5x cheaper query responses with Gemini context caching
$100B in annual housing payments processed
95% of member interactions handled autonomously by AI
96% of merchant guest inquiries handled using AI concierge
Bilt partners with Google Cloud to power AI concierges for Bilt Neighborhood OS and scale $100B in annual housing spend.
Bilt partners with Google Cloud to power AI concierges for Bilt Neighborhood OS and scale $100B in annual housing spend.
Scaling a unified neighborhood commerce layer for where people live
Nearly half of U.S. residents today struggle to keep up with their monthly rent or mortgage payments.1 At last count, almost 23 million renter households spent 30% or more of their income on rent and utilities, and 12.1 million of those spent more than half of their income on housing costs.2 Bilt was founded in 2021 as a way to offset this cost-burden crisis. Its platform brings high-value rewards to renters and homeowners, enabling them to earn points on housing payments—often their largest monthly expense—without fees.
Today, Bilt powers one in four U.S. apartment buildings through its Bilt Alliance, and more than 8 million members use the platform to earn and redeem the highest value rewards currency on the market across real estate, travel, dining, fitness, sports, and more. The company is on track to process more than $100 billion in annual housing spend and routed nearly $12 billion to neighborhood merchants to date.
To connect the customer journey and help partners attract local business with a network of 50,000 merchants nationwide, Bilt decided to expand into a neighborhood-focused membership program. This required an enterprise infrastructure capable of handling extreme, recurring traffic spikes. This is especially important on the first of every month, when rent is commonly due, and daily active users in the Bilt app roughly double, putting excessive strain on the system.
To overcome database capacity constraints and infrastructure scaling limits during these peak moments, Bilt turned to Google Cloud. Using Gemini alongside pay-as-you-go overflow capacity, Bilt has now launched an AI concierge that reserves bandwidth for the busiest day of the month without compromising speed or reliability.
1 Source 1
2 Source-2
We chose ADK because we'd rather run a flexible system with guardrails as opposed to a rigid workflow. We've seen amazing tool-calling accuracy, and that's what matters when the agent is moving someone's reservation or helping answer a question about their upcoming housing payment.
Abhi Maddineni
Tech Lead, Concierge, Bilt
Building an AI-driven Neighborhood OS with Gemini and ADK
To expand beyond housing payments, on November 10, 2026, Bilt launched Neighborhood OS, an AI-driven platform that connects multifamily property managers, sports organizations, travel brands, and local merchants with nearby residents. This interconnected ecosystem helps businesses find new customers while enabling customers to spend their Bilt Points to support their communities.
Most local merchants rely on disconnected systems for reservations, customer outreach, payments, and marketing. Neighborhood OS replaces historically fragmented tools with an AI agent interface that acts on the merchant's behalf.
Bilt created this AI concierge using Google Cloud’s Agent Development Kit (ADK) and Gemini, featuring a primary routing agent that orchestrates 13 specialist agents across dining, travel, rent, rewards, and fitness. The agentic concierge now handles approximately 95% of member conversations without requiring human handoff, up from 86% just six months prior.
The concierge tool can answer guest questions, check availability, and manage reservations around the clock. For example, if a guest is running late for a 7 PM dinner reservation, they can text the restaurant at any time to request a change and can even adjust the party size. The dining agent will then check availability, move the reservation or send other available options, and confirm. In September 2026, the merchant concierge handled 96% of these guest conversations autonomously across more than 100 restaurants.
To maintain this high performance at scale, Bilt also uses advanced Gemini features, including function calling, structured outputs, grounding with Google Search for menus and reviews, and multimodal understanding for documents and uploads. By implementing context caching through the Gemini API, 89% of input tokens for dining queries are served from cache, making these responses approximately 5x cheaper.
We want agents in nearly every interaction between a member and their neighborhood, and we want to know, conversation by conversation, whether it actually helped. Every agent interaction lands in BigQuery, and Gemini-based judges grade those conversations where the data already lives.
Andrew Fitz
Engineering Manager, Concierge, Bilt
A unified data and AI ecosystem on AlloyDB and BigQuery
Behind Bilt’s autonomous agents is a fully managed Google Cloud database and analytics architecture designed for operational simplicity. Every concierge service runs on its own dedicated AlloyDB for PostgreSQL cluster, handling chat history, agent services, outreach, and long-term memory. Agent memory is stored using pgvector on AlloyDB paired with Gemini embeddings, eliminating the need to deploy and manage a separate vector database. Data from AlloyDB is replicated seamlessly into BigQuery using Datastream for real-time analytics and continuous evaluation.
Because agent interactions and model evaluation reside in the same ecosystem, Bilt uses Gemini inside BigQuery to evaluate customer conversations and resolutions. “We want agents in nearly every interaction between a member and their neighborhood, and we want to know, conversation by conversation, whether it actually helped,” says Kosta Krauth, CTO at Bilt. “Every agent interaction lands in BigQuery, and Gemini-based judges grade those conversations where the data already lives.”
By offloading database administration and infrastructure management to Google Cloud's managed services, Bilt's engineering team redirected its focus away from database operations toward refining agent quality, tool accuracy, and product innovation.
This end-to-end, autonomous agent workflow will keep track of open requests and close the loop instead of waiting for guests to ask again. For merchants, that means repeat visits without a marketing team. For us, it means we can add merchants and scale support to everyone.
Kush Parikh
VP, Concierge, Bilt
Looking forward, Bilt plans to move the agent even further from answering questions to acting on its own. Merchants can set up goals within the concierge and the agent will work toward them, including the follow-up when a guest replies.
“This end-to-end, autonomous agent workflow will keep track of open requests and close the loop instead of waiting for guests to ask again,” says Kush Parikh, VP, Concierge at Bilt. “For merchants, that means repeat visits without a marketing team. For us, it means we can add merchants and scale support to everyone.”
Bilt is the membership for where you live and the Neighborhood Operating System (OS) powering everything around it.
Industry: Financial Services, Technology
Location: United States
Products: Agent Development Kit, AlloyDB for PostgreSQL, BigQuery, Gemini, Google Cloud
