Enterprises want more value from their data, but research from Salesforce shows how silos, gaps in strategy and low data trust continue to limit how far AI can scale.
Enterprises want more value from their data, but recent research from Salesforce shows how silos, gaps in strategy and low data trust continue to limit how far AI can truly scale.shows how much work is still ahead.
Many business leaders do feel confident about their data, with 63% stating that their company is data-driven; however, the same share of data leaders report struggling to connect data efforts to real business priorities. This is particularly pointed in the case of AI: the research found that 84% of data and analytics leaders believe their organization needs a complete reset of its data strategy for AI to succeed. The gap between being “data-driven” and actually being able to support new technologies is shaping how quickly companies can move forward with AI and broader IT modernization efforts. The consequences are real. Nearly nine out of 10 organizations using AI in production have seen inaccurate or misleading outputs. Technical leaders estimate that more than a quarter of their data cannot be trusted, and almost one-fifth of the data is either stuck in silos or is unusable. Even more challenging, many of the most valuable insights sit inside these disconnected or inconsistent datasets. All of this reinforces the insight that AI can scale only as far as the underlying data allows.Overcoming Data Fragmentation To Enable AI If we focus on the most promising advances made by IT vendors, we see that enterprise systems are shifting toward event-driven architectures and agentic AI, where intelligent systems work alongside humans to make decisions, trigger workflows, and take action in real time. That, at least, is the conceptual goal. But practically speaking, most companies in the real world still operate in a highly fragmented environment. Salesforce’s research found an average of 897 applications per enterprise, with only 29% of those apps integrated. With so much fragmentation, even the most advanced analytics packages or AI agents will fail to deliver sustained outcomes when the data feeding them remains incomplete, stale, or inconsistent. In line with these findings, ERP Today, and many do not have a clear data strategy or defined ownership around it. As Salesforce’s chief data officer, Michael Andrew, put it, “You can’t build automation or agents on sand. If the data isn’t trusted and harmonized, the system will always fall back to humans to sort out the noise.” He emphasized that as companies move toward increased automation, interoperability, and a single view of information, it has become non-negotiable to be “data fluent” and build connected systems.Andrew described this as a fundamental shift for data teams. Rather than focusing on backward-looking reporting, they now need to build data environments and practices that fuel real-time signals and flows that support the business. In the study, 81% of analytics leaders said their work now directly supports real-time decision-making, signaling a shift from periodic analysis to continual operational enablement. Based on my experience, this change also alters how success is measured; traditional indicators such as dashboard usage or the size of the data warehouse now matter much less than consistency, speed, and responsiveness. Andrew added that for Salesforce itself, this means maintaining a single source of truth, operating the infrastructure that supports it effectively, and ensuring that clean, contextual data is continuously available to both human and automated systems. He defines this idea as “data activation,” which implies getting the right data into the correct format so it can actually drive work. Salesforce estimates that 80% to 90% of enterprise data is unstructured; meanwhile, 70% of organizations say their most valuable insights are trapped in unstructured sources.As organizations make progress connecting core data sources, many are realizing that a large share of the insights they need will come from data that sits outside those systems. Unstructured documents, e-mails, and other sources — which make up 80% to 90% of all enterprise data, according to Salesforce — often explain what is actually happening in the business. Bringing that information into everyday decision-making is becoming a necessary step in making AI useful beyond limited use cases. However, unstructured data is often an area of weakness in companies’ data strategy. Enterprise analytics has historically focused on structured data from ERP, CRM, and financial systems because it was consistent, governed, and reliable for reporting. And that foundation still matters — very much so. The gap is that much of what explains performance, risk, and outcomes lives outside those records. Sources like service transcripts, meeting notes, chat conversations, and technician logs often contain the context behind delays, exceptions, and individual decisions, yet they have rarely been usable at scale. The organizations Salesforce identifies as leaders focus on connecting systems so that structured transactions and contextual data work together. On this topic, the Salesforce State of Data and Analytics report reflects what I hear regularly from customers and vendors. Enterprises know that valuable insight exists in unstructured data, as reflected by the report’s finding that 70% of data leaders believe that the best insights are locked away in those sources. But putting it to work is difficult. The data is messy, spread across multiple systems, and often lacks clear ownership or governance. Simply applying AI to raw documents or transcripts does not solve this. Without the appropriate metadata and context, using unstructured data can create as many problems as it solves. Where I see progress is in how enterprises are using unstructured data more deliberately. Instead of analyzing it after the fact, more organizations are connecting unstructured sources to structured records and workflows — in line with the best practices Salesforce identified. E-mails and notes are being linked to orders or customer cases. Maintenance logs and technician comments are being tied to specific assets . Customer conversations are being associated with granular service outcomes. When these kinds of connections are made, teams can better understandAI increases both opportunity and pressure. Agent-based and conversational AI systems are only as effective as the data they can access. Enterprises that properly organize and govern unstructured data can use AI to summarize prior interactions, surface relevant documents during workflows, and identify patterns that would otherwise be missed. Meanwhile, though, organizations that try to move too fast, without addressing data quality, ownership, and integration, often see inconsistent results and loss of trust. Although it isn’t addressed specifically in Salesforce’s report, this shift has direct implications for ERP. Based on my conversations across the market, vendors such as SAP, Oracle, Microsoft, QAD, Infor, Acumatica, Epicor, Sage, and IFS are moving toward architectures that support the thoughtful use of unstructured data. Shared data layers, zero-copy approaches, and embedded intelligence make it easier