# Is Your Data Ready Before You Launch AI? **By Dan Cumberland** · Published August 27, 2026 · Categories: AI Strategy ![Featured image for Is Your Data Ready Before You Launch AI?](https://cms.dancumberlandlabs.com/uploads/is_your_data_ready_before_you_launch_ai_c94844c5d6.webp) Illustration: Dan Cumberland Labs with Gemini. > A founder-led firm I know of rolled out AI-powered forecasting last year. By month three, the recommendations were consistently off— enough that the... A founder\-led firm I know of rolled out AI\-powered forecasting last year\. By month three, the recommendations were consistently off— enough that the operations team stopped checking them\. By month six, the tool was open in a browser tab nobody scrolled to\. The AI wasn't broken\. The model was fine\. The data going into it wasn't\. Most AI implementations fail because the data feeding the AI is wrong— and the AI doesn't know the difference\. That's the whole problem\. And it means that upgrading your AI tool when this happens doesn't help— you're just adding a faster engine to a car with no wheels\. Before you spend another dollar on AI, here's how to tell whether the problem is the tool or the input\. \[DAN FIRST\-PERSON: 1\-2 sentences on seeing this pattern in client engagements— request Dan supply at HITL\] If that pattern sounds familiar, you're not alone\. According to Gartner[1](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-1), 63% of organizations either don't have or aren't sure they have the right data management practices for AI\. The firms investing in better AI tools before solving this are digging in deeper\. To understand why, it helps to understand what AI actually does with data— because it's not what most people expect\. ## What AI Actually Does with Your Data AI does not evaluate the quality of your data before using it\. It takes whatever you give it, finds patterns in it, and produces answers— confidently— whether or not the underlying data is accurate\. Think of it as a mirror, not a filter\. AI doesn't clean up what it sees before reflecting it back\. It shows you your data at scale, with confidence, and without a single flag that says "this might be wrong\." That's the core mechanism behind the principle garbage in, garbage out \(GIGO\)— and AI makes the problem worse than any manual process could\. There are three specific ways bad data breaks AI[2](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-2): - **Inconsistency destroys pattern recognition\.** When the same entity— a vendor, a project, a customer— is recorded multiple ways, AI can't find reliable patterns\. If lead times are accurate for some vendors but missing for others, the system can't make reliable supply recommendations\.[2](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-2) - **False confidence without warning\.** AI doesn't behave like a cautious employee who says "I'm not sure about this\." It gives confident answers, even when the data behind them is wrong\.[2](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-2) - **Errors at scale\.** Automation amplifies rather than corrects flawed inputs\. Advanced systems process bad data faster, scaling errors across workflows, financial systems, and operations\.[3](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-3) The counterintuitive part: a better AI model on bad data produces more confident wrong answers\. ERP systems— the operational platforms where your business data lives— require data that is complete, consistent, and current before an AI layer can do anything useful\. As Support One puts it[4](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-4): AI cannot fix poor ERP data\. If the underlying information is incomplete, inconsistent, or inaccurate, AI recommendations may also be unreliable\. Understanding the mechanism explains why AI fails\. But it doesn't explain why the data was bad in the first place— and that's where most advice falls short\. ## The Root Cause Nobody Talks About Bad data is a human behavior problem, not a technology one— and that's what makes it harder to fix\. Most content on this topic says "clean your data before implementing AI\." True enough\. But useless without the next part: your data is bad because your people are entering it in ways that don't serve them\. That's not carelessness\. It's a rational response to a broken feedback loop\. Employees experience data entry as administrative overhead with no visible payoff\. When the system is rigid— fields that don't match how work actually gets done— workarounds develop\. Notes land in the wrong fields\. Fake part numbers get created\. Spreadsheets grow alongside the ERP because the ERP doesn't work for the people using it\. There's no personal consequence for entering data badly, and no personal reward for entering it right\. The result, documented across ERP implementations[4](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-4), includes: - Duplicate customer and vendor records with inconsistent spellings - Missing required fields left blank at entry - Outdated inventory data, pricing, and bills of materials - "Workaround data"— real information stored in the wrong place because the right place doesn't exist The scale of this is significant\. A Qlik survey[5](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-5) of 500 U\.S\. AI professionals found that 81% say their company still has significant data quality issues\. And 85% believe leadership isn't adequately addressing them\.[5](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-5) Data quality as the top AI challenge more than doubled in a single year— from 19% of organizations in 2024 to 44% in 2025\.