by Meredith Delaware
Despite significant investment in AI, only a quarter of government organizations have successfully integrated it across their operations, and nearly half say poor data infrastructure is holding them back. The promise of AI is real, but the reality is simple: “junk in, junk out” has never been more accurate. If organizations want AI to deliver on its potential, they must start by getting their data house in order because no algorithm can fix what’s problematic at the source.
The stakes for AI implementation are high as technology is transforming national security, both on and off the battlefield. Artificial intelligence (AI)-enabled tools promise to speed decision-making and increase situational awareness, adding critical efficiency to combat effectiveness. From safeguarding the military and its industrial base against espionage to adding unprecedented precision to unmanned aircrafts’ firing solutions, AI integration is essential to national security and success in combat.
Achieving data readiness for AI is complex: organizations face a multitude of challenges in bringing data together from different formats and systems, and coordinating with cross-organization data owners. Government organizations must take the time to assess and enhance their data readiness if their AI investments are going to succeed.
In this article, we will:
● Explain why AI’s value in government depends upon a focus on data readiness.
● Identify signs that an agency’s data is not ready for AI.
● Present the business case for investing time and resources in delivering data readiness today.
● Help governmental organizations begin to understand their current data readiness and develop a roadmap for moving forward.
Data Readiness, Defined
Data readiness describes the quality of data being collected and prepared for use, with the goal of accurate, complete, well-structured, and accessible data. Poor data readiness can impact the reliability and accuracy of AI systems. AI models rely on quality data to learn patterns and make reliable predictions. Models can expose poor-quality data by delivering inaccurate results and data readiness is exponentially more important in complex, multi-modal AI systems where inconsistencies compound across domains and data types.
One of the primary advantages of implementing AI is its ability to deliver more accurate data processing and analytics outputs. Achieving this advantage is why data readiness is a foundational requirement for AI systems. It is the bridge between collecting data and reaching mission outcomes. It is a step that cannot be skipped if an organization wants to reach the intended return on its AI investment.
Key Elements of Data Readiness
To deliver data readiness, organizations need to align data assets with operational and mission requirements. This may not be the most exciting element of implementing AI, but its importance can’t be overstated. Organizations need to spend resources and attention on these four areas:
-
Data Processing: This essential first step builds a data ecosystem of complete, accurate, and meaningful data. For an AI tool to make sense of data, organizations must transform unstructured multimodal inputs into structured, analyzable formats, which might include text, images, video, or specialized sensor data. This raw data must first be collected, cleaned, and formatted. This step ensures later accuracy by removing duplicates, addressing errors, and managing missing values.
-
Data Categorization: For AI systems to function optimally, data must be organized usefully. This is done by grouping data into specific categories based on shared attributes. This step enables AI systems to identify patterns and key features when analyzing data. In this way, appropriate categorization unlocks the ability of AI tools to deliver accurate predictions.
-
Data Integration: Effective data integration enables organizations to bring disparate data together so that they can gain deeper insight from existing assets. Organizations of all sizes and missions use, store, and process data, with varying levels of communication between tools and platforms. Organizations need to consolidate data from various sources, including Excel spreadsheets, CRM systems, sensors, specialized databases, and other specific-use systems, among others. They need to develop interfaces that orchestrate data across sources and flows to balance fusing insights with security.
-
Data Governance: Data governance establishes the infrastructure for and policies dictating how data is entered, managed, stored, and used. These rules ensure data quality and security throughout the data lifecycle, maintaining compliance with applicable regulations. Organizations that haven’t determined their AI posture and policies prior to adopting tools risk introducing bias and exposing vulnerabilities.
Preparing People to Deliver Data Readiness
In addition to attending to the mechanics of improving data for AI implementation, organizations must also acknowledge and address the intrinsic role of the humans who interact with that data. Analysis from S&P Global Market Intelligence reports that organizations have scrapped 46% of AI proof-of-concept projects before they reach production. Forty-two percent of companies have abandoned most AI projects. The analysis found that organizations with high project failure rates tend to encounter resistance from employees.
In ways large and small, implementing AI requires team members to change their patterns of working. For example, changes in data categorization schemas may require people to label data differently using methodologies, tools, or validation tasks in systems or applications new to them. Data owners will need to collaborate with other departments and organizations to integrate data and processes. Alternatively, teams may require training or new types of expertise to deliver data readiness for AI. Considering how changes to data policies will impact people is essential at this foundational stage. Without attention to how people are impacted by data readiness initiatives, and applying accompanying user adoption and change management techniques, professionals will return to old patterns of data interaction, risking realizing the benefits of data and AI investments, and worse, risking mission failure.
