Grafy News
Try Grafy free

Enterprise AI Shifts from Pilot Projects to Scalable Platforms

Enterprises are stuck in AI pilots; experts urge a unified data layer, clear governance, and a platform mindset to scale AI agents across workflows.

Illustration for the news story: Enterprise AI Shifts from Pilot Projects to Scalable PlatformsSource: MIT Technology Review — AI

Enterprises are still wrestling with AI pilots, but leaders say the next step is treating AI as an enterprise platform that can be woven into everyday workflows. The shift matters because without a solid data foundation, clear governance, and scalable use‑case evaluation, AI agents risk becoming siloed experiments rather than reliable digital coworkers.

Why are so many AI projects stuck at the pilot stage?

Both speakers cite surveys that show a high failure rate for AI pilots. A BCG report finds 74% of organizations remain in pilot mode with no clear plan to scale. An Anthropic survey adds that 42% struggle to get access to the data and quality needed for enterprise‑wide AI. Infosys and HFS research of 2,000 enterprises reveals only 14% have successfully scaled AI, leaving 86% in early exploration or fragmented pilots. The common thread is a fragmented approach: isolated proof‑of‑concepts built on inconsistent data foundations.

What does treating AI as an enterprise platform entail?

Thorsten Leiduck (SVP, SAP Business AI Platform Ecosystem) argues that AI must be anchored to a "connected, semantically‑rich" data layer that mirrors the business. This means moving from ad‑hoc projects to a unified platform where AI services are embedded across applications. SAP’s Autonomous Enterprise Framework is presented as a concrete implementation that links data, processes, and AI models under common governance.

How should companies evaluate which AI use cases to pursue?

Satinder Pal Singh (Associate VP, SAP BAIP & Infosys) recommends a three‑lens framework:

  1. Impact – Does the agent move meaningful business metrics (e.g., cycle‑time reduction, cost per transaction, revenue uplift)?
  2. Feasibility – Are the required data, talent, and model capabilities available?
  3. Scalability – Can the solution be replicated across lines of business or geographies?

He warns that a pilot that cannot scale is merely an experiment. Starting with a small set of high‑value use cases and defining success metrics up front helps avoid wasted effort.

What governance structures are needed for AI agents?

As AI moves into mission‑critical workflows such as supply‑chain planning or financial close, the tolerance for error drops dramatically. Singh emphasizes the need for clear policies, access controls, monitoring, and oversight to keep agents trustworthy. Without governance, organizations create a “risk backlog” that can force disruptive audits later. Governance should be baked into the AI lifecycle—from model selection to deployment and ongoing monitoring.

Why does a strong data foundation matter?

Leiduck stresses that AI is "absolutely dependent on the data foundation." A semantically rich, enterprise‑wide data layer enables consistent model performance and reduces duplication. Customer‑specific knowledge graphs are highlighted as future‑critical assets, allowing agents to retrieve contextual information quickly and make better decisions. In practice, this means consolidating data silos, standardizing schemas, and enriching data with business semantics before training or deploying agents.

How to explore these concepts with Grafy

Grafy’s Chat app lets you experiment with multiple AI models side‑by‑side, helping you compare how different agents respond to the same business prompt. You can also use the Reader app to ingest the MIT Technology Review transcript (or any PDF) and ask specific questions about data foundations, governance, or use‑case evaluation. For teams that want to map out an AI workflow, the Graph app can visualise the branching steps from data ingestion to agent deployment, making it easier to spot gaps in governance or data quality.


Source: https://www.technologyreview.com/2026/11/16/1146070/building-ai-that-can-work-across-the-enterprise/

Frequently asked questions

What is the most common reason AI pilots fail to scale?

The primary reason is a fragmented data foundation and lack of a unified platform, which leads to duplication, inconsistent performance, and unclear paths to scale.

How can organizations decide which AI use cases are worth investing in?

Apply the impact‑feasibility‑scalability framework: choose cases that move clear business metrics, have the necessary data and talent, and can be replicated across the enterprise.

Why is governance essential when deploying AI agents in critical workflows?

Governance provides policies, access controls, and monitoring that ensure agents remain trustworthy, compliant, and aligned with business requirements, preventing risk backlogs and costly audits.


Source: MIT Technology Review — AI — Building AI that can work across the enterprise