[5](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-5) This is not a problem trending toward solved\. For AEC firms specifically— architecture, engineering, and construction— per a Dodge survey cited by For Construction Pros[6](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-6), 74% of U\.S\. contractors rate their data quality as poor or only moderate\. Decades of project data sit in fragmented systems never designed for machine analysis\. This behavior pattern has a predictable consequence when AI enters the picture— one that's expensive and surprisingly fast to set in\. ## The Trust Erosion Trap When AI produces bad recommendations, users don't switch tools\. They stop using the tool entirely\. And it happens fast\. TeccWeb's analysis of ERP implementations[2](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-2) documents this directly: if a planner or buyer follows two bad AI suggestions in a row, they stop using the system\. The trust is gone\. It doesn't come back— even after the data is corrected— because the association is already formed\. As TeccWeb's research puts it in practice: > "If a planner or buyer follows two bad AI suggestions, they stop using it\. The firm paid for AI and now nobody's using it— not because the AI failed, but because the data did\." The financial picture is proportional\. Gartner estimates that poor data quality costs the average enterprise $12\.9 to $15 million annually\.[7](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-7) For founder\-led firms, the raw dollar figures are different\. But the AI investment loss and the trust erosion are proportionally the same\. You've already paid for the AI\. The question is whether it's being used\. The [hidden costs of AI projects](/blog/hidden-costs-ai-projects) aren't just in licensing fees\. They're in the organizational trust you spend trying to rebuild adoption after data\-quality failures— and in the shadow credibility problem that follows when leadership backed a tool the team stopped using\. Through 2026, Gartner projects that 60% of AI projects without AI\-ready data will be abandoned\.[1](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-1) That trajectory is already underway\. The fix is the work that should have happened before the AI was purchased\. ## What You Actually Need Before AI The fix for a data\-input problem is process accountability— not a software upgrade\. Before adding AI, organizations that succeed invest in the foundations: data quality, governance, people, and change management\. The numbers are stark\. Gartner's April 2026 survey of 353 data and analytics leaders[8](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-8) found that organizations with successful AI outcomes invest up to four times more— as a percentage of revenue— in these foundational areas\. The AI tool itself is the last mile, not the starting point\. The barriers to building those foundations are human, not technical\. A 2024 Gartner survey, cited by ERP Today,[9](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-9) identified the top obstacles to AI adoption as lack of training \(30%\) and change resistance \(30%\)— ahead of poor AI quality \(29%\) and missing process integration \(26%\)\. Notice what comes first\. It's people\. **A 4\-question self\-diagnostic:** ```html-table
QuestionWhat it reveals
Do your team members skip required fields in your CRM or ERP regularly?Whether data entry is treated as a requirement or a suggestion
Is the same entity— customer, vendor, project— recorded multiple ways in your system?Whether inconsistency has compounded to the point that pattern recognition fails
Do you have "unofficial" data sources: spreadsheets, personal notes, workarounds alongside the system?Whether the system works for the people using it, or whether they've built around it
When AI makes a recommendation, can you trace the data behind it?Whether you have the visibility to identify and correct bad inputs before they compound
``` If you answered yes to any of these, you have a data\-input problem that a better AI tool won't fix\. What the minimum viable fix looks like: - **Point\-of\-entry validation:** Make bad data entry impossible, not just discouraged\. If a field matters to your AI, it needs to be required at input\. - **Accountability structures:** Connect data quality to outcomes the team can see\. When people see what their inputs produce— or fail to produce— behavior changes\. - **A pre\-AI audit:** Know what's in your system before the AI goes live\. Not after\. This is a project, not a checkbox\. Don't automate workflows until they're well\-established manually first\. That applies to data too\. The [AI governance strategy](/blog/ai-governance-strategy) you build now determines what AI can do for you in two years\. And [building an AI\-ready culture](/blog/building-ai-culture) is the organizational work that makes any tool— not just the one you're evaluating today— useful\. The question to ask isn't "which AI tool should we buy?"— it's "do we have the process discipline to make any AI tool work?" ## The Real Question to Ask Data quality is a management problem, not a data one— and that's good news, because management problems have solutions that don't require waiting for better technology\. Firms that layer AI on top of broken data processes aren't implementing AI\. They're amplifying their dysfunction at speed and scale\. Leaders who recognize this stop chasing better AI tools and start asking better questions about their own processes\. No matter the question, people are the answer\. The AI tools are infrastructure\. The people who enter the data, the managers who establish accountability, the leaders who audit what's in the system before they buy the software— that's where AI success or failure actually lives\. If you're navigating the [right sequence of decisions](/blog/ai-decision-framework-founders)— whether to build data foundations first or run a limited pilot to learn what you have— an [AI implementation partner](/services/ai-implementation/) who has worked through this sequence with founder\-led firms can compress the timeline considerably\. The work doesn't have to take as long as it sounds\. But it does have to come first\. ## Frequently Asked Questions ### Can AI automatically clean up bad data? No\. AI systems use whatever data exists and produce outputs based on it— without flagging poor quality\.