Zendesk is an AI-powered resolution platform helping organizations improve customer and employee service. In 2025 alone, the company served more than 20,000 AI customers, supported nearly 5 billion service resolutions, processed 2.4 trillion data requests, and powered 788 million AI interactions.
At that scale, reliable pipelines alone aren't enough. “To support this scale of data, we don't just need a reliable data pipeline,” explains Suhas Jangoan, Senior Data Engineer at Zendesk. “We need a platform that can build reusable assets, house shared definitions, and provide governance.” Its enterprise data platform brings together approximately 70 SaaS sources alongside product and event data, using S3, Snowflake, dbt Core, and Airflow to build data products the business can reuse.
As the organization grew, however, the same business concepts were increasingly being modeled differently across functional teams. Marketing, sales, and finance teams could each have valid but slightly different definitions of the same metric, creating dependencies between teams and making shared data harder to govern and reuse.
Zendesk addressed that challenge by moving commonly used logic into a foundational data layer with shared definitions, data contracts, testing, and operational guardrails. Then it went a step further: using AI skills and dbt MCP to automate repeatable parts of the engineering lifecycle while keeping business alignment and approvals in human hands.
The result is a workflow that has increased engineering velocity by 2.5x, accelerated testing and reviews, and created reusable context that can support Zendesk's next generation of conversational and proactive analytics.
Turning competing definitions into reusable enterprise data
Zendesk's "created pipeline" metric illustrates why shared definitions matter. Marketing uses pipeline data to understand campaign performance, sales uses it to track opportunities and forecast revenue, and finance needs to connect those opportunities to bookings and cash. Each team has legitimate requirements, but when they independently model the same metric, different versions can spread across the business.
Previously, marketing might build its own pipeline model, sales would consume and extend it, and finance would depend on both. That created circular dependencies without clear contracts, making it harder to change shared logic without affecting downstream teams or forcing stakeholders to reconcile inconsistent numbers.
Zendesk's enterprise data team identifies logic that has become valuable across functions and promotes it into the platform's foundational layer. Teams align on a shared definition, while the data team standardizes naming, models the data in dbt, maintains history, and adds testing and operational guarantees. Marketing, sales, and finance can then adapt the same governed asset for their individual use cases rather than independently rebuilding the underlying logic.
That process turns business alignment into reusable enterprise context. Definitions that once varied across teams become governed dbt models that people — and increasingly AI — can consistently understand and use.
Automating repeatable engineering work with AI
Promoting models into the foundational layer required engineers to gather requirements, trace lineage across repositories, develop and test dbt models, complete UAT, deploy changes, and update downstream dependencies. Zendesk saw an opportunity to automate the repeatable parts of that process without removing human judgment.
The team created reusable AI skills that divide the workflow between people and agents:
- Humans own business context and decisions, including aligning stakeholders on definitions, approving requirements, and signing off on the final output.
- AI handles repeatable engineering tasks, including lineage analysis, dbt development, code reviews, testing, deployment, and documentation.
- dbt MCP provides the technical context agents need to work effectively, exposing lineage, node details, manifests, catalogs, tests, and other dbt artifacts.
As Jangoan notes, this wasn't intended to be complete automation. "Working with the stakeholders on alignment of business definitions is still human," he says. Business sign-off also remains with people, while AI skills take on repeatable development and deployment tasks. That division allows Zendesk to accelerate engineering without automating the decisions that require business expertise.
Instead of every engineer repeatedly solving the same procedural problems, Zendesk can improve shared skills that are reused across model-promotion projects. The existing dbt environment becomes useful in a new way: models, metadata, lineage, and tests built for data engineering also provide structured context AI can use to understand the environment it's being asked to change.
Scaling agentic development without scaling inconsistency
As more engineers began building AI skills, Zendesk ran into what Staff Data Engineer Ayan Putatunda calls “skill sprawl.” In an enterprise, he explains, you don’t want every developer inventing their own skills; you have to think about how they’re created, shared, and governed.
“When you're working in an enterprise, you don't want every developer coming up with their own skills. You have to think about how those skills are created, shared, and governed,” Putatunda explained.
Zendesk addressed that challenge by creating its Data Platform AI Skills Library, giving developers across geographies and time zones a central place to contribute, improve, and reuse skills and workflows. Those capabilities can be distributed directly into development environments through plugins, making a workflow created by one engineer available to others without rebuilding it from scratch.
The company applies the same enterprise discipline to AI artifacts that it does to software development. Shared skills go through centralized review, governance, and evaluation before broader distribution, with humans reviewing results against sandbox data. That allows Zendesk to scale reusable AI workflows while maintaining consistent standards across teams.
More than doubling engineering velocity
Zendesk is already seeing measurable improvements from its AI-assisted development lifecycle:
- Increased data engineering velocity by 2.5x, based on sprint performance measured over 6 months.
- Accelerated UAT cycles with automated testing, helping teams move changes through development faster.
- Standardized code reviews across developers with a shared review skill, improving consistency without requiring every engineer to create their own process.
- Scaled reusable AI workflows across multiple teams, allowing improvements made by one developer to benefit others.
- Automated repeatable development and deployment tasks while keeping business alignment, stakeholder sign-off, and final approvals human-led.
The 2.5x improvement reflects more than faster code generation. By automating repeatable development tasks, standardizing reviews, and accelerating UAT, Zendesk can move shared business logic into governed enterprise models faster without requiring engineers to recreate the same processes for every project. As Jangoan explains, the improvement was measured through 6 months of sprint performance, giving the team a concrete way to evaluate the productivity gains from AI-assisted development.
Giving AI the context to understand the business
Zendesk's experience shows how the data platform's role is expanding. The same governed dbt models, metadata, lineage, and tests that help engineers maintain reliable data products can now provide context for AI agents working within the development lifecycle.
That foundation is already helping Zendesk automate repeatable engineering tasks and improve development velocity. Looking ahead, the company is investing in metric governance and centralized context management to make conversational intelligence a first-class capability of its data platform. Its longer-term vision includes giving employees trusted answers through natural language and proactively surfacing meaningful changes in business metrics.
For other data teams, Zendesk's experience offers a practical starting point: align on shared business definitions, establish governance around reusable data products, and make technical context accessible to AI. Then automate the repeatable work while keeping business decisions and approvals in human hands. That approach can help teams deliver trusted data faster without sacrificing the oversight required at enterprise scale.
