AI layers are landing on legacy data instead of replacing it

Keep the ERP, add the intelligence

Oracle says it has launched E-Business Suite Intelligence to provide prebuilt analytics and AI-assisted experiences for E-Business Suite customers, combining packaged E-Business Suite data, business semantics, analytics and AI-assisted experiences. The company says the solution aims to reduce the manual effort of building data pipelines and modeling application data by providing packaged business semantics out of the box. m:…1455

The solution supports Finance, Procurement and Projects business areas, Oracle says, with finance intelligence helping teams understand financial performance and working capital, procurement intelligence helping teams improve visibility into spend and supplier performance, and projects intelligence helping teams improve visibility into cost, billing and funding. Users can continue running E-Business Suite while using the solution, the company says, and for customers using Intelligence, the solution can bring E-Business Suite and Fusion Applications insights together. m:…1455

Finance data, joined without rebuilding it

Oracle has also introduced EPM Integration in Fusion Data Intelligence, which the company says is designed to reduce data preparation work for finance teams. It provides a managed path for bringing data and dimensions from supported Oracle Fusion Cloud EPM applications into Fusion Data Intelligence, supports recurring scheduling and incremental data options, and supports plan-versus-actual analysis and AI-enabled experiences. m:…6695

Oracle says Fusion Cloud ERP actual financial activity is available for analysis in Fusion Data Intelligence, that customers define the semantic model and analytics that fit their requirements in EPM Integration, and that finance teams can move from high-level variance toward supporting detail using the integration. This suggests the value here is not a new ledger, but a shorter path from a variance a finance team already sees to the detail behind it. m:…6695

Productivity data, governed for agents

Databricks says it has expanded Lakeflow Connect to include native ingestion connectors for productivity tools, providing fully managed native connectors across SaaS applications, databases and file sources. Data ingested via Lakeflow Connect is stored as governed, managed tables in Unity Catalog, the company says, with connectors for Jira, Monday.com, Smartsheet, Aha! and GitHub, and for Confluence, Notion, Glean, SharePoint and Google Drive. m:…2875

SharePoint and Google Drive connectors land unstructured files as rows with raw bytes in a content column, Databricks says, and Databricks uses ai_parse_document to turn raw bytes from document files into text for indexing. The company cites Toyota Finance Australia using Lakeflow Connect to ingest Jira data into Databricks, and VaxCare using the Confluence connector to feed its internal knowledge base into Databricks for AI agent grounding. m:…2875

Databricks says Orizon synchronized over one million Jira tickets from 168 projects into Databricks, that the OpenAI connector allows attribution of AI spend by team and tracking of adoption, and that Lakeflow Connect includes connectors for OpenAI and Anthropic to govern AI usage. The pattern is consistent with the Oracle launches: the AI layer is useful only if the scattered data underneath it arrives governed, searchable and attributable, without a custom pipeline for each source. m:…2875

The same logic inside an AI builder

Writer, an enterprise AI platform, shows the same trade-off from the builder side. The company says its LLM tracing pipeline generates up to 600 rows per minute at peak, its primary traces table receives roughly 1.3 million rows per day, and it routes every customer request through LLM inference providers. Standard Snowflake tables experienced 45 percent failure rates for insertions due to lock contention, the company says, with P50 insertion latency on standard tables reaching 2,362 milliseconds. m:…7424

After switching to Snowflake Hybrid Tables, Writer says P50 insertion latency dropped from 2,362 milliseconds to 154 milliseconds, write failures dropped from 45 percent to near zero, and it downsized the warehouse powering its ingest pipeline from medium to extra small. Hybrid Tables use row-oriented storage with row-level locking, the company says, and Writer uses a Snowflake Task to incrementally copy rows from the hybrid table to a standard table, using Hybrid Tables for high-velocity ingestion and standard tables for downstream analytics. m:…7424

Writer says it evaluated writing tracing data to Postgres as an alternative to Snowflake, and is considering using Hybrid Tables for high-volume API logging data. The implication for tech executives is practical: when AI workloads stress an existing data estate, the fix may be a different table type or a governed connector on the estate already in place, rather than a migration or a new external store. m:…7424