[2](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-2)[4](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-4) Some tools can surface anomalies, but resolving them requires human judgment\. And they don't prevent the input behavior that creates bad data in the first place\. The root problem is behavioral, not technical— which means the fix has to be behavioral too\. ### Why do most AI projects fail? The most common reason is poor data quality, not flawed AI models\. Gartner's February 2025 research[1](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-1) found that through 2026, 60% of AI projects without AI\-ready data will be abandoned\. Most organizations are implementing AI on a data foundation that isn't ready to support it— and often don't know that until users stop trusting the output\. ### What is AI\-ready data? AI\-ready data is complete, consistent, current, and structured in ways AI can reliably analyze\.[1](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-1) This typically requires defined governance standards and point\-of\-entry validation— not just a data cleanup project before launch\. The distinction matters: a cleanup addresses what's in the system now\. Governance determines what goes in going forward\. ### What do successful AI organizations do differently? They invest up to four times more— as a percentage of revenue— in data quality, governance, AI\-ready people, and change management, according to a 2026 Gartner survey of 353 data and analytics leaders\.[8](/blog/blog-you-can-t-ai-your-way-out-of-a-data-input-problem#ref-8) The AI tool itself is where they spend last, not first\. The most common mistake is reversing that sequence\. ### Meta Title \(Final\) ``` AI Data Quality Problems Are Process Problems ``` ### Meta Description \(Final\) ``` AI data quality problems get fixed by process accountability. Gartner: 63% of firms lack AI-ready data. Take the 4-question test before you buy a new tool. ``` ## References 1. Gartner, "Lack of AI\-Ready Data Puts AI Projects at Risk" \(February 2025\)— [https://www\.gartner\.com/en/newsroom/press\-releases/2025\-02\-26\-lack\-of\-ai\-ready\-data\-puts\-ai\-projects\-at\-risk](https://www.gartner.com/en/newsroom/press-releases/2025-02-26-lack-of-ai-ready-data-puts-ai-projects-at-risk) 2. TeccWeb, "Why Bad Data Will Break Your Epicor AI Strategy" \(2024\)— [https://teccweb\.com/bad\-data\-breaks\-epicor\-ai\-strategy/](https://teccweb.com/bad-data-breaks-epicor-ai-strategy/) 3. Parseur, "Garbage In, Garbage Out: Why Bad Data Destroys Automation ROI" \(2024\)— [https://parseur\.com/blog/gigo](https://parseur.com/blog/gigo) 4. Support One, "Why Your ERP Data Quality Matters More Than Ever" \(2024\)— [https://supportone\.us/why\-your\-erp\-data\-quality\-matters\-more\-than\-ever/](https://supportone.us/why-your-erp-data-quality-matters-more-than-ever/) 5. Qlik, "Data Quality is Not Being Prioritized on AI Projects" \(March 2025\)— [https://www\.qlik\.com/us/news/company/press\-room/press\-releases/data\-quality\-is\-not\-being\-prioritized\-on\-ai\-projects](https://www.qlik.com/us/news/company/press-room/press-releases/data-quality-is-not-being-prioritized-on-ai-projects) 6. For Construction Pros, citing Dodge survey, "Construction Firms Face Data and Security Challenges in AI Adoption" \(2024\)— [https://www\.forconstructionpros\.com/business/business\-services/training\-education/article/22967287/creative\-itc\-construction\-firms\-face\-data\-and\-security\-challenges\-in\-ai\-adoption](https://www.forconstructionpros.com/business/business-services/training-education/article/22967287/creative-itc-construction-firms-face-data-and-security-challenges-in-ai-adoption) 7. Gartner \(widely cited estimate\), "Poor Data Quality Costs $15M Annually"— [https://www\.actian\.com/blog/data\-management/the\-costly\-consequences\-of\-poor\-data\-quality/](https://www.actian.com/blog/data-management/the-costly-consequences-of-poor-data-quality/) 8. Gartner, "Organizations with Successful AI Initiatives Invest Up to Four Times More in Data and Analytics Foundations" \(April 2026\)— [https://www\.gartner\.com/en/newsroom/press\-releases/2026\-04\-16\-gartner\-says\-organizations\-with\-successful\-ai\-initiatives\-invest\-up\-to\-four\-times\-more\-in\-data\-and\-analytics\-foundations](https://www.gartner.com/en/newsroom/press-releases/2026-04-16-gartner-says-organizations-with-successful-ai-initiatives-invest-up-to-four-times-more-in-data-and-analytics-foundations) 9. Gartner 2024 survey, cited via ERP Today, "How AI Is Forcing ERP Vendors to Rethink the Human Side of Transformation" \(2024\)— [https://erp\.today/how\-ai\-is\-forcing\-erp\-vendors\-to\-rethink\-the\-human\-side\-of\-transformation/](https://erp.today/how-ai-is-forcing-erp-vendors-to-rethink-the-human-side-of-transformation/) --- **How we made this article:** We use AI in our research and writing so our small team can share more of what we learn. We verify the sources and take responsibility for every article we publish. [Read how we use AI.](https://dancumberlandlabs.com/how-we-use-ai/) --- ## About the author **Dan Cumberland** — Founder, Dan Cumberland Labs Dan Cumberland helps engineering and construction firms see where they stand with AI and decide what to build first. He created Pacemark, the AI maturity model behind that work, from research on more than 300 companies. - Take the assessment: https://pacemark.ai/signal/assessment/?track=aec&utm_source=dcl-site&utm_medium=link&utm_campaign=pacemark-assessment - Book a call: https://book.dancumberland.com/ai-strategy --- Source: https://dancumberlandlabs.com/blog/ai-data-quality-problems/