getty
As organizations move quickly to adopt AI, many are discovering that the biggest obstacle isn’t the technology itself; it’s the condition of the data feeding it. When data is scattered across systems, inconsistently labeled, difficult to access or poorly governed, even advanced AI tools can struggle to produce useful, reliable results.
Before companies invest more heavily in AI platforms and applications, they need to make sure their data is ready to support them. Below, Forbes Technology Council members share practical first steps organizations can take to make their data more usable for AI initiatives.
The biggest AI integration challenge confronting businesses today is data readiness. Too many think they need a massive data migration to support AI applications. That’s not the case. Modern data ontologies and APIs make it possible to extract only the data necessary for certain functions, marry it with other datasets, and create a complete data-driven workflow. - Anand Logani, EXL
Start by prioritizing data validation and relevance. Most AI systems ingest noisy, unstructured inputs, often with up to 95% irrelevant data, driving cost and accuracy issues. By defining clear specifications and filtering for high-quality, task-relevant data, organizations can reduce noise and ensure AI operates on precise, reliable inputs. - Uri Knorovich, Nimble
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
Start by studying how your data actually lives across your systems. We did this and built an internal context library that automatically formats, sanitizes and structures data before it ever gets injected into our LLM context. Clean data in, reliable output out. - Rushil Nagarsheth, Hypercard
The first step depends on what you want AI to do. For example, for a SaaS company, the data that moves revenue is product behavior, not a centralized catalog—which features users touch, which trials activate and which accounts expand. Pipe those events into the same table as your CRM so AI can see the full buyer journey in one place. That single join beats a six-month governance initiative. - Osman Koc, UserGuiding
Start by clarifying data ownership and usage rights before trying to clean or reorganize everything. If legal, security, engineering and business teams are not aligned on who owns the data and what can be used for which AI use cases, even well-structured data will create friction later. - Murali Swaminathan, Freshworks
The context provided by internal data is key to developing accurate AI agents and applications. A couple of suggestions: Ensure a data delivery foundation that can provide the AI with the necessary data and context across all internal systems. This is the focus of my company. Further, develop a unified semantic layer that is needed to provide AI agents with the context needed. - Felix Liao, Denodo
Pick one high-value workflow—not a department, a single workflow—and map every data source it touches end to end. You’ll find three things: data that’s duplicated, data that’s missing, and data that lives in someone’s spreadsheet. Fix that one pipeline before buying any AI tooling. Clean data in one workflow beats a companywide data strategy that never ships. - Abhishek Gandotra, American Express
Begin by establishing a unified, well-governed data foundation—one that centralizes, standardizes and makes data readily accessible. AI systems are only as effective as the data they rely on; without consistency and accessibility, their outputs lack reliability. A cohesive data layer is essential to unlock reliable insights and scalable impact. - Lalena Nau, Zeta Global (NYSE: ZETA)
Start with orchestration. Data is fragmented because work is fragmented inside organizations. An orchestration layer unifies workflows and the data behind them across disconnected systems and teams, allowing AI to have context to act across entire processes, not just isolated use cases. That enables AI to scale, which is a mandate for ROI. - Alfred Kahn, OvationCXM
First things first: You need a single source of truth. Otherwise, any change or optimization you make to data management won’t cascade through the entire organization. You need one source of data that can be updated in real time across departments. If you move forward with AI before your system is ready, you’ll just get an inaccurate result faster than was possible before. - Bill Rokos, Parsec Automation
First principles thinking helps. If you audit and find your data is bad, go to the foundation of the data: contracts, purchase orders, invoices, log files and so on. When you rely on second- or third-order data, the likelihood of bad information is much higher. AI compounds and amplifies errors. The more removed your data is from the source, the more likely you are to have problems. - Praful Saklani, Pramata
Before you touch the technology, you need to review the decisions. Map the choices your organization makes and ask honestly what data informs them and whether it’s trusted. The problem is often discipline rather than structure: Data was never consistently captured because the workflow didn’t require it or specify what it should look like. Fix that first; AI can only surface what is there. - Naomi Lariviere, ADP
The old philosophy of “bring your data to a warehouse” is not true anymore. Instead of centralizing data, evaluate AI frameworks that can leverage disparate data sources and provide specific intelligence where needed while enabling visibility of both problems and issues across these diverse datasets to allow better, inclusive decisions. - Prashant Darisi, Octave
In the automated retail, kiosk and smart vending world, before adopting AI, companies need to ask whether their data can be sourced and accessed in a way that supports real-time outcomes. Operational data must be captured at the source and acted on immediately; otherwise, the moment passes, the data loses its value, and companies risk losing both customers and revenue. - Brett Beveridge, The Revenue Optimization Companies (T-ROC Global)
One of the most underrated first steps to make data more usable for AI initiatives is building a data provenance layer so you can understand where your data came from, how it was collected, and what decisions shaped it. Why add context instead of just fixing the data? Because AI models do not simply learn from data; they inherit the assumptions, distortions and biases embedded in it. - Dr. Adita Karkera, Deloitte Consulting
A practical first step is to unify access to data across people, systems and AI agents. The real value comes from seeing patterns: who is accessing what and how often and what looks abnormal. That behavioral view makes data more usable for AI because it adds the context, quality and governance needed to drive better decisions and reduce hidden risk. - Joel Burleson-Davis, Imprivata
The first practical step involves systematically cataloging all data sources—structured, semi-structured and unstructured—while mapping data lineage and quality metrics. By creating a centralized data catalog, organizations gain clear visibility into their data landscape. This baseline directly accelerates AI readiness by enabling targeted improvements like data cleansing pipelines and MDM. - Pulak De, Cardinal Health
Stop installing state-of-the-art faucets on rusted plumbing. The first step to AI readiness is not algorithmic; it is architectural. Leaders must aggressively audit their “data debt” and map exactly where mission-critical data lives. If your underlying data structure is fractured, deploying AI will not generate intelligence—it will simply process bad information faster. - Abhishek Kumar, ICONMA
Less is more, and clarity is key. Data needs to be structured and labeled to be used for AI. The first step is to fix the point where data is created by standardizing how it’s captured within a key workflow with universal inputs, formats and clear ownership. When data is consistent and governed at entry, it becomes reliable, and AI can actually deliver value rather than amplifying issues. - Jo Debecker, Akkodis
The key is to choose the right AI. Right now, AI is primarily associated with LLMs, which can do a lot but not everything. They are statistical aggregators that lack a deep understanding of what they generate and mimic behavior based on retrieved information. In contrast, autonomous AI agents using composite AI (combining LLMs with logic-based, verifiable, symbolic AI) can handle even differently structured data. - Filip Dvorak, Filuta AI
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。