How Second Measure turns billions of credit card transactions into consumer insights

Executive overview

Most investors want to understand company performance in real time, but traditional methods — surveys, quarterly reports, web traffic proxies — are slow, expensive, and imprecise. Second Measure solves this by ingesting billions of raw credit card transactions and building a self-serve analytics platform on top of them.

The core insight is borrowed from video games: instead of writing reports and guessing at questions, give analysts the tool and let them answer their own questions.

Credit card data, when properly cleaned and de-biased, lets investors and companies see real consumer spending behaviour across millions of people — immediately, not weeks later.

From video games to financial analytics

• Michael Babineau and co-founder Lillian both came from Electronic Arts, building high-scale infrastructure and data pipelines.
• A friend at a $30B hedge fund asked how to load two terabytes of data into Excel — the fund had no engineers.
• Most hedge funds have a handful of analysts and no in-house coders; they rely on crude proxies like Google Trends and Comscore for consumer insight.
• The video game pattern — instrument everything, build a self-serve tool, get out of the analyst's way — translated directly to the investment space.
• First product focus: transactional (credit card) data, because investors were excited by it but couldn't handle its messiness.

What the platform shows

• Direct observations of millions of US consumers allow instant answers to questions that previously required costly surveys taking weeks or months.
• Primary investor use case is diligence: verify a startup's numbers, benchmark against competitors, assess relative market share.
• Example: Bird vs. Lime — VCs can see which scooter company is winning in which cities and compare customer spend per month.
• Example: Chipotle food poisoning — revenue impact visible immediately from spending data, not from a sample of 100 survey respondents.
• Consumer behaviour queries include: where else do a company's customers shop, cohort retention curves, lifetime sales at 6/12/24 months, and cross-brand switching.

Editorial and media work

• Second Measure runs a dedicated editorial team — data scientists paired with journalists — to find and publish insights proactively.
• The Wall Street Journal, Financial Times, and others regularly cite Second Measure data to support reporting on companies like Uber and Lyft.
• Stitch Fix finding: becoming a Stitch Fix customer does not cannibalize department store spend — customers simply buy more clothes overall.
• Peloton finding: active Peloton subscribers have surpassed active SoulCycle riders on a monthly basis; SoulCycle issued a non-denial denial in response.
• Amazon Prime finding: an increasing share of Amazon's revenue comes from Prime subscribers; even lapsed subscribers keep spending more than they did before joining.
• The blog doubles as internal dog-fooding: hitting a wall while investigating a question is a direct signal to build that capability into the product.

The data cleaning problem

• A credit card statement description is written by a human, constrained to a short character field, and never designed for machine reading.
• Across 50+ billion transactions, Second Measure sees one billion unique transaction descriptions — for far fewer actual merchants.
• Macy's alone has roughly three million distinct description variants.
• Two root causes: humans set up point-of-sale systems inconsistently (including typos of their own company name), and processing chains introduce further perturbations (apostrophes become spaces, stars, or nothing).
• The entity resolution pipeline maps every variant back to a canonical merchant. It is machine-driven, not human-driven.
• A secondary problem: location. Uber transactions always show "San Francisco" regardless of where the ride occurred; the platform infers true location from surrounding purchases.
• A third layer: de-biasing. The consumer panel is not a perfectly representative US sample, so corrections are applied before any result is published.
• Two separate products: (1) the ingestion and cleaning pipeline (~10–15 people), and (2) the analytics platform — a hyper-specialised Tableau built on top of the clean output.

Expanding from investors to corporations

• Second Measure avoided corporate clients for years to maintain focus, following the YC principle of building something a small number of people love.
• Clayton Christensen's jobs-to-be-done framework shifted their thinking: the core job — helping people understand company performance — is shared by investors and the companies being invested in.
• VCs began bringing the product into boardroom presentations; portfolio company CEOs asked to sign up directly.
• Corporate use cases: competitive benchmarking, tracking competitor revenue, identifying fast-growing businesses to sell to.
• All 150 clients at the time of recording came through inbound — no outbound sales had been run.

Hiring and data science principles

• Second Measure looks for scientists with strong quantitative foundations (statistics, not just tooling); one-third of the team holds PhDs, with backgrounds ranging from statistical genetics to string theory.
• Tools can be taught; mathematical rigour cannot.
• The interview test: give candidates a large, messy, open-ended dataset and ask them to present findings.
• The most common failure: assuming the data is clean, loading it straight into pandas, and reporting results without checking for "dragons" — hidden distortions that invalidate conclusions.
• Strong candidates surface their assumptions explicitly and flag where simplifications may weaken findings.
• The same first-principles mindset applies across data scientists, engineers, and product managers: ask why before accepting the framing of a problem.

Business model and funding

• Revenue comes from subscriptions to the analytics platform; custom one-off research projects are sold separately but not proactively marketed.
• Series A led by Bessemer, co-led by Goldman Sachs, with Citi participating.
• Goldman's rationale: making a push into alternative data (credit card transactions, satellite imagery, geolocation, web traffic) and needed a credit-card-data investment.
• Citi's rationale: the messy-transaction problem is one they feel acutely themselves.
• Seed round was led by Jefferies — investment banks provide unusually strong introductions to both investors and corporates on the East Coast.
• The decision not to sell "signals" to hedge funds was deliberate: as more customers share the same data, any edge from a shared signal disappears. Selling the tool instead of the answer preserves value for every customer.