“Consolidating onto CockroachDB meant we could build our entire AI platform on infrastructure we already trusted. We didn’t need another vendor, another contract, or another compliance audit.” — Garrett Browne, CTO, BetterTracker
Based in Maryland, BetterTracker is an AI-powered business optimization platform and digital ecosystem built to help MSPs and SMBs gain complete visibility into their technology, vendors, contracts, subscriptions, expenses, and shadow IT.
By combining technology stack data, financial transactions, contracts, and vendor intelligence into one system, BetterTracker helps organizations uncover waste, reduce risk, improve accountability, and turn technology data into actionable business intelligence. Features like NaviStack, its vendor portfolio visualization tool, and an AI-powered tool categorization engine have helped thousands of MSPs and SMBs make smarter decisions about their technology stacks.
BetterTracker has been a CockroachDB customer since 2021, using it as the transactional backbone of their entire platform. The decision came down to five core requirements:
A fully managed database that could scale horizontally without operational overhead or manual sharding
High availability with no tolerance for downtime on a live marketplace
Guaranteed consistency across every transaction (no stale reads, no data conflicts, no edge cases)
Multi-region readiness built in from the start without needing to re-architect later
A compliance posture capable of handling sensitive customer data across SOC2, PCI, and HIPAA requirements
Challenge: Building AI on a foundation that was never designed for it
As BetterTracker expanded its product roadmap into AI, the engineering team faced a problem familiar to many fast-growing SaaS companies: their transactional database and their AI infrastructure were two completely separate systems.
BetterTracker was building several AI-powered capabilities on top of their existing CockroachDB data including:
RAG-based AI assistant that surfaces intelligent insights from platform data back to partners and vendors
Tool classification engine that automatically categorizes free-text software submissions into standardized taxonomies (e.g. mapping a vendor-supplied entry to “Accounting Software”)
Canonical company matching to identify and de-duplicate companies across records sourced from different customers and systems
OCR-based document retrieval for contracts and customer interaction records
To power these features, the team adopted a standalone vector store alongside CockroachDB. While this got them off the ground quickly, it introduced challenges that compounded over time.
Architectural complexity and extra latency
Two separate databases meant every AI-powered query required an additional external API call to a standalone vector store. This added latency and introduced a second system that had to stay in sync with CockroachDB through ETL pipelines. More moving parts meant more potential failure points and more engineering overhead to maintain.
Poor observability and debuggability
The standalone vector store’s UI made it difficult to inspect what was actually stored in an index, trace why a query returned unexpected results, or diagnose issues quickly. When something went wrong with an AI-powered feature, the team couldn’t easily see inside the vector store. This created a blind spot in their operational stack that grew more concerning as AI features moved closer to the core product experience. Changing or finding specific data required context-switching to a system with limited visibility and unfamiliar tooling.
Cost unpredictability
The standalone vector store’s pricing model scaled with the number of indexes and agents, creating a baseline monthly cost that existed regardless of usage and a growth curve that threatened to scale linearly as BetterTracker added customers. For a marketplace business, that model was a fundamental mismatch: adding a new customer should not mean database costs go up proportionally.
Compliance fragmentation
With financial data and AI workloads split across two systems, maintaining a clean compliance boundary for SOC2, PCI, and HIPAA requirements became increasingly difficult. Two vendors meant two audit scopes, two sets of controls, and two potential exposure surfaces.
“Every vendor you add is another thing to monitor, another contract to manage, another system your team has to know. Consolidating onto CockroachDB wasn’t just a technical decision — it was about keeping our team focused on building Betty, not babysitting infrastructure.” — Erica Lira, Director of Engineering, BetterTracker
Solution: CockroachDB as the unified OLTP + vector database
BetterTracker’s team began evaluating CockroachDB’s built-in vector search capabilities in mid-2025, working closely with the Cockroach Labs engineering and product teams to run a structured proof of concept.
The goal: determine whether CockroachDB could fully replace their standalone vector database and power their entire AI roadmap from a single system. Several capabilities proved decisive.
Unified OLTP + vector search
CockroachDB’s vector indexing meant BetterTracker could store, query, and search vector embeddings directly alongside their transactional data — eliminating the need for a separate vector store entirely. One codebase, one operational model, one team responsible for the full stack. No sync pipelines, no drift, no second system to keep running, and no extra API call on every query.
Native multi-tenancy at the index level
CockroachDB’s ability to scope vector indexes by primary key gave BetterTracker account-level data isolation out of the box which solved a key architectural requirement for their multi-tenant platform without custom workarounds. This directly unlocked their per-customer AI agent model: every partner on the marketplace can have their own isolated AI experience without BetterTracker building custom isolation logic.
Storage efficiency via quantization
CockroachDB’s vector quantization compressed embeddings from approximately 3KB to ~200 bytes per vector which is a critical advantage as BetterTracker plans to scale to hundreds of thousands of embeddings across dozens of tables. This is the difference between a manageable storage footprint and one that becomes a cost problem as the AI platform grows.
Predictable, consumption-based pricing
Vector queries are billed as standard CockroachDB transactions (Request Units), with no separate per-index or per-agent fees which gives BetterTracker cost predictability as their AI platform scales. The baseline cost of a dedicated vector vendor is gone entirely, and growth no longer triggers a proportional cost penalty.
Compliance continuity
By keeping everything inside CockroachDB Cloud Advanced, BetterTracker inherited the existing SOC2, PCI, and HIPAA compliance posture they had already established. One audit boundary, one vendor, one set of controls.
