top of page
Unified feature flags, experimentation, and analytics into one shared event-stream architecture rather than three separate tools — a bet that paid off so decisively OpenAI acquired the company for $1.1 billion in September 2025, with founder Vijaye Raji becoming OpenAI's CTO of Applications.
1
MODEL
BUSINESS MODEL
SaaS
model bm
HOW THEY BUILT IT
Built a unified experimentation and feature-management platform where a single event stream underpins feature flags, A/B experiments, and analytics dashboards — meaning any feature flag automatically functions as an experiment dimension, and metric anomalies can be traced directly back to specific feature releases. Processes over 1 trillion events daily across 300+ paying customers including OpenAI, Notion, and Rippling.
HOW TO ARCHITECT IT
1) Unify what competitors sell as separate point solutions (feature flags, experimentation, analytics) around one shared data architecture, so a customer replacing three tools with one gets a genuinely better product, not just a bundling discount. 2) Offer both cloud-hosted and warehouse-native deployment (data stays in the customer's own Snowflake/BigQuery/Databricks) so enterprises with strict data residency requirements aren't forced to choose between compliance and capability. 3) Price on metered events rather than seats, so revenue scales automatically as a customer's product usage and experimentation activity grows.
DISTRIBUTION MODEL
Self-Serve Website, Enterprise Sales
dm
HOW THEY OPERATIONALIZED
Offers a freemium self-serve tier that lets individual developers wrap a feature in a flag using one of 30+ open-source SDKs immediately, while enterprise sales targets larger organizations needing advanced statistical tools, compliance features, and dedicated support once usage crosses a threshold.
HOW TO REPLICATE WHAT WORKED
Worked: the shared event-stream architecture meant customers didn't have to reconcile data between separate flagging, testing, and analytics tools — a genuine technical differentiation, not just a marketing bundle. Caution: the company's own analysis notes enterprise buyers requiring strict production governance (audit trails, SSO, workflow approvals) sometimes still preferred incumbent LaunchDarkly's more mature governance tooling — proof that technical elegance alone doesn't automatically win the most risk-averse enterprise segment.
| PATTERNS OF THIS MODEL
PATTERNS IN UNIFYING POINT SOLUTIONS AROUND ONE DATA ARCHITECTURE:
1. UNIFY WHAT COMPETITORS SELL SEPARATELY AROUND A SINGLE SHARED DATA MODEL. Replacing three tools with one must deliver a genuinely better product, not merely a bundled discount.
2. OFFER BOTH HOSTED AND WAREHOUSE-NATIVE DEPLOYMENT so enterprises with data-residency requirements are not forced to choose between compliance and capability.
3. PRICE ON METERED EVENTS SO REVENUE SCALES WITH THE CUSTOMER'S OWN USAGE growth without renegotiation.
4. UNIFIED ARCHITECTURE IS HARD TO EXPLAIN AND EASY TO UNDERSELL. The advantage only becomes visible after adoption, so early proof must come from technical champions rather than positioning.
What companies with this model reveal
| OPPORTUNITY INTELLIGENCE
GOLDMINE 1 — UNIFY WHAT COMPETITORS SELL AS THREE PRODUCTS.
Standard: one event stream underpinning feature flags, experiments and analytics means every flag is automatically an experiment dimension and every metric anomaly traces to a release. Replacing three tools with one is a better product, not a bundling discount.
GOLDMINE 2 — OFFER WAREHOUSE-NATIVE DEPLOYMENT.
Standard: letting data stay in the customer's Snowflake, BigQuery or Databricks removes the forced choice between compliance and capability for regulated enterprises.
GOLDMINE 3 — METER EVENTS SO REVENUE TRACKS CUSTOMER ACTIVITY.
Standard: over 1 trillion events daily across 300+ paying customers scales automatically with their product usage.
THE PIT — YOUR MARQUEE CUSTOMERS ARE ALSO YOUR CONCENTRATION RISK.
OpenAI, Notion and Rippling are exactly the accounts that build experimentation in-house once scale justifies it. Sophisticated technical customers are the fastest to adopt and the most capable of replacing you.
THE SECOND PIT — EVENT-METERED PRICING IS THE LINE ITEM DATA TEAMS AUDIT FIRST.
MOVE WITH CAUTION — WAREHOUSE-NATIVE ARCHITECTURE INVITES THE WAREHOUSE VENDOR TO SHIP THE SAME PRODUCT.
