Why IBM thinks agentic AI enterprise adoption starts with fixing the data problem

• 
  
  IBM’s watsonx Orchestrate positions as the coordination layer for agentic AI enterprise deployments.
  
  


• 
  
  Biggest obstacle to agentic AI is data.
  
  

Most enterprises have invested heavily in AI. The decisions they’re making are still slow. That gap, between AI investment and actual decision velocity, is what IBM’s Institute for Business Value

identified

as one of the defining pressures heading into 2026: AI-enabled workloads are expected to rise from just 3% in 2024 to 25% by 2026, an eight-fold increase that signals how fast the change to agentic systems is moving.

IBM’s answer to that pressure is watsonx Orchestrate, its platform for building and coordinating networks of AI agents in enterprise workflows. The pitch is straightforward: instead of querying a dashboard or waiting for a report, an AI agent performs the retrieval, makes the evaluation, and hands it off to a human only when necessary.

The company will be showcasing its agentic AI approach at the upcoming

AI & Big Data Expo at TechEx North America

in San Jose this May.

The data layer nobody wants to fix

Before the agents can work, the data has to be ready. According to IBM’s own research, most organisations are not yet there. A global

study

from IBM’s Institute for Business Value, surveying 1,700 Chief Data Officers in 27 countries, found that while 81% say their organisation’s data strategy is now integrated with technology road maps – up from 52% in 2023 – only 26% are confident their data can support new AI-enabled revenue streams.

IBM says the barrier to effective agentic AI is not inference costs or the model itself. The problem is data. Organisations need trusted, company-specific data for agents to create real value, and that data is overwhelmingly unstructured, sitting inside emails, documents and internal videos. IBM

estimates

that 90% of enterprise data generated in 2022 was unstructured, with only 1% accounted for in LLMs.

This is what IBM’s watsonx.data platform is designed to address, giving agents access to both structured and unstructured sources without requiring organisations to rebuild their data stack from scratch.

What the orchestration layers do

Watsonx Orchestrate

integrates

with more than 80 enterprise platforms, including Adobe, AWS, Microsoft, Oracle, Salesforce, SAP, ServiceNow, and Workday, allowing AI agents to act in systems through a single interface. The multi-agent architecture lets agents from different vendors share context and hand off tasks without a human being as the relay.

IBM chief commercial officer Rob Thomas describes the challenge: “Our view is there was a big gap in the market for how you integrate agents. It’s not about us providing agents. It’s also letting a company that’s going be looking at a future where there’s hundreds, thousands, maybe millions of agents running at once.”

What IBM is betting on is that orchestration, not model performance, becomes the decisive ability. Analyst Mark Beccue of Enterprise Strategy Group has noted that IBM’s experience in enterprise systems integration gives it an advantage in this space: “IBM is better at holding hands. Their whole job is system integration.”

IBM as its own test case

IBM has leaned into being what it calls “client zero” – deploying its own agentic AI stack internally before offering it to customers. The results it has published are specific enough to be useful as a benchmark.

The company’s IT support agent, AskIT,

resolved

86% of queries with AI agents, reduced calls and chats by 74% since its 2023 launch, and delivered an initial US$18 million cost reduction with ongoing annual cost avoidance. Across the business, IBM says AI and automation have

helped

it unlock US$4.5 billion in savings and an estimated 3.9 million hours freed from manual tasks in 2024 alone.

In procurement, the company’s invoice review agents achieved over 95% first-pass accuracy in 18 of 19 invoice formats in testing, automating processes that previously required manual review at scale. Its sales assistant, AskSales, answered over 250,000 seller questions in 2025.

The agentic AI enterprise gap

The deployment reality in the broader market is less tidy. While 79% of enterprises have adopted AI agents in some form,

only 11%

are running them in production; a 68-percentage-point gap that represents the largest deployment backlog in enterprise technology history, according to research tracking the space.

IBM’s own CEO

study

, surveying 2,000 CEOs globally, found that 61% are actively adopting AI agents and preparing to implement them at scale, and 72% view their organisation’s proprietary data as the important to unlocking generative AI value. Yet half of the respondents acknowledged that the pace of recent investments has left their organisation with data environments that cannot yet support the AI ambitions in place.

The companies closing that gap fastest – those that have moved agents into production not pilots – are reporting returns of 171% on average, according to market benchmarks. Those still in pilot stages are watching the competitive distance widen.

IBM’s positioning is that the path from pilot to production runs through a governed, orchestrated data and agent layer, not through any single model. Whether that framing resonates with enterprise buyers at scale is the question its TechEx appearance this May will help answer.

IBM is exhibiting and speaking at the AI & Big Data Expo, part of TechEx North America, at the San Jose McEnery Convention Centre, 18 – 19 May 2026.

Want to experience the full spectrum of enterprise technology innovation?

Join

TechEx

in Amsterdam, California, and London. Covering AI, Big Data, Cyber Security, IoT, Digital Transformation, Intelligent Automation, Edge Computing, and Data Centres, TechEx brings together global leaders to share real-world use cases and in-depth insights. Click

here

for more information.

TechHQ is powered by

TechForge Media

. Explore other upcoming enterprise technology events and webinars

here

.