The information asymmetry problem
Trading runs on information advantage. The fund manager who learns about a regulatory change before it's priced into the market has an edge. The desk that's still reading last night's wire report when news breaks in real time is already behind. In the world of professional trading, information latency isn't an inconvenience — it's the difference between a profitable trade and a missed one.
In 2010, the dominant providers of real-time financial intelligence were Bloomberg and Thomson Reuters. Bloomberg's terminal was the industry standard — comprehensive, deeply integrated into trading workflows, and deeply expensive. Terminal costs ran in the thousands of dollars per seat per month, plus contract commitments. Bloomberg had a lock-in flywheel built over decades: the more traders used Bloomberg, the more the Bloomberg chat and data infrastructure became part of how markets operated, making it harder to leave. Reuters had comparable reach and similar pricing.
Neither was designed for the Indian retail trading market. Neither was priced for smaller brokerages, regional fund managers, or emerging-market traders who needed financial intelligence but couldn't justify Bloomberg-tier costs. The information gap between institutional traders with Bloomberg access and everyone else was real and large.
Heckyl Technologies was founded by four Merrill Lynch veterans — Abhijit Vedak, Jaison Mathews, Mukund Mudras, and Som Sagar — who had dealt with this problem from the inside. Their observation: the explosion of social media and the growth of open regulatory filings and public databases had created a new raw material for financial intelligence. Real-time news, blogs, financial forums, and regulatory disclosures were all publicly accessible. They just needed to be structured, filtered, and surfaced faster than a human analyst could do it manually.
The Decision
The product Heckyl built was a financial intelligence platform that scraped 1.5 million+ sources, ran NLP and sentiment analysis on the output, and presented structured, actionable signals to traders at a fraction of Bloomberg's cost.
By 2017, the product suite included real-time news analysis (company-relevant news aggregated and tagged by entity), market analysis (price movements correlated with news events), sentiment analysis (positive and negative signal from news and social content around specific stocks and sectors), global market trends, and predictive data analysis (forward-looking indicators built on historical correlation models). Heckyl had 14 institutional clients and had raised $1 million from Seedfund while maintaining positive cashflow — an unusual combination for a data analytics company.
The founding decision that shaped everything else was the choice to build on public data rather than proprietary data feeds. Bloomberg's moat is partly the exclusive data relationships — the direct feeds from exchanges, market makers, and institutional players that aren't available on the open web. Building on public data eliminated the exclusive data advantage but dramatically reduced the data acquisition cost and created a different kind of defensibility: the NLP and sentiment models built on top of public data were proprietary, even if the inputs weren't.
This was the right trade-off at this stage and this price point. A $1 seat-per-month product competing with Bloomberg doesn't need Bloomberg-quality data for every signal — it needs good-enough signals for the decisions its users are actually making. A regional fund manager tracking 30 Indian stocks doesn't need microsecond exchange feeds; they need to know when material news breaks on those companies faster than they'd find out manually.
The Product Management Challenge
Being a PM at a data analytics product for financial professionals is different from consumer product management in ways that matter for how you think about the role.
The first difference is that the user's standard for product quality is quantitative and adversarial. A consumer app's success metric might be DAU or NPS. A trader's success metric is P&L. If Heckyl's sentiment signals generate false positives — flagging positive news on a stock that then falls — the user doesn't write a 1-star review. They cancel, and they tell their colleagues. Quality control in financial data products isn't a UX problem; it's a trust problem with binary outcomes.
This creates a specific PM challenge: data quality versus data volume. Heckyl scraped 1.5 million sources. The volume was technically impressive, but volume is not signal. A PM at Heckyl had to define what "verified financial insight" means as a product specification: which sources are reliable enough to surface directly? Which require multiple confirmations before showing up? How do you filter the financial equivalent of spam — low-credibility blog posts, rumour-driven message boards — from the material news that actually affects prices?
