We Rebuilt Beton: Automated Signal Discovery for PLG Sales
Hi,
It’s Vlad, founder of Beton. Long time no see, huh?
Over the past two months, we’ve rebuilt the product from the ground up. Beton now focuses on automated signal discovery for product-led growth sales teams.
This post covers what changed, why we made this shift, and how the new system works.
Problem
PLG companies generate enormous amounts of first-party data: product usage events, billing records, CRM history, support tickets, and customer attributes. In theory it contains valuable signals about which accounts are ready to expand — but extracting those signals is harder than it looks.
Most of the time, revenue & sales teams rely on intuition: “visited pricing page 3+ times” or “invited 2+ team members.” Sounds reasonable, but rarely a reliable predictor of revenue.
Sales teams are usually the last to get a dedicated analyst, but there’s no budget for them at seed or Series A. Sales ops are armed with pen & paper, when it comes to finding product-qualified leads (aka PQLs).
So, inbound reps spend most of their time on leads that won’t convert. It wastes up to 80% of their capacity — which is money lost both salaries and missed opps.
What We Built
Beton plugs into your PostHog data warehouse and finds patterns that predict account expansion. We validate these patterns against historical conversion data by running 1000s of SQL queries, tracking them automatically, and creating new deals in your CRM when signals fire.
The core difference from existing tools: instead of tracking signals you manually define, Beton discovers signals you didn’t know existed.
Here’s how the technical flow works:
1. Data Schema Analysis An AI agent reads your data warehouse schema and generates hypotheses about potential expansion signals. This includes product usage patterns, billing changes, support ticket content, and cross-attribute combinations.
2. Statistical Testing The system runs thousands of statistical tests via your data warehouse’s API. We only pull aggregated test results — never raw customer data. This keeps sensitive information in your infrastructure while still enabling comprehensive analysis.
3. CRM Integration When an account triggers a validated signal, Beton creates a deal record in your CRM with context about why this account qualified. Sales reps get actionable leads instead of guesswork.
How It Works
Beton’s UI (called Inspector) sits on top of your Posthog DWH and queries data inside via Posthog’s Query API. Even in the cloud version, we don’t get row-level data – we explicitly prohibit this to our agent — and only use aggregated results of these queries.
Agentic system built with Google ADK that utilizes a multitude of agents, each with specific toolsets and goals, to comprehensively analyze signal data while being robust to any user data schema and business practices.
In the self-hosted version, users bring their own API keys. In the future, we will allow self-hosters to use our agent via API – this way Beton’s agent will tell the Inspector (that lives in your infra) what to do and all queries will happen on-premise. The agent will only see the results of tests.
Deployment Options
The codebase will be available on GitHub (we still need to make sure both agent and frontend work properly from one docker). Self-hosting is free if you provide your own LLM API keys.
The cloud version costs $0.5 per tracked user per month with monthly billing. Your customer data stays in your warehouse — we only receive statistical aggregates.
Quick note: The live demo at Beton shows the UI and auth, but the AI agent isn’t connected yet — we’re wiring it up next week along with real PostHog client integration. Feel free to poke around, just don’t expect it to actually work yet.
What’s Next
We’re onboarding the first pilot customers in Q1 2026. If you use Posthog as DWH and have inbound sales — we’re coming after your business.
On the roadmap: backtesting to validate signals against historical data, and continuous monitoring to detect when signals stop working as cohorts change.
We’re also raising an angel round and are already backed with OSS operators. Drop me a line at v@getbeton.ai for more details.
Updates will continue weekly.
Vlad
Links:
Product: https://getbeton.ai/
X: @getbeton
Documentation: Coming in Q1 2026
Book the demo: https://cal.com/beton