Securing buy-in from the humans involved in the data lifecycle is an often-overlooked and universally underfunded aspect of data readiness. It’s easy to focus on the processes, tools, and models of AI, but data owner, engineer, analyst, and user engagement is key in advancing these mission-critical initiatives.
Why Government Data Isn't AI-Ready: A Look at the Key Friction Points
Despite the importance of data readiness, many government organizations are struggling to establish the processes and infrastructure required to pull quality data across systems into an effective AI tool. As noted in a 2025 report by the U.S. General Accountability Office, government agencies ready to expand their use of generative AI often face challenges accessing sufficient technical resources and budget. Agencies, including the DoD, reported difficulty hiring and developing an AI workforce. Yet these resources will be essential to ensure data readiness.
This is a common issue among organizations seeking to implement AI. In a recent report by MIT Technology Review Insights, “Building a high-performance data and AI organization,” only 12% of surveyed organizations considered themselves “high achievers” in data readiness, a 1% decrease from the 2021 survey, despite significant advances in data management solutions.
Due to the frenetic push for AI adoption, data readiness is non-negotiable for all organizations, regardless of mission type or data maturity. Investing in AI technology without first achieving data and process readiness leads to wasted resources. It will automate existing inefficiencies and lead to inaccurate output. To break this pattern and prioritize data readiness actions, organizations must first identify roadblocks and friction points. Some of these common obstacles are listed below.
Missing and Unused Data
AI models require complete datasets to detect and predict patterns. Missing values can cause biased results, faulty recommendations, and hallucinations. It’s not uncommon for data to be locked in siloed systems or non-operable formats where AI pipelines can’t retrieve them. Users of these systems may be unaware of this gap in information. It’s hard to prove a negative.
The opposite case can also prove problematic. Many organizations are prone to collecting data in the belief that “someday” it will prove valuable. Unlocking the value from this data may depend on cleaning this workflow and appropriately categorizing data. But when data is collected without purpose, there’s rarely an incentive to begin the time-intensive process of cleaning and evaluating it for mission relevance.
To address these challenges, organizations should prioritize data based on its strategic relevance, then invest in cleaning and categorizing it in alignment with their mission priorities. Metadata tagging and indexing can accelerate the transformation of data into usable assets. When data is properly labeled and indexed, AI systems can retrieve and interpret it more effectively. Data needs to be treated as a living asset with continuous curation, governance, and alignment with mission needs.
Resource Constraints
Data readiness demands time and investment, but teams are rarely handed resources to address the labor-intensive and organizationally challenging process of ensuring data readiness. It’s up to leaders to recognize that launching AI without data readiness merely accelerates dysfunction. They must help teams communicate the business case for securing necessary resources more effectively.
Teams can strengthen their case by assigning tiger teams, including junior or new hire teams, to experiment with small batches or sample datasets to demonstrate improved outcomes. Experimentation in low-risk, low-cost areas can inform how and to what scale organizations will need to clean, label, and process data to provide suitable AI input. Many organizations are surprised to find that even a brief amount of time spent focused on this work can uncover powerful insights.
Lack of Data Expertise and Guidance
Every area of government is under pressure to explore potential AI applications, but many organizations are still building the expertise and creating the policies required to move forward. Even when teams are provided AI tools, such as the generative AI tools in enterprise commercial products like AWS GovCloud or Microsoft 365 Copilot GCC, there can be a prolonged learning curve to realizing value. Access to AI tools won’t drive success if the organization doesn’t know how to create the processing and governance structures required for effective use. Organizations must invest in appropriate training and talent management to close these skills gaps.
In addition, organizations must recognize that any change in how data is processed, categorized, and integrated will demand a change in how people operate. Before tasking users with new ways of working with data, consider how to effectively manage this change for lasting success.
Conflicts Around Data Ownership
AI solutions demand that data be categorized and structured in a consistent format. If the team assigned to an AI initiative lacks ownership of all required data or does not have access to essential systems, it may be difficult to advance the changes needed to achieve the desired AI outputs.
Operating within siloed teams can also foster a territorial view of departmental data. This was once the case within the U.S. Postal Service, CIO Pritha Mehra noted during a Federal IT Efficiency Summit in July 2025. “With an organization as large as ours, we have data sets that everybody [individually] owns. ‘Don’t touch my data, it’s mine,’ and then finance says, ‘that’s mine,’” Mehra commented. “We had to break down a lot of the data silos…”
Culture change, aligning incentives, and making the case for data sharing may be required. As Mehra found, the work to break down these siloes is an essential step toward creating a usable foundation for the data availability needed for successful AI implementation.
