How we're modernizing the foundations of Zendesk analytics to make business data more accessible, flexible, and ready for AI.
Zuye Zheng
Chief Data Architect at Zendesk
最後更新 2026年11月19日
The demands we place on data platforms today are vastly different from those of a decade ago. Businesses need to analyze more data, connect more systems, and increasingly make that information available to AI agents that can act on it.
Zendesk Explore, our reporting and analytics solution, launched in 2018 following the acquisition of BIME in 2015. It gave businesses a way to analyze their service data, track performance, and make more informed decisions.
These years also marked a major transformation in data architecture and technology. Snowflake and Databricks were emerging,Apache Pinot was open sourced in 2015, dbt was founded as Fishtown Analytics andClickhouse was open sourced in 2016. Apache Iceberg followed in 2018. New approaches to data lakes, lakehouses, and open table formats were making it faster and more efficient to process and analyze increasingly large and complex datasets.
Zendesk Explore was a product of its time. Built on traditional OLTP Postgres, its architecture was designed around technology optimized for fast, real-time updates and individual record lookups. That approach works well for powering applications, but becomes less efficient for analytics as data volumes, complexity, customizations, and AI use cases grow.
We needed a different foundation.
In November, Zendesk’s Agentic Analytics will become generally available and we’ll begin transitioning Zendesk customers from Explore to Agentic Analytics over the following year. Built on the foundations of HyperArc, Agentic Analytics introduces a new self-learning analytics experience that learns from how businesses use their data through memories.
Behind this experience is a complete re-architecture of the Zendesk data platform. Built from the ground up using modern data technologies, the new platform is designed to support more flexible analytics at scale and make business data accessible to both people and AI agents.
Built for all of your data
Zendesk is one of the few customer and employee service platforms that allows you to build your entire business on it, from custom fields and objects (including for external data) to custom automations and action flows.
If your data is important enough to model and bring into Zendesk, it's just as important to make it fully accessible in reporting and analytics. Understanding what happened is only part of the picture. You also need the data and context behind your operations to understand why.
Historically, Explore relied on predefined datasets designed to work across all Zendesk customers. Its underlying multi-tenant OLTP architecture made it difficult to support the unique data models and customizations of individual businesses. Some data couldn't be included at all, while other elements, such as custom fields, came with performance trade-offs.
Agentic Analytics changes this with the new Dataset Builder, which lets businesses model their data around how they actually operate. It’s the same tool we use to build Zendesk’s core datasets, so customers can customize existing datasets or create entirely new ones.
Dataset Builder also lays the foundation for managed ETL (Extract, Transform, Load) capabilities within Zendesk. Over time, we're expanding support for external data connectors and transformations, from familiar operations like joins and data preparation to more advanced capabilities such as AI-powered anomaly detection.
The new Dataset Builder lets businesses model their data around how they actually operate.
ETL allows us to prepare and combine data ahead of time, reducing the amount of computation required when someone runs a query. Combined with modern online analytical processing (OLAP) technologies and open table formats, this makes it possible to deliver faster, more flexible analytics across larger and more complex datasets.
For businesses, that means being able to bring more of their data together, analyze it in ways that reflect their operations, and get answers faster.
Built for flexibility and AI/ML scale
There's a familiar saying in engineering: of good, fast, and cheap, you can only have two. Data platforms face similar trade-offs.
The PIE theorem describes this tradeoff between pattern flexibility, infinite scale, and efficiency. Optimizing for one often means compromising on another. To support all three, the Zendesk data platform combines multiple analytical processing (OLAP) technologies, each optimized for different workloads, behind a common set of APIs and a unified analytics experience.
The Dataset Builder provides the pattern flexibility to model any data around your business needs. By performing computationally expensive operations, such as joins, ahead of time through ETL, the platform can query hundreds of millions of rows in seconds. The trade-off is data freshness, with datasets refreshed every hour.
Zendesk Real-time monitoring takes a different approach. Built on a complementary OLAP engine, it updates data within 5–10 seconds, giving businesses visibility into what's happening across their operations. In exchange, it supports a shorter historical lookback period and doesn't allow user-defined data models.
The advantage of combining these technologies is that customers don't have to navigate the underlying complexity. Through zero-copy analytics, which allows data to be queried without moving or duplicating it, Agentic Analytics provides a consistent experience across both real-time and historical datasets.
Whether you're exploring a real-time dataset or one you've customized, you can use the same report builder to create aggregations and expressions, then bring everything together in a single dashboard.
Making business data accessible to AI agents
The new Zendesk data platform is designed to make analytics available beyond traditional dashboards.
Real-time and historical datasets are unified through ArcQL, a single declarative query interface that makes data accessible through APIs and Model Context Protocol (MCP). Queries can be made through structured JSON requests or natural language, giving both people and AI agents ways to access the information they need.
With Zendesk Custom Agents, or your own AI agents connected through Zendesk Bridge, AI can query and analyze Zendesk data using deterministic APIs or natural language. These interactions can also draw on memories that capture how your business uses analytics, helping ground responses in relevant business context.
This creates new possibilities for how businesses use their data. Rather than relying on someone to build a report or consult a dashboard, AI agents can access analytical information directly as part of their workflows, using that information to investigate problems and inform decisions.
What it means for your business
As businesses generate more data and AI takes on more service work, the ability to access and analyze information quickly becomes increasingly important.
The new Zendesk data platform provides the foundation to meet those demands, making business data more accessible, flexible, and actionable for both people and AI agents.
The future of analytics depends on more than what we can learn from our data. It depends on how quickly and reliably we can put that knowledge to work, whether the next decision is made by a person or an AI agent.
Zuye Zheng
Chief Data Architect at Zendesk
Zuye leads Agentic Analytics and Zendesk’s new Data Platform. He’s shaping how people and AI work with data and take action on it, starting with the information in Zendesk. Before joining Zendesk, he co-founded HyperArc and led analytics at Salesforce and Tableau, where he helped launch Salesforce Reports and Einstein Analytics. When he’s not joining data, he likes joining wood—and hopes to spend more time doing it post-ASI.
Build business intelligence for the agentic era
AI performs best when it's grounded in trusted business context. Learn how Agentic Analytics helps AI investigate, reason, and act with confidence.