AI-Native, Self-Service Dashboards on Live SAP Data

Executive Summary

Getting a straight answer out of SAP usually means going through a queue. Operational data (CDS views) and planning data (SAP IBP — Integrated Business Planning) sit behind specialist skills, so a new report or dashboard typically waits on a BI or SAP team, and often ends up as yet another extract copied into a separate analytics tool — a second, ungoverned copy of sensitive business data.

We built a dashboard capability inside the client’s AI agent platform that removes both problems. A non-technical user — or an AI agent acting on their behalf — can create a live SAP dashboard end to end through a guided wizard: pick a source, browse the catalog, build a query, bind it to a widget. The whole dashboard is a single portable definition that agents can read and write at the same granularity as the human. Crucially, no SAP data is stored at rest — widgets fetch live, in a credentialed request context, every time they open.

The Challenge

Self-service analytics on SAP is hard for reasons that compound:

• Specialist bottleneck: Building a dashboard against CDS views or IBP planning data requires knowing the catalog, the field semantics, and the query mechanics — skills that live with a small BI/SAP team, so business users wait in a queue for every change.
• A second copy of the data: The common workaround — extract SAP data into an external BI tool — creates a governed system of record and an ungoverned copy, with all the residency, freshness, and access-control problems that follow.
• Read-only by default: Even good SAP dashboards only show numbers. Acting on them — adjusting a demand plan, for example — means leaving the dashboard and re-entering the data by hand in SAP, with no safety rail between “considering a change” and “committing it.”
• AI that can’t reach the data: Teams increasingly want to ask for a dashboard in natural language, but an assistant with no governed, live path into SAP can only guess.

The Solution

We built the dashboard as a first-class part of the client’s agent platform, reusing the platform’s existing SAP connectivity (Model Context Protocol servers) rather than bolting on a new integration.

Key Components

1. Guided, no-code build flow: A wizard walks the user from “New Dashboard” to a working widget — the project’s SAP source is auto-enabled, so there is no manual connection step. The user browses the CDS catalog (or the project’s IBP planning intents), introspects a view’s fields, and builds a query — fields, filters, sort, row cap, and aggregation (COUNT/SUM/MIN/MAX/AVG) — then binds it to a widget. The query lives inside the widget.

2. Live fetch, nothing at rest: CDS and IBP widgets execute their query against SAP on open and refresh, inside a credentialed request context. No SAP data is persisted — viewers without a valid SAP session see an explicit “requires connection” state rather than stale cached numbers. Credentials are resolved server-side per request and are never exposed to the frontend.

3. Portable, agent-authorable definition: The entire dashboard — widgets, their embedded queries, layout, and variables — serializes to a single secret-free JSON bundle that can be exported and imported. Dashboard-enabled AI agents can read and write that bundle at the same granularity as the human wizard, so a user can ask for a dashboard and refine it conversationally.

4. Two live SAP sources: SAP CDS for operational data and SAP IBP for Integrated Business Planning (demand/supply). Both flow through the same widget, transform, and inspection pipeline, so the experience is identical regardless of source.

5. Safe write-back to planning: IBP widgets support an “adjust key figures” decision flow modelled on a planning cockpit — edit the records, run a dry-run to preview the impact, then explicitly confirm before the change is committed back to SAP with a transaction id. Write access is gated by a strict owner-only check and a project-level write-intent allowlist, enforced again server-side (defense in depth).

6. Shared variables and filters: A dashboard variable bar re-parameterises many widgets at once (${var} interpolation in queries), so a single “Region” or “Product Family” control drives the whole board across both CDS and IBP widgets.

The Outcomes

The outcomes below are described qualitatively — no performance figures are claimed for this engagement. See the note that follows.

• Self-service without the specialist queue: Business users build live SAP dashboards through the wizard without waiting on a dedicated BI/SAP team.
• No second copy of SAP data: Live-fetch-only means there is no data-at-rest extract to govern, secure, or keep fresh — the SAP system remains the single source of truth.
• Dashboards you can ask for: AI agents author and adjust dashboards conversationally, using the same governed, live path into SAP that the human UI uses.
• From insight to action, safely: Planners can adjust IBP key figures directly from a widget, with a dry-run impact preview and an explicit confirm step before anything is committed — and owner-only plus allowlist controls on every write.
• Portable and versionable: A dashboard is a single, secret-free JSON bundle that can be exported, imported, reviewed, and round-tripped between environments.

Scope & Honesty

• What this is: A production-grade capability we delivered within a client’s AI agent platform, built on the platform’s existing, security-hardened SAP MCP connectivity. It is described here as a delivered capability, not as a measured rollout.
• No metrics claimed: We have deliberately not attached percentages or time savings to this study. Any such figures would depend on the client’s SAP configuration, data quality, and usage, and none have been independently measured for publication.
• Governance posture: Data sources and dashboards are project-scoped; credentials are resolved server-side and never sent to the browser; write-back is owner-only and constrained to an explicit per-project allowlist. These are design properties of the system, verifiable on review.
• Anonymised: Client and environment specifics are omitted by default; architecture and controls can be walked through in more detail under NDA.

Technology Stack

• Platform: Python (FastAPI) backend, React frontend, PostgreSQL — the client’s AI agent platform
• SAP Connectivity: Model Context Protocol (MCP) servers for SAP CDS and SAP IBP, called live over HTTP in a credentialed request context
• Data Sources: SAP CDS views (operational) and SAP IBP planning intents (Integrated Business Planning)
• Dashboard Definition: Single versioned, secret-free JSON bundle (widgets, embedded queries, layout, variables); export/import and agent read/write
• AI Authoring: Platform agents with dashboard create/modify tooling, operating at the same granularity as the human wizard
• Controls: Server-side credential resolution, project-scoped sources, owner-only write checks with a per-project write-intent allowlist enforced server-side