Untapped Business Model / Gaps / Goldmines / Pits
Patterns & Insights
2
MARKET
mkt mt es
MARKET TYPE
Fragmented Market
WHY THEY WON
Feature flagging, A/B testing, and product analytics were historically separate tool categories (LaunchDarkly for flags, Optimizely for testing, Amplitude for analytics), each requiring separate integration and data reconciliation. Statsig won by unifying all three around one data model, a genuinely different architecture rather than a repackaged version of any single incumbent's approach.
ENTRY STRATEGY
Greenfield Entry
EXECUTION
Founded in 2021 by Vijaye Raji (former Meta engineering leader) building the unified architecture from scratch — the technical bet (one event stream powering three previously-separate product categories) required original infrastructure design rather than acquiring any one of the existing point-solution vendors.
FOOTHOLD STRATEGY
fs
Beachhead Strategy
Started with data-forward engineering and product teams at high-growth tech companies (Series A-C startups, AI-native products) who had outgrown manual release processes and were actively replacing fragmented point solutions — a beachhead chosen because these teams valued the unified architecture's technical elegance immediately, without needing convincing.
GROWTH CAMPAIGN
CAMPAIGNS THAT WORKED
Landing OpenAI as both a customer and eventual acquirer created an extraordinarily credible reference point — a company at the center of the AI industry choosing Statsig for its own experimentation needs became a self-reinforcing proof point for every subsequent AI-native prospect.
KEY LEARNING
If you're selling infrastructure to the AI industry specifically, landing one of the frontier AI labs as a genuine customer (not just a logo) is worth disproportionate sales effort — their adoption is read by every other AI company as a signal about what serious infrastructure looks like.
gc
Market Context
| MARKET INTELLIGENCE
THE STANDARD: Unifying historically separate tool categories around one data model is a genuinely different architecture, not a repackaged incumbent.
RULE 1 — SEPARATE TOOLS FORCE DATA RECONCILIATION, WHICH IS THE HIDDEN COST. Flags, experiments and analytics in three systems means three integrations and inconsistent numbers.
RULE 2 — ONE EVENT STREAM MAKES THE WHOLE MORE THAN THE PARTS. Shipping a flag and measuring its effect in the same system is what point solutions cannot assemble.
RULE 3 — UNIFICATION PLAYS LOSE EVERY SPECIALIST COMPARISON. Accept it and compete on the consolidated workflow, not on feature parity.
RULE 4 — EXPERIMENTATION IS A CULTURE PURCHASE AS MUCH AS A TOOL PURCHASE. Organisations that do not test will not renew regardless of product quality.
MARKET TYPE: Fragmented Market (experimentation and product analytics).
| MARKET ENTRY PLAYBOOK
THE STANDARD: UNIFYING SEPARATE POINT CATEGORIES ON ONE DATA PIPELINE IS AN ARCHITECTURAL ENTRY EXISTING VENDORS CANNOT COPY.
RULE 1 — ONE EVENT STREAM POWERING THREE PRODUCTS IS A COST AND ACCURACY ADVANTAGE.
Customers stop reconciling conflicting numbers between tools — the actual pain, not feature gaps.
RULE 2 — FOUNDER PROVENANCE FROM A COMPANY FAMOUS FOR THE PRACTICE IS THE CREDIBILITY.
"Built the way that company does experimentation" is a claim no competitor can assert.
RULE 3 — CONSUMPTION PRICING ON EVENTS ALIGNS REVENUE WITH CUSTOMER SCALE.
Design it in at entry; retrofitting usage pricing onto seats costs churn.
How to enter
| FOOTHOLD STRATEGY PLAYBOOK
THE STANDARD: Sell architectural elegance to teams sophisticated enough to recognise it without persuasion.
RULE 1 — TARGET TEAMS ALREADY REPLACING FRAGMENTED TOOLING. Companies running separate feature flagging, experimentation and analytics have felt the integration cost and are actively consolidating.
RULE 2 — A UNIFIED DATA MODEL IS THE PRODUCT, NOT THE FEATURE SET. Where the same events power flags, experiments and metrics, the value is structural and hard to replicate.
RULE 3 — DATA-FORWARD EARLY-STAGE COMPANIES ARE FAST TO ADOPT AND FAST TO SCALE. They evaluate on architecture, decide quickly, and grow into significant accounts.
