Seed round · 2026

Winnow

Make millions of automated decisions — faster, and far cheaper.

Nassor Frazier-Silva · nassor<at>gmail.com

The problem

Almost every instant "yes or no" online runs through a rules engine.

Is this card payment fraud? What price do you see? Which ad? Do you qualify for the loan? Is this post allowed?

Companies make millions of these decisions every second — and a piece of software has to check each one against a long list of rules.

Why it's a problem

That software hasn't really changed since the 1980s.

Think of it like this: for every single decision, today's engines re-read the rulebook one line at a time — over and over, billions of times a day.

It's slow

Every extra millisecond is a customer waiting at checkout, or fraud slipping through.

It's expensive

It burns enormous computing power — one of the biggest lines on the cloud bill.

The solution

Winnow answers the same questions in a smarter way.

The shift: instead of reading the rulebook line by line, Winnow checks hundreds of records in a single glance, like scanning a whole spreadsheet column at once instead of cell by cell.

Same answers. Same rules. A fraction of the time and cost.

The payoff

Thousands of decisions in the time others make one.

up to 22,000×

faster than a database

~2.4 µs

per decision

1 server

does the work of thousands

Faster decisions win customers; cheaper decisions widen margins. Winnow does both at once.

Measured head-to-head against real databases — the numbers are on the next two slides.

Proof · measured, not promised

The same rule, the same data — against the tools you'd use today.

One real-time decision: a 10-condition rule across 10 columns, over 3,000,000 records, with no index — because real rule books reference too many column combinations to pre-index.

Winnow~2.4 µs = 0.0024 ms

Answers from pre-built column summaries — it never reads the data row by row.

Analytics database3.3 ms · ≈1,400× slower

Must scan, plus milliseconds of fixed overhead on every query.

Traditional database~54 ms · ≈22,000× slower

Reads every row to answer one decision.

Bars drawn to scale — shorter is faster. Winnow's is the green sliver: too small to see at this scale, which is the point.

And one machine sustains ~12 million decisions a second even while the data is updated live — no locks, no slowdown. Databases stall under that load.

Proof · the cloud bill

Same answers. A fraction of the servers.

To sustain 100,000 decisions per second, around the clock, at typical cloud rates — what a year costs:

Winnow · ¼ of one processor~$90/year

A rounding error on the cloud bill.

Analytics database · ~330 processors~$124K/year

+$124K every year for the same work.

Traditional database · ~5,400 processors~$2.0M/year

+$2.0M every year for the same work.

Bars drawn to scale — the year's bill for the same 100,000 decisions every second.

At a million decisions a second, multiply by ten — the database path runs to $20M a year.

Measured head-to-head at 3M records on a 16-core (32-thread) workstation, July 2026.

Market · where it matters most

The high-volume, real-time decisions.

Fraud & paymentsbanks · fintech

Every transaction screened against thousands of rules in real time. Volume is staggering, and a slow check means lost sales or missed fraud.

Ad targetingadtech

Each ad opportunity matched against millions of campaign rules in milliseconds — the textbook case Winnow already demonstrates at 2M rules.

Dynamic pricingretail · travel

Every page view priced against many rules and segments, constantly, at scale.

Eligibilitylending · insurance

Instant underwriting decisions over rich applicant data — speed is the product.

Market · the sweet spot

Winnow wins biggest where three things meet.

Huge volume

Millions to billions of decisions a day — where a small saving per decision is an enormous saving overall.

Real-time

Answers needed in milliseconds, in the live path of a payment, a page, an ad.

Many rules

Thousands to millions of rules checked at once — exactly what old engines handle worst.

That's fraud, adtech, and pricing first — the workloads with the most decisions, the tightest time limits, and the biggest cloud bills.

Why now · AI

AI doesn't replace the rules. It multiplies them.

How real AI systems work: the model estimates ("this looks 87% like fraud") — the rules decide ("block above 80, unless it's a long-standing customer"). Stripe, HSBC, Amazon, and Visa all ship exactly this split.

Every score, one more decision

Each model score must pass business policy — thresholds, exceptions, legal bounds — on every event, instantly.

AI agents act at machine speed

One request fans out into dozens of automated actions, and every action needs a permission check on the hot path.

The rules layer is where a business steers its AI — and it has to run in microseconds. That layer is Winnow.

Market · size

Multi-billion-dollar software markets, all running this workload.

$32B→ $66B by 2030

Fraud detection & prevention software — growing ~15.5% a year.

$28B→ $276B by 2033

Real-time-bidding infrastructure — ~33% a year — carrying well over half a trillion dollars of programmatic ad spend.

$1.8B+12% a year

Business-rules / decision-management software (BRMS) — the core decisioning tier; broader decision-intelligence platforms run several times larger.

Bottom-up: every company making high-volume decisions already pays for this tier — in software licences and in the cloud compute to run it. Winnow sells into both budgets.

Sources: MarketsandMarkets (fraud detection & prevention, 2025→2030; BRMS, 2025), Market Data Forecast (real-time bidding, 2025→2033), Statista (programmatic ad spend, 2024). Third-party analyst estimates; sizes vary by definition.

Traction

The engine works today.

rules matched in one demo

use-case areas mapped out

design partners in pipeline

A real, working engine with native libraries for Rust, Python, Java, Go, Node.js, C#, and C — teams adopt Winnow from the language they already use, no rewrite required.

Why now

The timing is right.

Decision volumes have exploded — more transactions, ads, and users than ever, all needing instant answers.
Cloud costs are under the microscope — every team is hunting for efficiency, and decisioning is a fat target.
The hardware finally suits the approach — today's chips are built for exactly the way Winnow works.

Team

Why me.

Nassor Frazier-Silva — I previously engineered a similar solution at Unity Technologies, generating tens of millions of dollars in annual cost savings. Recognizing a widespread industry need, I built Winnow from the ground up with an optimized architecture to help other organizations solve comparable challenges.

The ask

Raising US$4M seed.

To turn a working engine into a company. Roughly:

Engineering — cloud product & benchmarks~50%

Go-to-market — land design partners~30%

Runway & operations~20%

Nassor Frazier-Silva · nassor<at>gmail.com