker (now part of a larger company) dominates BI. Narrator AI was an early entrant in the “semantic layer for analytics” space. Census and Hightouch are doing adjacent things with reverse ETL. Eppo and Statsig focus on experimentation. Lopus is trying to sit across all of these by being the layer that understands what the business actually means when it talks about its metrics.

That is an ambitious positioning for a two-person team. But the timing might be right. LLMs have made natural language interfaces genuinely usable for the first time, and the agentic pattern of chaining together multi-step reasoning is getting reliable enough to trust with business-critical queries.

The Verdict

I think the core insight is strong. Analytics tools that understand company-specific semantics are fundamentally more useful than tools that just track events. Every growth team I have talked to has the same complaint: they spend more time explaining what the data means than actually analyzing it.

The risk is scope. Unifying structured and unstructured data, building an agentic query system, and maintaining a semantic layer that stays current as the business evolves is an enormous technical challenge. Most companies that try to be the “one layer to rule them all” end up being mediocre at everything instead of excellent at one thing.

Thirty days from now, I want to see a live demo with a real customer’s data. Not a canned dataset. Show me a growth team asking a question they could not answer with their current tools. Sixty days, I want to know the accuracy rate. When the agentic system decomposes a question and synthesizes an answer, how often is it actually right? Ninety days, the question is whether customers are replacing their existing analytics stack or adding Lopus on top of it. The former is a much bigger business than the latter.