RULE 4 — EXPERIMENTATION PLATFORMS ARE ATTACHED TO ENGINEERING VELOCITY. When release cadence slows, so does the value the product can demonstrate.
How to get the first strong position
MARKET PATTERNS & PLAYBOOK
3
MONEY
money rev pri
REVENUE MODEL
Usage-Based
PRICING MODEL
Freemium, Usage-Based Pricing
WHY THEY WON
Reached $40M ARR by May 2025 with an average revenue per customer of approximately $133,000 annually across 300+ paying customers, valued at a 27.5x revenue multiple in its $1.1B acquisition by OpenAI — reflecting the market's expectation of continued rapid growth, not just current revenue.
Metered pricing charges based on the volume of feature flag exposures, experiment participants, and analytics events processed, with a self-serve freemium tier for individual developers and small teams, scaling into enterprise contracts for organizations needing warehouse-native deployment and advanced statistical rigor.
TARGET AUDIENCE
CUSTOMER BUYING BEHAVIOUR
tg cb
Product and engineering teams at high-growth technology companies (Series A-C startups through mid-market SaaS, 20-5,000 employees) who ship software frequently and need to measure real release impact — with a growing high-value segment of AI-native companies testing LLM outputs and AI-powered UX features.
Self-serve, developer-led adoption starting with feature flags alone, expanding to the full experimentation and analytics suite once a team sees the value of closing the feedback loop between a release and its measured outcome — a classic land-and-expand pattern within engineering organizations.
| PRICING INTELLIGENCE
What makes this model effective & make customers pay
Experimentation infrastructure priced on events lets small teams start free and large teams pay proportionally to their data.
RULE 1 — EVENT-VOLUME PRICING TRACKS BOTH INFRASTRUCTURE COST AND CUSTOMER SCALE.
The honest meter for anything ingesting telemetry.
RULE 2 — AGGRESSIVE FREE TIERS ARE THE WEAPON AGAINST INCUMBENTS WITH ENTERPRISE PRICE LISTS.
Where established experimentation vendors quote six figures, a generous free tier takes the entire startup market.
RULE 3 — BUNDLING FLAGS, EXPERIMENTS AND ANALYTICS ATTACKS THREE VENDORS AT ONCE.
Consolidation pricing is evaluated on subscriptions deleted.
RULE 4 — ACQUISITION BY A PLATFORM IS COMMON FOR DEVELOPER INFRASTRUCTURE.
Statsig was acquired by OpenAI in 2025. Infrastructure adjacent to a platform's core need is bought rather than partnered with.
A product team is buying the ability to know whether a change worked. Where the alternative is arguing from opinion, measurement is priced against the decisions it settles, not against a competing tool.
PRICE & REVENUE
| Revenue Risk - The biggest threat to revenue stability
Usage pricing with ~$133,000 average revenue per customer across 300+ customers means a small number of relationships carry the business.
Experimentation platforms are bought by companies with mature product cultures — a narrow buyer population that shrinks when product teams are cut.
Open-source and in-house feature-flagging is always available, which caps price permanently.
Being acquired for infrastructure rather than revenue is a legitimate outcome: a 27.5x revenue multiple prices future growth, not current business.
$40M ARR (May 2025); acquired by OpenAI for $1.1B — the acquirer's strategic need, not the standalone model, set the price.
Where the model can break
4
MOTION
GROWTH EXPANSION MODEL
COMPETITIVE STRATEGY
motion ge cs
Ecosystem Expansion
HOW THEY EXPAND
Expanded from core feature flagging into the full experimentation and analytics suite by unifying them around one architecture from the start, then extended into warehouse-native deployment options — growth came from deepening the single unified platform rather than bolting on unrelated new products.
Differentiation
HOW THEY COMPETE
Differentiates from LaunchDarkly (flags-first, strong on enterprise governance) and Optimizely (testing/personalization-first, marketing-team oriented) by refusing to specialize in just one layer — the unified architecture is the entire competitive thesis, not a single standout feature.
GROWTH ENGINE
GTM
ge n gtm
Product-Led Growth, Freemium User Acquisition
A developer wrapping their first feature flag in Statsig's free tier experiences the unified architecture immediately — seeing flags, experiments, and analytics connected in one place creates an internal case for expanding usage across the team long before any sales conversation happens.
