Concept card

BARC — Lessons from the Leading Edge: Successful Delivery of AI/GenAI

As of 2026-09-27

Edited by Cédric Mary, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

What is BARC?

Use-case selection is the single highest-leverage decision in AI delivery — five specific criteria separate a high-probability-of-success use case from the predictable failure modes that keep sinking enterprise AI projects.

BARC's "Lessons from the Leading Edge" distills a pattern that every SAP analytics consultant now has to internalise: enterprise AI and generative AI initiatives fail far more often than they succeed, and the failure modes are neither random nor mysterious — they recur across industries, vendors, and use cases in ways that are avoidable if a delivery team knows what to check for before committing budget.

Why AI delivery fails differently from conventional software

Conventional enterprise software projects fail from familiar causes — scope creep, weak sponsorship, poor requirements. AI systems add a distinct failure surface on top of those: they degrade silently as the data distribution they were built on drifts away from the data they encounter in production, they produce confidently wrong answers (hallucination) under conditions that are difficult to anticipate at design time, and their realised value depends on a human-adoption behaviour — will the domain expert actually trust and act on the output — that project plans routinely overestimate. The gap between an impressive prototype demo and a production system that keeps delivering value six months later is wider in AI than in almost any other software category, which is exactly why understanding what separates the two is among the most commercially valuable judgment calls a consultant can offer a client.

Why it matters

The five success criteria are concrete, not generic: bounded task with clear success criteria, sufficient historical data volume, an available domain expert evaluator, detectable/recoverable errors, and augmenting rather than replacing human judgment.
The named failure modes are equally specific: fully autonomous high-stakes decisions with no human review, sparse/unstructured/unknown-quality data, use cases chosen for technical interest rather than business impact, and success defined as "the model is accurate" instead of "the business outcome improved."
For SAP analytics specifically, strong starting use cases are named: automated narrative generation from structured financial/operational reports, intelligent search over knowledge bases, anomaly detection, and natural-language queries over governed semantic layers like Joule-on-Datasphere.

Key points

BARC research focused on actual production AI/GenAI delivery (vs. pilots).
Leading-edge sample = self-selected high-maturity organisations.
Covers both classical AI and GenAI delivery patterns.
Maps onto the Joule + Knowledge Graph delivery wave SAP customers are entering.
Specific lessons paywalled — use card to anchor the conversation, not to quote percentages.
The underlying study surveyed 421 organisations worldwide; data quality is now the top-cited barrier to AI success (44%), up from a secondary concern in the prior year's edition (BARC, 2026).
More than half of surveyed organisations reported software costs exceeding initial expectations, driven largely by validation, quality-control and training costs rather than the model licence itself (BARC, 2026).
Delivery-partner performance diverges sharply in the survey: internal IT is the default delivery driver but scores lowest on project satisfaction, while regional/global consulting firms score highest for delivery effectiveness — a data point worth citing in a build-vs-partner conversation (BARC, 2026).

Terms used on this page

Leading-edge sample: BARC's self-selected subsample of organisations that have shipped AI/GenAI in production — distinct from pilots. Used to extract delivery lessons that the broader market hasn't yet encountered.
Delivery model: The operating model wrapping an AI/GenAI product in production: change management, trust-building, data-quality controls, governance process, and model-lifecycle management. Separate from the ML/LLM technology itself.
GenAI: Generative AI — AI systems that produce text, code, or structured output from prompts. In the SAP context: Joule, Joule for Consultants, and BDC's embedded AI copilots.
AI leadership maturity: BARC's term for organisations demonstrating stronger practices across leadership dimensions (sponsorship, governance, change management); the survey found 53% of these 'AI leaders' run more than five AI projects in production, versus 28% of other organisations.
Data readiness: The cleansing, structuring and governance work required before data can reliably ground or train an AI system; BARC's research names underestimated data readiness a second-order failure cause that typically surfaces months into delivery rather than at kickoff.

Sources

Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks · the facts worth quoting.

Open in the app →