The answer to that question required both editorial judgment (which sources have a track record of accuracy?) and technical judgment (how do you model source credibility at scale when you're ingesting 1.5 million feeds?). The PM sat at the intersection of both. Getting it wrong either direction — filtering too aggressively and missing real signals, or filtering too loosely and surfacing noise — directly harmed the product's core value proposition.
The second PM challenge was latency as product. In consumer apps, a 500ms page load is a UX problem. In trading data, a 500ms delay on a breaking news event is the product failure. The trader who acts on Heckyl's news signal 30 seconds after another platform's users have already moved the price has no edge. Latency spec had to be defined by the PM before the engineering architecture was designed, not discovered during testing.
This required the PM to understand the trading use case precisely: what decisions are users making with Heckyl's signals, and what's the time window between when a signal becomes available and when acting on it generates value? For equity traders reacting to news, that window might be minutes. For risk managers monitoring macro events, it might be hours. Different latency requirements, different infrastructure implications, different product promises. A PM who couldn't specify this precisely was building for a product spec the engineering team couldn't design to.
The third challenge was defining the regression model. Heckyl's predictive analysis features required specifying which variables were hypothesised to predict price movement. This is a product decision before it's a data science decision: the PM defines the hypothesis, the data scientist tests it.
The candidate independent variables — company-level news volume, sentiment score, macro indicators, competitor news, regulatory filings, CEO statement tone — each had different data availability, different model complexity requirements, and different validation methodologies. Choosing which to include wasn't a purely technical choice; it was a product bet about what traders actually cared about and what signals would translate into trades. A model that predicted well in backtesting but used variables traders found implausible would fail in the market because users wouldn't trust signals they couldn't understand.
What Worked
The pricing and positioning worked. Bloomberg cost thousands of dollars per month. Heckyl offered financial intelligence at a fraction of that cost for users who couldn't justify Bloomberg pricing but still needed better information than a Google News alert. This wasn't competing with Bloomberg — it was serving a segment Bloomberg didn't serve and had no incentive to serve. The addressable market was large and genuinely underserved.
The institutional client focus worked at early stage. Rather than selling to retail traders — a large, diffuse market with high support costs and unpredictable churn — Heckyl focused on institutional clients: brokerages, fund houses, and corporate treasury teams. Institutional clients have procurement processes, longer contracts, and higher average contract values. Fourteen institutional clients with stable contracts was a better starting point than 14,000 retail users with monthly subscriptions.
The Seedfund raise at positive cashflow also worked as a signal. A data company that is cashflow positive before its Series A has demonstrated that someone is paying real money for the product — not a subsidised or free tier, but genuine commercial exchange. That's a meaningful proof point in a category where many companies burn cash for years before demonstrating revenue.
What a PM Should Take From This
Heckyl represents a PM challenge that's increasingly common but rarely taught: building data products where the raw material is public but the value is in the processing. The product isn't the data — Bloomberg has better data. The product is the intelligence layer built on top of publicly available data, and the quality of that layer is entirely a product management question.
The key principle: define what "good signal" means before building the data pipeline. This sounds obvious but is consistently violated. Data product teams often start by ingesting as much data as possible — build the pipeline first, define quality later. In financial products, this approach ships noise to users who are making real decisions with real money. Quality specification precedes data collection.
The KPI question is also worth working through explicitly: what metrics tell you the product is providing genuine financial insight, not noise? In a consumer app, you might measure DAU, retention, and NPS. In a financial data product for professional traders, the relevant metrics are different: signal-to-noise ratio (what fraction of surfaced signals turn out to be material?), latency distribution (what percentage of signals reach users within the target time window?), and client outcome tracking (do users who act on Heckyl's signals achieve better performance than their baseline?). These metrics require a closer relationship with users and more sophisticated measurement, but they're the right metrics for the product promise.
The line between PM and data scientist at a company like Heckyl is deliberately thin. PMs in data products need to be analytically fluent without being analysts — able to specify a hypothesis, read a confusion matrix, and understand what a correlation coefficient means without writing the model themselves. That's the competency bar for this kind of product work, and it's higher than most PM role descriptions acknowledge.