Product-led growth through generous self-serve access to the full platform, supplemented by targeted enterprise sales into AI-native and high-growth tech companies, with the OpenAI relationship functioning as an implicit trust signal across the entire AI industry.
SUSTAINING MOATS
Switching Costs, High Customer Lock-In, Brand Power, Technology Advantage (complex enterprise scenarios)
moat
The unified event-stream architecture means a customer's flags, experiment history, and analytics all live in one interconnected data model — untangling that to migrate to a competitor selling three separate tools is a meaningfully harder technical migration than switching a single-purpose point solution, and the accumulated experiment history itself becomes a data asset a new tool can't replicate on day one.
| MOAT INTELLIGENCE
THE STANDARD: Experimentation infrastructure becomes a moat when the results feed decisions the company cannot revisit without the same measurement.
RULE 1 — THE EXPERIMENT ARCHIVE IS INSTITUTIONAL KNOWLEDGE. What has already been tried and failed prevents teams repeating years of work, and it exists nowhere else.
RULE 2 — FEATURE FLAGS SIT IN PRODUCTION CODE, which makes removal an engineering project rather than a subscription decision — a far stronger position than an analytics dashboard.
RULE 3 — STATISTICAL RIGOUR IS THE ENTERPRISE PURCHASE CRITERION, because once results drive roadmap and budget, methodology defensibility outranks interface quality.
THE SIGNAL: being acquired into a platform that already owns developer workflow is the logical outcome for experimentation tooling. It is a layer of how software is built rather than a category with an independent budget.
Why this company remains defensible
ARR & TAKEAWAY
ARR Journey - what to do at each stage
PRE-$1M ARR — GIVE AWAY THE INFRASTRUCTURE, CHARGE FOR SCALE
Experimentation and feature flags were sold as expensive enterprise tools. Free at generous volumes with usage-based pricing above it undercut the entire category.
Founded by an engineering leader from a company that ran experimentation at scale — the credibility is the go-to-market.
$1–5M ARR — BUNDLE FLAGS, EXPERIMENTS AND ANALYTICS
Competitors sold these as three products from three vendors. One platform at one price is the wedge.
WATCH: events processed and experiments running per account.
$5–10M ARR — WIN THE ENGINEERING-LED COMPANIES FIRST
Atlassian, Notion, Brex, Bloomberg and OpenAI itself became customers — reference logos that sell the category to everyone else.
$10–50M ARR — PRICE ON EVENTS, NOT SEATS
Consumption pricing scales with the customer's product usage and requires no renegotiation.
$50–100M ARR — YOUR BIGGEST CUSTOMER MAY BE YOUR ACQUIRER
OpenAI acquired Statsig in September 2025 in an all-stock deal valued at $1.1B, with founder Vijaye Raji becoming CTO of Applications. OpenAI was already a customer.
Building infrastructure a fast-moving platform depends on is the most reliable route to a strategic exit.
$100M+ ARR — THE ACQUIRED PRODUCT'S INDEPENDENCE RARELY SURVIVES
Operational independence was promised at acquisition. In May 2026 Amplitude announced it would take on the Statsig brand and customers, with the remaining team at OpenAI contributing to an integrated roadmap.
Meanwhile the category consolidated around it — Everstone merged VWO with AB Tasty in January 2026 to form a platform reported at $100M+ ARR.
Rule: when a platform acquires infrastructure it uses internally, it is buying the team and the capability. Assume the standalone product is a transitional asset and negotiate for your customers accordingly.
COPY PLAYBOOK : What Worked → What Failed → What to Replicate → What to Avoid
THE STANDARD: A shared underlying architecture that removes reconciliation between adjacent tools is genuine technical differentiation, not a marketing bundle. It still does not win the most risk-averse buyer.
SEQUENCE:
1. Find adjacent tools whose data customers must manually reconcile.
2. Unify them on one event stream so the reconciliation disappears.
3. Build governance tooling in parallel, or cede the enterprise segment.
WORKED: A shared event-stream architecture eliminating data reconciliation across flagging, testing and analytics.
CAUTION:
1. TECHNICAL ELEGANCE DOES NOT WIN THE RISK-AVERSE ENTERPRISE. Buyers requiring strict production governance — audit trails, SSO, approval workflows — still preferred the incumbent's more mature tooling, by the company's own analysis.
2. GOVERNANCE MATURITY TAKES YEARS and is invisible until you lose a deal on it.
bottom of page