An Affiliate Marketing Platform Built From Scratch
How Brown Boots went from zero in-house engineering to owning a multi-tenant adtech platform, with one managed delivery team accountable end to end.
- AdTech
- Multi-Tenant SaaS
- Managed Delivery
- Industries
- AdTech, Affiliate Marketing
- Platform
- GoKart, multi-tenant SaaS for offer serving, tracking, and partner management
- Delivery scope
- Managed delivery: product analysis, architecture, front end, back end, UX/UI design, QA, DevOps
ABOUT THE PROJECT
Project overview
- Engagement
- 2023 - ongoing
- Team today
- PM/BA, 2 Full-Stack developers, QA, UI/UX designer, DevOps
- Core technologies
- .NET/C#, React/TypeScript, PostgreSQL, Redis, Azure
Brown Boots is a publisher-first US affiliate marketing agency with 10k+ campaigns delivered and $200M+ in managed budget. They partnered with Limestone to replace fragmented third-party tools with GoKart, a unified platform built from scratch.
After a collaborative discovery phase in 2023, GoKart launched in 2024 and is continuously developed and operated by the same Limestone team through two-week sprints and weekly to biweekly releases.
Challenge
Build the platform the market never had, for a company with no engineers. Four constraints shaped the engagement from day one.
-
One platform from a patchwork
Publishers, advertisers, and networks each stitch together third-party tools that solve one slice of the problem. GoKart's founding idea was to compile the patchwork into one platform.
-
Publisher-first economics
The incumbent tools constrained exactly what Brown Boots needed: control over offer serving and prioritization, transparency into partner performance, and flexibility to run publisher-first economics.
-
Zero in-house engineering
Product analysis, architecture, front end, back end, design, QA, and DevOps had to come from one team, with someone accountable for the whole thing.
-
Build while operating
GoKart had to grow from MVP into a commercial SaaS with paying tenants, enterprise controls, and integrations, without ever pausing the live roadmap.
What we built
Solution
The shipped platform, not the roadmap. From a publisher-only MVP in 2024 to a full multi-persona platform: publishers, advertisers, and affiliate networks on one system, with versioned APIs and an AI-agent surface on top.
-
Offer Server & Display
Three delivery surfaces: a link-out offer wall, an iframe embed, and an API feed. Placement-aware positioning, pinned slots, and EPC-driven ordering.
-
Offer Management
Create, edit, clone, and share offers across tenants. Multi-event offers with per-event capping, scheduled revenue and reward changes, budget auto-pausing, full change log.
-
Conversion System
Click tracking, server-to-server postbacks, attribution including multi-event and repeat conversions, approval rules, snapshots, bulk operations.
-
Partner Platform
The multi-persona pivot: offer sharing with two-way synchronization, granular partner permissions, partner payouts, and partner-scoped reporting.
-
GoKart Exchange
The marketplace surface. Publisher and advertiser profiles with T&C management live since June 2026.
-
Reporting
Timezone-aware across every report, dashboard, and export. EPC metrics, configurable columns, history logs, daily summaries, XLS exports.
-
Integrations
Cost reporting from AppsFlyer, Adjust, and Singular. Offer providers RevU and Wannads, tracking platforms including Impact and CJ Affiliate, plus postback presets for 40+ systems.
-
Platform APIs
Versioned Management and SDK APIs behind dedicated key types, published Swagger, and partner self-service API keys.
-
Enterprise Controls
Two-factor authentication, mandatory T&C acceptance, API key scoping, and rules-based fraud control: IP validation, rate limiting, HMAC verification, event sequence checks.
System Architecture
Multi-tenant by design, scaled without scaling the bill.
C# .NET with BLL/DAL architecture on Azure, React and TypeScript up front, PostgreSQL with a production read replica carrying reporting, conversion, and auto-rewarding workloads, Redis caching, Cloudflare at the edge, Docker-first development with CI/CD quality gates. The performance program bought growth headroom without a proportional rise in infrastructure cost: hot-path cache work, batched tenant refreshes, pre-aggregated reporting queries, and virtual writer and reader endpoints for safe promotion.
DELIVERY TIMELINE
Three years in, still shipping weekly.
The strongest thing about this engagement is that it has held for years and still ships. Everything else supports that.
-
2023
Collaborative Discovery
From zero engineering
-
2024
GoKart MVP Live
Publisher-first offer platform
-
Q4 2025
Multi-Persona Pivot
Publishers, advertisers, networks
-
JUN 2026
Exchange + MCP Live
Agent-ready platform
-
Q3 2026+
Roadmap Continues
Same team, still shipping
DELIVERY RECORD
Predictable, measured, and it does not break.
-
CADENCE
Small releases, relentlessly.
22 production releases in Q1 2026 with roughly 120 tickets completed, zero rollbacks, and all five planned major features delivered. 34 releases from February through June, six to nine every month. Hotfixes deploy within hours of discovery.
-
CONTINUITY
Three role transitions, zero missed releases.
Over seven months the team absorbed a tech lead departure, a DevOps replacement, and a frontend replacement, each with structured knowledge transfer and pre-planned cover. Cadence held through all three.
-
QUALITY
A measured, improving QA function.
72% of defects caught before production across Q1, trending from 64% in January to 86% by early April, with an 88% resolution rate. Automated coverage across the three core domains, integration tests against real PostgreSQL.
THE AI ELEMENT
The platform is agent-ready. So is the engineering process behind it.
-
GOKART MCP
An adtech platform AI agents can operate.
A tenant-scoped MCP service, live since June 2026 on the SDK API: read tools first by design, write tools following in July. Endpoints purpose-built for language models, with response-size and latency work most teams skip. Ready for direct agent consumption, which is still rare in performance marketing.
-
JOINT AI-ERA ENGINEERING
The client's CTO ships through his own agent.
Same repository, same quality gates: the CTO delivers production code through his agent harness while Limestone owns architecture, review, QA, and release. The first production delivery from that workflow validated itself by diffing old and new SQL against production data, row for row.
-
THE HONEST TRADE-OFF
AI made QA the bottleneck. We said so first.
AI-assisted development raised throughput faster than validation could follow. We flagged it in our own monthly memo, ran a QA retrospective, wrote a tiered improvement strategy, and started closing the gap with automation. We measure what AI does to a delivery system, including the inconvenient parts.
ONE OF THE FIRST PILOTS OF VELOCITY CORE, LIMESTONE'S AGENTIC DELIVERY SYSTEM, TESTED ON LIVE GOKART TICKETS
Impact & Results
- A services-only agency became a software vendor: GoKart is a commercial SaaS platform with its own brand, pricing, documentation, status page, and paying tenants.
- Brown Boots now owns its full stack: offer server, tracking, partner platform, GoKart Exchange marketplace, versioned APIs, and an MCP surface for AI agents.
- 22 production releases in Q1 2026 with zero rollbacks; 34 releases February through June, six to nine every month, sustained.
- QA catch rate at 72% across Q1 and climbing to 86%, with automated coverage across Offer Management, the Conversion System, and the Partner Platform.
- Three key-role transitions absorbed in seven months without a single missed release.
- Platform scaled without scaling the bill: read replica, cache, and query work bought growth headroom without a proportional rise in infrastructure cost.
- New tenant onboarding turned into a repeatable, planned delivery stream rather than a one-off scramble.
-
production releases in Q1 2026
22
-
releases February through June
34
-
QA catch rate
72%