Prioritizing Outcomes over Exploration Limits Discovery
In many cases, organizations encounter issues when tasked with applying AI to achieve a specific outcome before exploring the available data to understand the models it can support. AI can, and should, be applied to better understand the data at hand before building toward specific outcomes. Large AI models that are available to government organizations, such as Microsoft Copilot and the recently launched USAi evaluation suite, prove helpful in understanding available data sources and testing data discovery techniques.
This discovery phase can also help identify potential AI projects that strategically advance organizational goals. For example, an Interior Department generative AI project tailored to help improve internal search is also encouraging employees to consider how they could use the tool to streamline their day-to-day work. To encourage this discovery phase, the agency created a prompt script tailored to each business line. Using the initial version of the tool is also intended to help teams become accustomed to using AI.
A Government Leader’s Guide to Justifying the Investment in Data Readiness
The reality is that government organizations cannot successfully implement AI until their data is ready. Without this investment, the output generated will not reach the desired levels of accuracy or provide actionable outputs.
The good news is that government organizations have faced data readiness challenges before. To move to paperless forms and reports, they’ve had to manually enter data into new software programs. They’ve had to update processes and infrastructure to migrate data to the cloud. During each of these previous shifts, organizations have recognized that the change would be complex and perhaps less glamorous than other investments, but would ultimately prove necessary to achieve their goals. Whether it was to improve service to taxpayers, increase computing power, or save processing time, the upfront resource allocation to prepare data ensured the success of their efforts. The business case for advancing these previous technology shifts is similar to the argument for advancing data readiness for AI. Additionally, the data management technologies available today make it easier than ever for organizations to enhance their data readiness.
● Investing in accuracy now will reduce the risk of costly mistakes and rework later. A study from Columbia Journalism Review's Tow Center for Digital Journalism found that generative AI models used for news searches incorrectly cited sources in more than 60% of queries. While this inaccuracy is disturbing in a search context, it would be dangerous in government mission applications.
● The process of data discovery can guide next steps for effectively using AI to support specific outcomes. Informatica found that more than 97% of enterprises increasing generative AI investments admit to finding it difficult to demonstrate AI's business value. The report also found that 86% of data leaders expect increased data management investment in 2025, with 44% citing data readiness for GenAI as the primary driver of these investments. Data readiness work can help identify potential use cases that can guide AI investments and realize the potential for GenAI.
● Investing in data readiness today improves outcomes sooner.A cross-functional data readiness process can verify that all relevant data is being utilized, providing a broad foundation of accurate data from which to draw. With this foundation in place, decision-makers gain access to the smarter, faster insight promised by AI.
Advancing Data Readiness Demands Time and Expertise
Leaders tasked with advancing AI initiatives must have time and expertise available to dedicate to data readiness. To launch this process, consider requesting a set amount of time to perform a deep dive into available data. This early assessment can help begin to identify some of the following needs:
● Where data is housed. Knowing all available relevant data sources will help organizations understand the information available to guide future AI use cases.
● What types of data exist. This overview of data modalities, formats, and quality can help project leaders better know the condition of their data ecosystem and design their data pipelines and AI models accordingly.
● How data is categorized. Understanding the qualities of the data, including source, classification, and domain, as well as the mission categorization schemas, will help teams build processing workflows to optimize outputs.
● How data flows from each source. With a high-level look at system architecture, organizations can begin to understand how infrastructure works together, as well as hurdles to data integration.
● The scale of data. AI solutions require organizations to have a significant volume of historical data available to properly train or deploy models, workflows, and validations.
● Potential gaps and inconsistencies. This work can help identify areas where leaders believe data is being collected, or where data collection isn’t living up to its intended purpose.
Through this work, project leaders can also begin to identify data owners who can help advance data readiness work.
AI Investment Already Underway? PCI-GS Builds on What Works
A significant part of AI’s value lies in its ability to help organizations do more with less. We believe this should begin at the earliest stages of an AI launch. At PCI-GS, we prioritize working with organizations to maximize the value of their existing investments. Our goal is to meet organizations where they are, with recommendations on the tools and expertise needed to achieve their AI goals within their unique project constraints.
No matter what stage of the AI journey your organization is in, it’s not too late to take steps to enhance data readiness for a stronger foundation and more accurate AI results. The experts at PCI-GS have helped advance data readiness for organizations at every level of data maturity.
Whether you’re beginning to assess your data ecosystem or ready to address interoperability gaps to help data flow across systems, we can help. Contact PCI-GS to move forward.
