Business leaders have reported increased pressure from investors to deploy AI at scale (from 68% to 90% since last quarter) according to a recent report. But with increased pressure comes the risk of getting the foundations wrong. When scaling AI in business, it’s tempting to focus on the technology itself rather than the realities of how that tech will function at the organizational level.
Organizational readiness for AI is imperative. Teradata, an autonomous AI knowledge platform, outlined four foundations of enterprises that are prepared for autonomous AI. The technology makes people more productive — but productivity is a capacity unlock, not automatically a value unlock. Freed capacity only becomes value when the organization is designed to absorb and redeploy it:
- Trusted data at the foundation
- Operational infrastructure that scales with demand
- Clear accountability for model behavior
- Value redeployment that aligns with strategic objectives
The role of reliable data in AI success
1. Trusted data at the foundation
Reliable data is fundamental for scaling AI. Gartner research estimated that poor data quality costs organizations “at least $12.9 million a year on average.” A McKinsey report states that “88% of organizations now use AI in at least one function, yet only 33% scale it enterprise-wide” — and data quality is one of the biggest barriers to adoption.
AI is only as reliable as the data feeding it. Enterprises that scale successfully treat data quality, lineage, and access controls as non-negotiable.
One example is Zillow’s data science “disaster” of 2021. The real estate company launched a program where it would use a machine learning algorithm to estimate home prices and then flip the houses for a profit. But it turned out that because they were using incomplete datasets, the company failed to accurately calculate prices. “We’ve determined the unpredictability in forecasting home prices far exceeds what we anticipated and continuing to scale Zillow Offers would result in too much earnings and balance-sheet volatility,” said Zillow Group co-founder and CEO Rich Barton in a statement at the time. Zillow ended up losing $304 million in one quarter on the project and laid off 25% of its workforce.
Building systems that grow with AI demands
2. Operational infrastructure that scales with demand
Reliable data is foundational for success in scaling AI, but building infrastructure to support that growth and demand is also critical.
McKinsey found that 88% of enterprises have launched AI pilots, using proofs of concept (POCs) to ideally guide full deployments with production environments. Despite the enthusiasm for AI, studies from IDC and MIT paint a grim picture: AI pilots are rarely turning into revenue-generating, full production systems.
The IDC report suggests organizational readiness is a key reason: “The high number of AI POCs but low conversion to production indicates the low level of organizational readiness in terms of data, processes and IT infrastructure.” This makes sense, because POCs don’t stress-test systems. But there is a second gap the data doesn’t surface: even when infrastructure is technically ready, organizations frequently haven’t specified what changes operationally once the AI is running—and without that, a successful POC has nowhere to land.
Beyond reliable data, Teradata suggests that enterprises need an intelligence architecture that grows with the demands of AI. These models comprise a hybrid architecture (on premises and multi-cloud) with elastic compute; high-performance analytics engines that enable AI-grade real-time model scoring, large-scale data transformation, and parallel processing; and strong data governance and security.
Establishing responsibility for AI decisions
3. Clear accountability for model behavior
Who owns an AI decision when something goes wrong? Who’s responsible for checking model outputs and monitoring performance? Who will manage course correction?
A lack of strong governance is a major barrier to scaling AI. A Deloitte report found that although 74% of companies plan to deploy agentic AI within two years, only 21% believe they have a mature model for governance of autonomous agents. The same report stated that although AI is rapidly evolving beyond generative AI into agentic AI, only “42% of companies believe their strategy is highly prepared for AI adoption, and 30% say the same about risk and governance.”
There are both protective and proactive stances enterprises can take. For example, Forbes points to the federal case of Garcia v. Character Technologies Inc., in which a court allowed that makers of an AI chatbot could potentially incur product liability based on design choices such as a failure to build in safeguards. The article noted, “The absence of human oversight and structured safety mechanisms not only creates operational risk — it creates legal and financial risk.”
A Harvard Business School article sums it up: “Integrating AI into day-to-day tasks requires leaders who can acquire, scale, and build a workforce that uses AI responsibly, rather than allowing the technology to take over.”
The National Institute of Standards and Technology (NIST) provides a framework and a playbook that emphasize the importance of accountability and keeping a human in the loop to maintain governance and establish accountability. An article explaining the framework states its three core principles succinctly: “ensuring transparency, applying oversight that matches the level of risk, and clearly defining who is responsible for what.”
Defining where freed capacity gets redirected
4. Value redeployment that aligns with strategic objectives
Data quality, scalable infrastructure, and sound governance address whether AI can work. The fourth foundation addresses whether it will actually generate value—and it is the one most consistently missing from enterprise AI programs.
A BCG analysis found that only 26% of AI initiatives at large companies achieve significant financial impact. The gap is not technical. It is organizational. When AI makes a team more productive, that freed capacity has three possible destinations: it gets redirected to a higher-value named activity, it enables a downstream function to move faster, or it simply dissipates into lower-priority work and meetings. That third outcome—evaporation—is the default wherever no one has deliberately designed an alternative.
The solution is specific. Before an AI initiative is approved, there should be a named answer to one question: where does the freed capacity go? Not “people will focus on higher-value work,” but a concrete redeployment pathway—signed off by a named manager—describing exactly what the team does differently once the AI is running. Without it, productivity gains are real, but value creation is not.
This also means targeting the right constraint. AI productivity gains convert to business outcomes most reliably when they address the function that is actually the bottleneck—the evaluation capacity that limits how many deals can progress, the relationship coverage that limits executive engagement, the signal quality that determines whether at-risk customers are caught in time. Gains applied elsewhere in the organization create backlog rather than results. Identifying the binding constraint and directing AI investment there is what separates organizations generating meaningful returns from those running impressive pilots.
Although some organizations are struggling to launch and scale AI in the enterprise, the research indicates that there are clear steps companies can take to create a strong foundation as they adopt AI at scale. Companies must start with clean data, build the operational infrastructure to help AI scale into production, and implement strong human-in-the-loop governance to minimize risk. But those three foundations answer the question of whether AI can work. The fourth — named accountability for where freed capacity goes — answers the question of whether it will. The enterprises pulling ahead are the ones asking both.