I'm Kyle Lazurko. I build the systems that run go-to-market, then put agents in them so the system runs itself.

I work as a go-to-market engineer. Part operator, part systems builder, and a translator between what users actually do, what the product is changing, and what the business needs to earn.

My work lives where growth breaks most teams. Metrics that cannot answer why. Dashboards that report activity instead of truth. Funnels that hide the real constraint. Fragmented stacks that breed chaos, unreliable reporting that erodes trust, and fragile automations that demand constant fixes.

The core is simple. Define the value moment and instrument the path to it. Create a single source of truth. Centralise data across CRM, payments, funnels, ads and fulfilment. Then run disciplined experiments that compound instead of chasing wins.

What has changed is who does the running. I design swarms of AI agents that operate the pipeline itself, qualifying and routing leads, keeping the CRM honest, triaging operations, drafting and testing marketing, and reporting back in numbers a leader can act on. Not a chatbot bolted onto a funnel. A staffed system where each agent owns a job, hands off cleanly, and leaves an audit trail.

The principle stays the same. Measure what matters. Decide faster. Compound learning into revenue. The difference is that the system no longer waits on a person to move it.

If you can't measure it, you can't manage it.

Selected work

Click any project to open the numbers.

2026

Smaller builds running in parallel. Each one is a place to test the same idea: staff the system with agents and let it run itself.

An offline RAG device, built for environments with no connectivity. The hard question was never the technology. It was who the buyer actually is, which budget line pays for it, and what has to be true before a program office can say yes.

Federal procurement is its own funnel. Registration, solicitation notices, a contracting authority who is the real gatekeeper, and a proposal that has to survive review by people you never meet. I built the positioning, the deck and the path through it.

Bank and brokerage statements, lab results, prescriptions, food logs and training all arrive in different shapes and no two sources agree on a format. Ingestion agents parse each one, a categorisation pass labels what came in, and a reconciliation pass checks it against what the system already believes to be true.

The rule is that uncertainty escalates. Anything the model is not confident about is surfaced for review rather than quietly written, which is the only reason the numbers are trustworthy enough to act on.

Money and health end up measured the same way, in one place, because both compound and both are usually tracked badly.

A CRM that runs a field service business end to end. Contacts, quotes, invoices, warranty certificates, drip campaigns and inventory, with tenants siloed so several businesses run on one instance without touching each other.

The agents sit inside the system rather than beside it. A supervisor routes inbound leads, worker agents own the campaigns and the paperwork, each with scoped tool access, and every action lands in an audit trail so a human can see what the system did and why.

2025 to 2026SaaSYC F25

Go-to-market engineering for a Y Combinator F25 AI research platform. Most specifics are under NDA; the shape of the work is not.

CTRClick through rate

SRSignup rate

CRConversion rate

CPMCost per thousand impressions

CPCCost per click

ROASReturn on ad spend

CACCustomer acquisition cost

MoMMonth over month growth

RRRetention rate

ChurnMonthly churn rate

DAU/MAUEngagement ratio

LTV:CACLifetime value to acquisition cost

Instrumentation, activation, and the experiment cadence that turns a product with users into a product with revenue.

The motion was B2C, self serve researchers signing up one at a time. I built and ran the first structured testing into B2B, designing the strategy tests and campaigns aimed at labs, departments and enterprise research teams, and instrumenting them so each test was judged on pipeline and retention rather than clicks.

The product is now a research harness for vertical research agents. Retrieval over academic corpora, citation grounding at the line level, agent orchestration across search, synthesis and drafting, and an evaluation harness that scores what comes back. Selling that is a different job than selling a tool. It is a system of record for a research team, so the buyer, the security review and the proof of value all change.

The method did not change: find the value moment, measure the path to it honestly, and run experiments that compound instead of experiments that chase wins.

AnswerThis on Y Combinator Launch YC answerthis.io

2024 to 2025FSaaS

Took a service business from tribal knowledge to an instrumented pipeline. One source of truth, automated handoffs, and metrics leadership could act on.

$150K+Revenue in under 14 days

94%Close ratio

21%Pre-qualification rate

15%+Month over month growth

$72Cost per lead

$365CAC

< 5 minSpeed to first contact

8.2:1ROAS

$3,000Average revenue per client

The brief was growth. The actual problem was that nobody could agree on what a lead was, which meant no two reports matched and no decision could be defended.

Defined the value moment, instrumented the path to it, consolidated the stack into a single source of truth, and automated the handoffs that were eating the week. Once the measurement layer existed, the growth work became obvious.

2023 to 2024Marketplace

Two sides of one marketplace. Owner-to-owner platforms for buying and selling distributing properties, without the intermediary taking the spread.

700Users on the platform

Wound downThe rest tells the tale

BuySmarter and SellSmarter were built as a matched pair, one for each side of the transaction. The thesis was that owners of distributing properties were paying a spread to intermediaries for a matching problem that software solves better.

A two-sided marketplace is the hardest go-to-market problem there is, because neither side shows up for an empty room. Seven hundred users signed up and both sides never filled the room at the same time, so the platform wound down.

That is the honest version, and it is the most useful thing on this page. It is where I learned that a marketplace lives or dies on liquidity in one narrow segment before it earns the right to be broad.

2018 to 2021Logistics

The fulfilment arm. Same-day delivery across the region with accuracy held above 99% at both pick and drop.

< 2 hrsAverage fulfilment time

99%Fulfilment accuracy

99.9%Delivery accuracy

2Part-time drivers

600Orders per week

80%OTR

Under two hours from order to doorstep. Two part-time drivers moved roughly 600 orders a week on a routing process that held 99% accuracy on fulfilment and 99.9% on delivery.

Those two numbers are not the same thing and the gap between them is where most logistics businesses quietly lose money. Fulfilment accuracy is whether the right thing left the building. Delivery accuracy is whether it reached the right place. Measuring them separately is what let us fix the right problem.

2014 to 2021Distribution

Built an import and wholesale distribution business from a standing start to a catalogue of 1,200 SKUs, then ran the operation that moved them.

1,200SKUs available for distribution

34%Gross margin

$35AOV

2 weeksLead time, order to shelf

$1M+Annual revenue at peak

< 1%Return rate

Sourcing, importing, warehousing and wholesale distribution under one roof. The catalogue grew to 1,200 SKUs, which is the point where inventory stops being a spreadsheet problem and becomes a systems problem.

Everything I do now in revenue operations started here. When you own the stock, the margin and the delivery promise at the same time, you learn very quickly which numbers are real and which ones are decoration.