Incident sheet · /data
- System
- Data & Analytics
- Filed
- Read
- 4 min · 795 words
- Severity
- Routine
Before You Sign: The Traceability Demands Data Leaders Should Put on Vendors
Column-level lineage, OpenLineage support and blast-radius isolation: what data leaders should demand from vendors before approving any new platform layer — and why agents make it harder.

Event log
- Gartner analysts say agent deployments are creating bigger data challenges than generative AI experiments did
- Kanerika's evaluation test: ask a vendor to pull a denied access request from 60 days ago with the policy version attached
- Banks under Basel Committee BCBS 239 must already trace critical data elements at column-level granularity
- Most lineage tools stop at the data platform edge, leaving BI reports, dashboards and downstream apps outside blast-radius view
- Gartner expects enterprises to need multiple tools — no single product traces the full AI lifecycle equally well
A denied access request from 60 days ago, policy version attached — that's the test data leaders should run on vendors before approving any new platform layer, according to Amit Chandak, chief analytics officer at consulting firm Kanerika. It's a concrete demand in a market where traceability failures get expensive fast.
Modern data estates pile on layers: meshes, fabrics, ingestion tools, warehouses, lakehouses, analytics and governance. Each layer adds dependencies and new points of failure. When a pipeline breaks or a governance rule is violated, and the data is scattered across platforms, isolating the signals needed for root-cause analysis gets hard.
"The proliferation of distributed environments makes the job almost infinitely more complex," said David Marco, president and executive advisor at EWSolutions, a data and AI governance consultancy.
What does agentic AI change?
Plenty. Generative AI let organizations limit exposure by restricting prompts or curbing user interactions. Agents don't sit still that conveniently. Lauren Kornutick, senior director analyst in Gartner's Analytics and AI team, has watched the shift among clients.
"When they were just experimenting with generative AI, they didn't have some of the data issues they're bumping into now," she said. "What we're seeing happen with agent deployments is the data challenges are becoming bigger."
Identity becomes a live question: how do you keep one user's agent from reading what only another user's agent may access? And if two employees share the same agent, how do you prevent leakage between their permissions?
Can you reconstruct a data element's life — after the fact?
That's Marco's baseline test. "Before approving a new data architecture or platform layer, I would want to know whether the organization can reconstruct the life of a critical data element after the fact — not simply determine where the data resides today," he said.
That reconstruction breaks into specific requirements. At ingestion, the architecture should capture:
- the authoritative source or system of record
- the data owner — a named owner, not a team alias
- provenance and the creation or ingestion point
- sensitivity classification and data quality status
"Reconstruct it afterwards and you are guessing from old Slack threads and whoever happens to remember," Chandak said. His prescription: make owner, source system and classification required fields at the first hop — fields that fail the pipeline when missing.
Insist on column-level, not table-level, lineage. "At 2 a.m., you need to know which field produced the wrong number," Chandak said. Banks under the Basel Committee on Banking Supervision's BCBS 239 risk-data framework already trace critical elements at that granularity.
Lineage also needs to cover what happened in transit: pipelines, integrations, APIs, ETL and ELT processes, replication, transformations — plus the business rules applied and their versions, Marco said. Vendor-internal lineage is usually fine; the breakdown happens at the seam between tools. Chandak's test: can the platform emit and consume OpenLineage events, the open-source standard for cross-environment lineage? If it only sees what it runs, every script, notebook and reverse-ETL hop is a visibility gap.
What do vendor demos leave out?
Blast radius. "Blast radius is the capability that gets the least demo time," Chandak said — and the one leaders lean on most during an incident. "Most lineage tools stop at the edge of the data platform, and your exposure keeps going into BI reports, embedded dashboards, downstream applications and the business processes sitting on top of them."
His evaluation tactic: invite vendor reps to list every downstream asset and process, including ones outside their product, and watch how much of the answer depends on manual work.
Unstructured data and the tooling picture
n Access control is easier for structured data than unstructured, Kornutick noted — unstructured data often isn't classified, so internal leakage can occur. "Guardrails run on metadata that classifies unstructured data," she said; Gartner clients are now exploring automated metadata tagging to close that gap.
The traceability toolkit is broad: metadata management platforms, data security posture management, AI gateways, privacy-enhancing technologies, AI governance platforms for agent deployment decisions and shadow-AI detection, observability tools, runtime AI application security, and — for regulated enterprises — cryptographic tracing that verifies how an agent outcome was produced.
Expect to buy several. "I don't think that you are going to be able to get one tool to trace every aspect of the AI lifecycle equally well," Kornutick said. Governance teams will weigh discovery, registry and policy management; data teams will prioritize data traceability and quality.
Her closing framing cuts buyers some slack: "Everyone feels like they're behind, but it's not because of them necessarily. It's because the AI is moving faster than we can figure out how to build the governance and security around it."
via techtarget.com (Source)
More from Olivia Hart
Show full bio
News editor covering money, funding and market shifts around enterprise information technology and the leaders who run at CIO Monitor.
207 articles
The newsroom digest
One email a day with the stories that matter. No noise.
Cross-references
- Five Data Management Trends for 2027: Agentic AI Drives Architecture Overhaul
- Agentic AI Is Forcing Analytics and Operations to Converge
- Microsoft, Google Move AI Agent Governance Into Enterprise IT Mainstream
- KubeCon 2026: Three Rules for Enterprises Running AI Agents
- Snowflake and Anthropic Expand Partnership for Governed Enterprise AI