to bring structured transactions and unstructured context together without duplicating data or weakening governance. Enterprises that treat unstructured data as an input to ERP workflows can follow this approach to automate routine work, surface exceptions earlier, and support faster, more consistent decisions.Salesforce’s research makes it clear to me that trust, not tooling or ambition, is what separates AI progress from initiatives stalling. Many organizations have already invested heavily in data platforms and pipelines, yet fewer than half of data and analytics leaders believe their data is actually ready for AI. Respondents say they lack confidence in fundamentals like data quality, integration across systems, consistent governance, and reliable business context. The problem is not just access — as with the siloed applications and untapped unstructured data already discussed. It is institutional confidence. It’s a fact of the computer science driving AI that when data is incomplete, inconsistent, or poorly governed, any model built on top of it becomes less reliable. But it’s also a reality of human judgment: if your peoplethat your data quality and data handling aren’t up to snuff, they won’t trust the AI built on top of the data — and rightly so. This connection shows up clearly in outcomes. In the Salesforce report, 86% of leaders said AI results depend directly on how well data is governed and maintained, and I was not surprised to see the related finding that organizations with formal data quality processes are twice as likely to report strong ROI from AI. The organizations reporting the highest confidence take a disciplined approach, investing thoughtfully in data lineage, routine data quality checks, and clearly documented ownership. These practices do more than prevent errors; they make AI outputs understandable and explainable — and more trustworthy. When teams know where data came from and how it was validated, adoption improves because the system earns credibility over time.As generative and agent-based AI move closer to being a regular part of daily operations, the same foundations determine success. Salesforce’s report notes that nearly three-quarters of organizations plan to expand AI-driven operations within the next year, yet many still lack clarity around boundaries, oversight, and accountability. The strongest performers use AI to recommend actions or summarize context, while keeping humans responsible for final decisions. Clear guardrails matter. Companies with defined policies on what AI can execute autonomously versus what requires human approval report fewer issues with inaccurate or biased outputs. In practice, this means anchoring automation to verified data sources, audit trails, and governed workflows rather than experimental models or disconnected datasets. What emerges is a more disciplined form of automation. When data quality, process clarity, and ownership are in place, agent-based systems can improve efficiency without increasing operational risk. When those basics are missing, automation does not resolve underlying issues. It simply exposes existing gaps more quickly and at greater scale.The Salesforce report is helpful because it provides specific figures that show how pervasive enterprise data challenges are. And it is encouraging that Salesforce is so explicit about the scope of the problem, particularly around data fragmentation, quality, and governance. Many enterprises still operate hundreds of disconnected applications, and all my experience says that companies that struggle tend to treat integration as a one-time IT initiative rather than an organization-wide priority. In those environments, data unification, architectural simplification, and governance projects will require real, sustained work, not just check-the-box initiatives. Salesforce’s findings show that long-term gains come when data, business, and compliance teams operate with shared accountability — not as a one-off thing, but on an ongoing basis. In those environments, coordination becomes routine, and efficiency improves as a result of clear ownership going hand in hand with all the new technology. Moving toward agent-based automation forces companies — the successful ones, anyway — to address gaps head-on, especially around process consistency, risk controls, and change management. As enterprises move deeper into AI, the biggest constraint I see is not the AI itself. It is the data — and the operational behaviors behind the data. The organizations making real progress are not adding an excess of tools or overexpanding their AI pilot programs. They are fixing fragmentation, reducing duplication, assigning ownership, and making sure data is accurate and available when decisions need to be made. That foundation determines whether automation becomes something the business relies on or remains a disconnected experiment. Moor Insights & Strategy provides or has provided paid services to technology companies, like all tech industry research and analyst firms. These services include research, analysis, advising, consulting, benchmarking, acquisition matchmaking and video and speaking sponsorships. Of the companies mentioned in this article, Moor Insights & Strategy currently has a paid business relationship with Acumatica, Infor, Microsoft, Oracle, Salesforce and SAP.
Data Management Enterprise Data Enterprise AI Data Fragmentation Unstructured Data Agentic AI
United States Latest News, United States Headlines
Similar News:You can also read news stories similar to this one that we have collected from other news sources.
Trump Confronted ‘Weak’ Prosecutors About Delaying His Revenge TourThe president singled out a few of them for making Pam Bondi’s job difficult, a report says.
Read more »
Weak cold front bringing a few showers to northeastern ColoradoLisa provides the latest First Alert Weather forecast on Denver7 starting at 4:30am every weekday morning.
Read more »
Litecoin Price Forecast: LTC whale and derivatives activity rises amid weak price actionLitecoin (LTC) has seen a surge in whale activity and derivatives interest over the past three days, despite subdued prices.
Read more »
Weak Reentry Plans Help Competitors Win Female TalentSurvey data shows that insufficient reentry structures after maternity leave pushes experienced leaders out of organizations, creating a talent advantage for competitors.
Read more »
AI Data Centers Aim for the Stars: Space Becomes the New Frontier for Data StorageDriven by the soaring energy demands of artificial intelligence (AI), companies are increasingly turning their sights to space for data center deployment. SpaceX, Blue Origin, and LoneStar Data Holdings are leading the charge, aiming to leverage the unique advantages of space, including efficient cooling and enhanced data security, to revolutionize data storage and processing. This shift promises to alleviate strain on terrestrial power grids and address the growing vulnerabilities of Earth-based data centers.
Read more »
Why the Browns can’t stop ‘taking swings’ at quarterback despite weak 2026 NFL Draft classCleveland's quarterback search must remain relentless even as draft options thin out, with Mary Kay Cabot emphasizing the need to continuously pursue solutions through drafts, trades and free agency.
Read more »