How We're Building Infrastructure That Improves Itself
We used the holiday period to build infrastructure that compounds: automated outreach, self-improving AI, and tools that eliminate manual work. Here's why building systems that build systems is the strategic play for a small team.
Systems That Build Systems
We used the holiday downtime to step back and ask: what infrastructure would let us move faster in 2026?
The context: 39% install growth over the past 30 days, January MosaicLive conversations already at 2x December levels. Growth is ramping up. But churn is still around 45% monthly - the existential problem we need to solve.
For a super small team, the answer isn't just working harder. It's building systems that compound - infrastructure that improves itself so we can focus on the hard problems.
Inbound-Led Outbound
Our scanner has generated 1,319 scans since November, with 601 converting to leads at a 46% rate. But what about the other 700+ who scanned without leaving an email?
We built automation that finds them anyway. When someone scans their store, the system analyzes the URL, uses the Apollo API to find decision makers at that company, and generates personalized outreach: "Someone at your company used our scanner - we noticed a few opportunities."
It's not quite your standard cold outreach. They already interacted with us. We're just following up with context they gave us. The pipeline runs daily at 6am - sync leads, enrich, personalize, upload to Instantly. Completely hands-off. (try the scanner yourself at metricmosaic.io/scanner)
The Self-Improving Flywheel
This is the piece I'm most excited about. We're building a system where MosaicLive can evaluate and improve its own output.
Here's how it works: we analyze churned user conversations to identify where MosaicLive fell short. Did it give a generic response when it could have pulled specific data? Did it miss an obvious follow-up? We feed these patterns into the system, and it generates fixes.
The infrastructure: internal MosaicLive MCP tools let the AI pull real-time merchant data during conversations. We use Claude Code to review platform output and MCP to evaluate responses against what the data actually shows. Instead of manually reviewing every conversation, we're building a flywheel that turns churn data into product improvements automatically.
Same principle as the outreach automation - don't chase one-off fixes, build systems that find and implement fixes on their own.
Quick Answers Without Dashboards
We hand-rolled an agent using the Rill MCP and linked it to Slack. Need to know cost per lead this week? Ask in Slack, get an answer. Which email variant has the best reply rate? Same thing.
No jumping into BI tools. Quick answers where we already are.
SEO That Actually Works
We migrated metricmosaic.io from Vite to Next.js 14 - better performance, proper SSR, dynamic sitemaps, canonical URLs. The result: search impressions doubled overnight. Organic traffic went from rounding error to real channel.
Why This Matters
Last episode I talked about how being a small AI-enabled team turned our weaknesses into advantages. This is the next step: building infrastructure that compounds.
The outreach pipeline runs without us. The self-improvement flywheel finds problems we'd miss. The Slack agent answers questions instantly. Each system we build reduces the marginal cost of the next one.
That's the strategic play for a small team: don't just work harder, build systems that multiply your output.
What's Next
With enablement infrastructure in place, February is about execution. Bedrock migration to reduce AI costs. Winback campaign for churned users once MosaicLive improvements ship. And we locked in our first two annual commitments - early validation that merchants stick around when the product works for them.
Try the scanner at metricmosaic.io/scanner, or start a 14-day free trial.
Shane
CEO & Co-Founder, MetricMosaic