The migration itself was methodical. The team started with a staging environment, leveraging CockroachDB's free serverless tier to test at zero cost, before moving to production. Working iteratively with Cockroach Labs engineers, they vectorized their first production tables (company records and a vendor lookup catalog), migrated embeddings from the standalone vector store into CockroachDB, and cut over their AWS Lambda functions to point directly to CockroachDB. Finally, they shut off the standalone vector database.
Results: What it actually meant for the business
Replacing a vector database is a technical milestone, but it unlocked a business one. With a single system handling BetterTracker’s full data stack, the team did not just eliminate a vendor, they removed a ceiling on how fast they can grow.
One database, not two
BetterTracker’s entire data stack — transactional, relational, and vector — now runs on a single CockroachDB cluster. Eliminating the standalone vector store removed an entire vendor relationship, along with the ETL sync pipelines that kept the two systems in step. The engineering team no longer maintains a separate data pipeline between their OLTP store and vector database, reducing the surface area for bugs, data drift, and unexpected outages.
Vector search live in production
Semantic similarity queries (cosine and L2 distance) are running in production across multiple tables, powering tool classification and canonical company deduplication.
Cost consolidation and predictability
Removing the standalone vector store eliminated its baseline monthly cost entirely, along with the per-index and per-agent fees that threatened to grow linearly as BetterTracker scaled its AI platform to new customers. CockroachDB’s consumption-based pricing means vector workloads now scale predictably with actual usage, not with the number of AI agents or indexes deployed. For a company whose business model scales with customer count, this was fundamental.
Engineering overhead eliminated
Managing a standalone vector store meant maintaining a separate operational context entirely: a different UI, different debugging tools, and a different mental model from the CockroachDB environment the team already lived in. When something needed to be changed, found, or fixed in the vector layer, it required context-switching to a system with limited visibility.
By consolidating onto CockroachDB, the team now debugs, queries, and inspects their vector data the same way they work with everything else: in one place, with the same tools. That reclaimed capacity is now going toward building the AI features their customers actually need.
The move away from the standalone vector store was part of a broader platform restructuring rather than an isolated migration, making it difficult to attribute specific time savings to any single change. What the team can say clearly is that operating one system instead of two removed an entire category of maintenance overhead, and that the engineering capacity that once went toward keeping two systems in sync is now going toward building a fully agentic AI platform.
Unified observability
With a standalone vector store as a separate system, BetterTracker’s operational visibility was split. Debugging an AI feature might require checking CockroachDB query logs, then switching to the standalone vector store’s dashboard, then reconciling what was seen across both with limited tooling on the standalone vector store side to make that easy. Now, vector data lives in the same cluster, queryable with the same SQL tooling the team already uses. When something needs to be investigated or changed, it is one system, one interface, and one team who already knows how to work in it.
Always-on reliability
For a live marketplace that thousands of MSPs and vendors depend on daily, downtime is not just a technical inconvenience, it is a direct threat to revenue and customer trust.
CockroachDB’s distributed architecture delivers automatic failover with no single point of failure, meaning the BetterTracker platform stays available even during hardware failures or cloud disruptions. By keeping vector search inside CockroachDB rather than routing through a separate vendor’s API, BetterTracker eliminated an entire external dependency that could have caused AI-powered features to degrade or fail independently of their core database.
In the time BetterTracker has run on CockroachDB, the team has never experienced a customer-facing outage. For a live marketplace that thousands of MSPs and vendors depend on daily, that reliability record is not just a technical metric, it’s the foundation of customer trust.
Zero-disruption migration
Adding vector columns and indexes to production tables caused no measurable regression on BetterTracker’s core transactional workloads. The migration was staged deliberately and validated on CockroachDB’s free Serverless tier before touching production. The marketplace continued processing vendor transactions, partner activity, and contract data without interruption throughout the entire cutover.
The migration was completed without a single incident. The team began with their vendor lookup catalog vectorizing it, migrating embeddings, and cutting over without any disruption to the live marketplace. The platform kept processing transactions throughout the entire cutover.
Built for what’s next
Replacing the standalone vector store solved an immediate problem. But the bigger value of the decision is what it makes possible going forward.
BetterTracker is now building a fully agentic AI platform — one where every partner on the marketplace gets their own AI-driven intelligence layer, pulling real-time insights from platform data. The flagship of that vision is BetterTracker Betty: “Your stack spend, contract, and customer intelligence AI agent.” Betty is designed to surface the insights channel professionals need most such as financial benchmarking, vendor recommendations, contract analysis, directly within the platform, in real-time. The architecture built on CockroachDB is designed to support thousands of logical AI agents running as tenants in a single cluster, without per-agent cost penalties and without compromising data isolation between customers.
As the platform scales, so does the partnership. Their roadmap to extend vector search across 100+ tables and hundreds of thousands of embeddings is already underway. Their long-standing ambition to expand internationally to the UK, Australia, and New Zealand, which points toward CockroachDB’s native multi-region capabilities becoming an increasingly important part of the story. And as BetterTracker takes on more regulated channel partners, the compliance-first foundation they built will matter more, not less.
“We’re expanding into new markets, onboarding new regulated partners, and building AI features our customers have never had before. The fact that we don’t have to think about our database — that it just scales, stays compliant, and keeps running — is what lets us move as fast as we do.” — Erica Lira, Director of Engineering, BetterTracker
